diff --git "a/checkpoint-37100/trainer_state.json" "b/checkpoint-37100/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/checkpoint-37100/trainer_state.json"
@@ -0,0 +1,25618 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 20.339361556377586,
+  "eval_steps": 100,
+  "global_step": 37100,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "learning_rate": 5.399999999999999e-06,
+      "loss": 11.7783,
+      "step": 10
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 1.0799999999999998e-05,
+      "loss": 12.9293,
+      "step": 20
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 1.68e-05,
+      "loss": 13.9825,
+      "step": 30
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 2.2199999999999998e-05,
+      "loss": 16.1366,
+      "step": 40
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 2.7599999999999997e-05,
+      "loss": 34.2401,
+      "step": 50
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.36e-05,
+      "loss": 10.2244,
+      "step": 60
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.96e-05,
+      "loss": 8.0319,
+      "step": 70
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 4.56e-05,
+      "loss": 6.303,
+      "step": 80
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 5.1599999999999994e-05,
+      "loss": 5.358,
+      "step": 90
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 5.76e-05,
+      "loss": 7.887,
+      "step": 100
+    },
+    {
+      "epoch": 0.03,
+      "eval_loss": 3.6606574058532715,
+      "eval_runtime": 1231.7206,
+      "eval_samples_per_second": 9.485,
+      "eval_steps_per_second": 1.186,
+      "eval_wer": 1.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 6.359999999999999e-05,
+      "loss": 3.6845,
+      "step": 110
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 6.96e-05,
+      "loss": 3.4054,
+      "step": 120
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 7.56e-05,
+      "loss": 3.4159,
+      "step": 130
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 8.16e-05,
+      "loss": 3.3657,
+      "step": 140
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 8.759999999999999e-05,
+      "loss": 4.0007,
+      "step": 150
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.36e-05,
+      "loss": 3.3271,
+      "step": 160
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.96e-05,
+      "loss": 3.185,
+      "step": 170
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00010559999999999998,
+      "loss": 3.1758,
+      "step": 180
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00011159999999999999,
+      "loss": 3.124,
+      "step": 190
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0001176,
+      "loss": 4.0646,
+      "step": 200
+    },
+    {
+      "epoch": 0.05,
+      "eval_loss": 3.173530101776123,
+      "eval_runtime": 1202.0614,
+      "eval_samples_per_second": 9.719,
+      "eval_steps_per_second": 1.215,
+      "eval_wer": 1.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0001236,
+      "loss": 4.6529,
+      "step": 210
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000129,
+      "loss": 3.8886,
+      "step": 220
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.000135,
+      "loss": 3.6103,
+      "step": 230
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00014099999999999998,
+      "loss": 3.2837,
+      "step": 240
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000147,
+      "loss": 3.7727,
+      "step": 250
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00015299999999999998,
+      "loss": 3.7147,
+      "step": 260
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000159,
+      "loss": 3.3716,
+      "step": 270
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.000165,
+      "loss": 3.135,
+      "step": 280
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00017099999999999998,
+      "loss": 3.1926,
+      "step": 290
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00017699999999999997,
+      "loss": 3.7509,
+      "step": 300
+    },
+    {
+      "epoch": 0.08,
+      "eval_loss": 3.2335991859436035,
+      "eval_runtime": 1163.562,
+      "eval_samples_per_second": 10.041,
+      "eval_steps_per_second": 1.256,
+      "eval_wer": 1.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00018299999999999998,
+      "loss": 3.3192,
+      "step": 310
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00018899999999999999,
+      "loss": 3.0958,
+      "step": 320
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000195,
+      "loss": 3.0755,
+      "step": 330
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000201,
+      "loss": 3.1681,
+      "step": 340
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00020699999999999996,
+      "loss": 3.7908,
+      "step": 350
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00021299999999999997,
+      "loss": 3.4041,
+      "step": 360
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00021899999999999998,
+      "loss": 3.1383,
+      "step": 370
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.000225,
+      "loss": 3.1267,
+      "step": 380
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00023099999999999998,
+      "loss": 3.2255,
+      "step": 390
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0002364,
+      "loss": 3.8832,
+      "step": 400
+    },
+    {
+      "epoch": 0.11,
+      "eval_loss": 3.147369623184204,
+      "eval_runtime": 1163.4758,
+      "eval_samples_per_second": 10.041,
+      "eval_steps_per_second": 1.256,
+      "eval_wer": 1.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 5.999999999999999e-06,
+      "loss": 4.262,
+      "step": 410
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 1.1999999999999999e-05,
+      "loss": 4.2338,
+      "step": 420
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 1.7999999999999997e-05,
+      "loss": 4.1712,
+      "step": 430
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 2.28e-05,
+      "loss": 4.4312,
+      "step": 440
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 2.88e-05,
+      "loss": 4.9701,
+      "step": 450
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.48e-05,
+      "loss": 3.8904,
+      "step": 460
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 4.08e-05,
+      "loss": 3.7617,
+      "step": 470
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 4.68e-05,
+      "loss": 3.6052,
+      "step": 480
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 5.279999999999999e-05,
+      "loss": 3.5166,
+      "step": 490
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 5.88e-05,
+      "loss": 3.9646,
+      "step": 500
+    },
+    {
+      "epoch": 0.14,
+      "eval_loss": 3.334455728530884,
+      "eval_runtime": 1100.3955,
+      "eval_samples_per_second": 10.617,
+      "eval_steps_per_second": 1.328,
+      "eval_wer": 1.0,
+      "step": 500
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 6.479999999999999e-05,
+      "loss": 3.234,
+      "step": 510
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 7.079999999999999e-05,
+      "loss": 3.1796,
+      "step": 520
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 7.68e-05,
+      "loss": 3.1659,
+      "step": 530
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.28e-05,
+      "loss": 3.1283,
+      "step": 540
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.879999999999999e-05,
+      "loss": 3.2447,
+      "step": 550
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 9.479999999999999e-05,
+      "loss": 3.0545,
+      "step": 560
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0001008,
+      "loss": 3.08,
+      "step": 570
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00010679999999999998,
+      "loss": 3.0577,
+      "step": 580
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00011279999999999999,
+      "loss": 3.0595,
+      "step": 590
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0001188,
+      "loss": 3.1657,
+      "step": 600
+    },
+    {
+      "epoch": 0.16,
+      "eval_loss": 3.055232048034668,
+      "eval_runtime": 1107.3435,
+      "eval_samples_per_second": 10.55,
+      "eval_steps_per_second": 1.319,
+      "eval_wer": 1.0,
+      "step": 600
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00012479999999999997,
+      "loss": 3.0392,
+      "step": 610
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00013079999999999998,
+      "loss": 3.03,
+      "step": 620
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0001368,
+      "loss": 3.0355,
+      "step": 630
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00014279999999999997,
+      "loss": 3.104,
+      "step": 640
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00014879999999999998,
+      "loss": 3.3866,
+      "step": 650
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0001548,
+      "loss": 3.043,
+      "step": 660
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0001608,
+      "loss": 3.0396,
+      "step": 670
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0001668,
+      "loss": 3.0713,
+      "step": 680
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00017279999999999997,
+      "loss": 3.0453,
+      "step": 690
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00017879999999999998,
+      "loss": 3.0641,
+      "step": 700
+    },
+    {
+      "epoch": 0.19,
+      "eval_loss": 3.091907024383545,
+      "eval_runtime": 1074.3754,
+      "eval_samples_per_second": 10.874,
+      "eval_steps_per_second": 1.36,
+      "eval_wer": 1.0,
+      "step": 700
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0001848,
+      "loss": 3.045,
+      "step": 710
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00019079999999999998,
+      "loss": 3.0312,
+      "step": 720
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00019679999999999999,
+      "loss": 3.0153,
+      "step": 730
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0002028,
+      "loss": 2.9866,
+      "step": 740
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00020879999999999998,
+      "loss": 2.9827,
+      "step": 750
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00021479999999999996,
+      "loss": 2.9405,
+      "step": 760
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00022079999999999997,
+      "loss": 2.7764,
+      "step": 770
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00022679999999999998,
+      "loss": 2.5073,
+      "step": 780
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0002328,
+      "loss": 2.1963,
+      "step": 790
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0002388,
+      "loss": 1.9443,
+      "step": 800
+    },
+    {
+      "epoch": 0.22,
+      "eval_loss": 1.5123839378356934,
+      "eval_runtime": 1077.0854,
+      "eval_samples_per_second": 10.847,
+      "eval_steps_per_second": 1.356,
+      "eval_wer": 1.0,
+      "step": 800
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0002448,
+      "loss": 1.5669,
+      "step": 810
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00025079999999999997,
+      "loss": 1.2878,
+      "step": 820
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00025679999999999995,
+      "loss": 1.1475,
+      "step": 830
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0002628,
+      "loss": 1.0235,
+      "step": 840
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0002688,
+      "loss": 1.0418,
+      "step": 850
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0002748,
+      "loss": 0.916,
+      "step": 860
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0002808,
+      "loss": 0.8304,
+      "step": 870
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0002868,
+      "loss": 0.6735,
+      "step": 880
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00029279999999999996,
+      "loss": 0.6997,
+      "step": 890
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0002988,
+      "loss": 0.6772,
+      "step": 900
+    },
+    {
+      "epoch": 0.25,
+      "eval_loss": 0.6391938328742981,
+      "eval_runtime": 1088.2,
+      "eval_samples_per_second": 10.736,
+      "eval_steps_per_second": 1.343,
+      "eval_wer": 0.8844946784317292,
+      "step": 900
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0002999779755896118,
+      "loss": 0.697,
+      "step": 910
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00029995044507662654,
+      "loss": 0.5928,
+      "step": 920
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00029992291456364134,
+      "loss": 0.5868,
+      "step": 930
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00029989538405065613,
+      "loss": 0.588,
+      "step": 940
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0002998678535376709,
+      "loss": 0.6109,
+      "step": 950
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00029984032302468567,
+      "loss": 0.5986,
+      "step": 960
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00029981279251170047,
+      "loss": 0.5192,
+      "step": 970
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0002997852619987152,
+      "loss": 0.5146,
+      "step": 980
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00029975773148573,
+      "loss": 0.5384,
+      "step": 990
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00029973020097274474,
+      "loss": 0.5144,
+      "step": 1000
+    },
+    {
+      "epoch": 0.27,
+      "eval_loss": 0.4253242611885071,
+      "eval_runtime": 1098.9682,
+      "eval_samples_per_second": 10.631,
+      "eval_steps_per_second": 1.329,
+      "eval_wer": 0.7255526243973438,
+      "step": 1000
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00029970267045975954,
+      "loss": 0.5067,
+      "step": 1010
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00029967513994677434,
+      "loss": 0.4826,
+      "step": 1020
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0002996476094337891,
+      "loss": 0.5047,
+      "step": 1030
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0002996200789208039,
+      "loss": 0.4862,
+      "step": 1040
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00029959254840781867,
+      "loss": 0.4903,
+      "step": 1050
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0002995650178948334,
+      "loss": 0.524,
+      "step": 1060
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0002995374873818482,
+      "loss": 0.4635,
+      "step": 1070
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00029950995686886295,
+      "loss": 0.4518,
+      "step": 1080
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00029948242635587774,
+      "loss": 0.4541,
+      "step": 1090
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00029945489584289254,
+      "loss": 0.4361,
+      "step": 1100
+    },
+    {
+      "epoch": 0.3,
+      "eval_loss": 0.3405020236968994,
+      "eval_runtime": 1102.4518,
+      "eval_samples_per_second": 10.597,
+      "eval_steps_per_second": 1.325,
+      "eval_wer": 0.6595788228872919,
+      "step": 1100
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0002994273653299073,
+      "loss": 0.4521,
+      "step": 1110
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0002993998348169221,
+      "loss": 0.4303,
+      "step": 1120
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0002993723043039368,
+      "loss": 0.4547,
+      "step": 1130
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0002993447737909516,
+      "loss": 0.4971,
+      "step": 1140
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0002993172432779664,
+      "loss": 0.4177,
+      "step": 1150
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00029928971276498115,
+      "loss": 0.4208,
+      "step": 1160
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00029926218225199595,
+      "loss": 0.4125,
+      "step": 1170
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0002992346517390107,
+      "loss": 0.3863,
+      "step": 1180
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0002992071212260255,
+      "loss": 0.4037,
+      "step": 1190
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0002991795907130403,
+      "loss": 0.4239,
+      "step": 1200
+    },
+    {
+      "epoch": 0.33,
+      "eval_loss": 0.2857191860675812,
+      "eval_runtime": 1081.4677,
+      "eval_samples_per_second": 10.803,
+      "eval_steps_per_second": 1.351,
+      "eval_wer": 0.5968798326207586,
+      "step": 1200
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.000299152060200055,
+      "loss": 0.3721,
+      "step": 1210
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0002991245296870698,
+      "loss": 0.3766,
+      "step": 1220
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00029909699917408456,
+      "loss": 0.4146,
+      "step": 1230
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00029906946866109936,
+      "loss": 0.4146,
+      "step": 1240
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00029904193814811415,
+      "loss": 0.3439,
+      "step": 1250
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0002990144076351289,
+      "loss": 0.3952,
+      "step": 1260
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0002989868771221437,
+      "loss": 0.3558,
+      "step": 1270
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00029895934660915843,
+      "loss": 0.3923,
+      "step": 1280
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0002989318160961732,
+      "loss": 0.3893,
+      "step": 1290
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.000298904285583188,
+      "loss": 0.3226,
+      "step": 1300
+    },
+    {
+      "epoch": 0.36,
+      "eval_loss": 0.279046893119812,
+      "eval_runtime": 1089.4065,
+      "eval_samples_per_second": 10.724,
+      "eval_steps_per_second": 1.341,
+      "eval_wer": 0.5880332939143091,
+      "step": 1300
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00029887675507020276,
+      "loss": 0.4045,
+      "step": 1310
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00029884922455721756,
+      "loss": 0.3552,
+      "step": 1320
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00029882169404423235,
+      "loss": 0.3442,
+      "step": 1330
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0002987941635312471,
+      "loss": 0.3695,
+      "step": 1340
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0002987666330182619,
+      "loss": 0.3378,
+      "step": 1350
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00029873910250527663,
+      "loss": 0.3464,
+      "step": 1360
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00029871157199229143,
+      "loss": 0.3532,
+      "step": 1370
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0002986840414793062,
+      "loss": 0.3477,
+      "step": 1380
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00029865651096632097,
+      "loss": 0.3431,
+      "step": 1390
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00029862898045333576,
+      "loss": 0.3508,
+      "step": 1400
+    },
+    {
+      "epoch": 0.38,
+      "eval_loss": 0.24916307628154755,
+      "eval_runtime": 1079.8194,
+      "eval_samples_per_second": 10.819,
+      "eval_steps_per_second": 1.353,
+      "eval_wer": 0.5586168470845083,
+      "step": 1400
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00029860144994035056,
+      "loss": 0.372,
+      "step": 1410
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0002985739194273653,
+      "loss": 0.3597,
+      "step": 1420
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0002985463889143801,
+      "loss": 0.3585,
+      "step": 1430
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00029851885840139484,
+      "loss": 0.3502,
+      "step": 1440
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00029849132788840963,
+      "loss": 0.3097,
+      "step": 1450
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00029846379737542443,
+      "loss": 0.3237,
+      "step": 1460
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00029843626686243917,
+      "loss": 0.3071,
+      "step": 1470
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00029840873634945397,
+      "loss": 0.3479,
+      "step": 1480
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002983812058364687,
+      "loss": 0.3608,
+      "step": 1490
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002983536753234835,
+      "loss": 0.3091,
+      "step": 1500
+    },
+    {
+      "epoch": 0.41,
+      "eval_loss": 0.25719356536865234,
+      "eval_runtime": 1086.8697,
+      "eval_samples_per_second": 10.749,
+      "eval_steps_per_second": 1.344,
+      "eval_wer": 0.5655075957427453,
+      "step": 1500
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002983261448104983,
+      "loss": 0.3485,
+      "step": 1510
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029829861429751304,
+      "loss": 0.3394,
+      "step": 1520
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00029827108378452784,
+      "loss": 0.3418,
+      "step": 1530
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002982435532715426,
+      "loss": 0.3772,
+      "step": 1540
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002982160227585574,
+      "loss": 0.3438,
+      "step": 1550
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00029818849224557217,
+      "loss": 0.3139,
+      "step": 1560
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002981609617325869,
+      "loss": 0.3213,
+      "step": 1570
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002981334312196017,
+      "loss": 0.3382,
+      "step": 1580
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00029810590070661645,
+      "loss": 0.3355,
+      "step": 1590
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00029807837019363124,
+      "loss": 0.3094,
+      "step": 1600
+    },
+    {
+      "epoch": 0.44,
+      "eval_loss": 0.2237711250782013,
+      "eval_runtime": 1040.3304,
+      "eval_samples_per_second": 11.23,
+      "eval_steps_per_second": 1.404,
+      "eval_wer": 0.5121668334394615,
+      "step": 1600
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00029805083968064604,
+      "loss": 0.3198,
+      "step": 1610
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0002980233091676608,
+      "loss": 0.3048,
+      "step": 1620
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002979957786546756,
+      "loss": 0.3149,
+      "step": 1630
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00029796824814169037,
+      "loss": 0.3298,
+      "step": 1640
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002979407176287051,
+      "loss": 0.3102,
+      "step": 1650
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002979131871157199,
+      "loss": 0.2989,
+      "step": 1660
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00029788565660273465,
+      "loss": 0.3399,
+      "step": 1670
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00029785812608974945,
+      "loss": 0.2812,
+      "step": 1680
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00029783059557676424,
+      "loss": 0.4134,
+      "step": 1690
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.000297803065063779,
+      "loss": 0.305,
+      "step": 1700
+    },
+    {
+      "epoch": 0.47,
+      "eval_loss": 0.23389825224876404,
+      "eval_runtime": 1030.0489,
+      "eval_samples_per_second": 11.342,
+      "eval_steps_per_second": 1.418,
+      "eval_wer": 0.5010688620030929,
+      "step": 1700
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002977755345507938,
+      "loss": 0.3217,
+      "step": 1710
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002977480040378085,
+      "loss": 0.2761,
+      "step": 1720
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0002977204735248233,
+      "loss": 0.3057,
+      "step": 1730
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002976929430118381,
+      "loss": 0.3093,
+      "step": 1740
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00029766541249885286,
+      "loss": 0.2508,
+      "step": 1750
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00029763788198586765,
+      "loss": 0.3088,
+      "step": 1760
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002976103514728824,
+      "loss": 0.2762,
+      "step": 1770
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002975828209598972,
+      "loss": 0.2767,
+      "step": 1780
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.000297555290446912,
+      "loss": 0.3105,
+      "step": 1790
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002975277599339267,
+      "loss": 0.258,
+      "step": 1800
+    },
+    {
+      "epoch": 0.49,
+      "eval_loss": 0.2067476361989975,
+      "eval_runtime": 1030.7384,
+      "eval_samples_per_second": 11.335,
+      "eval_steps_per_second": 1.417,
+      "eval_wer": 0.5007845901937596,
+      "step": 1800
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002975002294209415,
+      "loss": 0.3283,
+      "step": 1810
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002974726989079563,
+      "loss": 0.2985,
+      "step": 1820
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00029744516839497106,
+      "loss": 0.2884,
+      "step": 1830
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00029741763788198585,
+      "loss": 0.2912,
+      "step": 1840
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002973901073690006,
+      "loss": 0.2704,
+      "step": 1850
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002973625768560154,
+      "loss": 0.2642,
+      "step": 1860
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002973350463430302,
+      "loss": 0.2884,
+      "step": 1870
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00029730751583004493,
+      "loss": 0.2991,
+      "step": 1880
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002972799853170597,
+      "loss": 0.2994,
+      "step": 1890
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00029725245480407447,
+      "loss": 0.2673,
+      "step": 1900
+    },
+    {
+      "epoch": 0.52,
+      "eval_loss": 0.2200978547334671,
+      "eval_runtime": 1028.7302,
+      "eval_samples_per_second": 11.357,
+      "eval_steps_per_second": 1.42,
+      "eval_wer": 0.5077321932138633,
+      "step": 1900
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00029722492429108926,
+      "loss": 0.3358,
+      "step": 1910
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00029719739377810406,
+      "loss": 0.2968,
+      "step": 1920
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002971698632651188,
+      "loss": 0.2756,
+      "step": 1930
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002971423327521336,
+      "loss": 0.3264,
+      "step": 1940
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002971148022391484,
+      "loss": 0.2562,
+      "step": 1950
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00029708727172616313,
+      "loss": 0.2717,
+      "step": 1960
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00029705974121317793,
+      "loss": 0.283,
+      "step": 1970
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00029703221070019267,
+      "loss": 0.2728,
+      "step": 1980
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00029700468018720747,
+      "loss": 0.2867,
+      "step": 1990
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00029697714967422226,
+      "loss": 0.2256,
+      "step": 2000
+    },
+    {
+      "epoch": 0.55,
+      "eval_loss": 0.18709534406661987,
+      "eval_runtime": 1029.6709,
+      "eval_samples_per_second": 11.346,
+      "eval_steps_per_second": 1.419,
+      "eval_wer": 0.46329482397889565,
+      "step": 2000
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.000296949619161237,
+      "loss": 0.2914,
+      "step": 2010
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002969220886482518,
+      "loss": 0.3636,
+      "step": 2020
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00029689455813526654,
+      "loss": 0.2804,
+      "step": 2030
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00029686702762228134,
+      "loss": 0.313,
+      "step": 2040
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00029683949710929613,
+      "loss": 0.2664,
+      "step": 2050
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002968119665963109,
+      "loss": 0.2472,
+      "step": 2060
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00029678443608332567,
+      "loss": 0.2605,
+      "step": 2070
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002967569055703404,
+      "loss": 0.2489,
+      "step": 2080
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002967293750573552,
+      "loss": 0.2753,
+      "step": 2090
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00029670184454437,
+      "loss": 0.2697,
+      "step": 2100
+    },
+    {
+      "epoch": 0.58,
+      "eval_loss": 0.19330532848834991,
+      "eval_runtime": 1030.6142,
+      "eval_samples_per_second": 11.336,
+      "eval_steps_per_second": 1.418,
+      "eval_wer": 0.4696283998908396,
+      "step": 2100
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00029667431403138474,
+      "loss": 0.2843,
+      "step": 2110
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00029664678351839954,
+      "loss": 0.2692,
+      "step": 2120
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002966192530054143,
+      "loss": 0.2718,
+      "step": 2130
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002965917224924291,
+      "loss": 0.2774,
+      "step": 2140
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00029656419197944387,
+      "loss": 0.2596,
+      "step": 2150
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002965366614664586,
+      "loss": 0.272,
+      "step": 2160
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002965091309534734,
+      "loss": 0.3172,
+      "step": 2170
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002964816004404882,
+      "loss": 0.3228,
+      "step": 2180
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00029645406992750295,
+      "loss": 0.2719,
+      "step": 2190
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00029642653941451774,
+      "loss": 0.233,
+      "step": 2200
+    },
+    {
+      "epoch": 0.6,
+      "eval_loss": 0.18349601328372955,
+      "eval_runtime": 1030.0621,
+      "eval_samples_per_second": 11.342,
+      "eval_steps_per_second": 1.418,
+      "eval_wer": 0.45199217683980714,
+      "step": 2200
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0002963990089015325,
+      "loss": 0.2861,
+      "step": 2210
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0002963714783885473,
+      "loss": 0.2984,
+      "step": 2220
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0002963439478755621,
+      "loss": 0.2395,
+      "step": 2230
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0002963164173625768,
+      "loss": 0.2607,
+      "step": 2240
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0002962888868495916,
+      "loss": 0.2739,
+      "step": 2250
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00029626135633660636,
+      "loss": 0.2549,
+      "step": 2260
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00029623382582362115,
+      "loss": 0.266,
+      "step": 2270
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00029620629531063595,
+      "loss": 0.2786,
+      "step": 2280
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0002961787647976507,
+      "loss": 0.2524,
+      "step": 2290
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0002961512342846655,
+      "loss": 0.2078,
+      "step": 2300
+    },
+    {
+      "epoch": 0.63,
+      "eval_loss": 0.1761041283607483,
+      "eval_runtime": 1030.565,
+      "eval_samples_per_second": 11.337,
+      "eval_steps_per_second": 1.418,
+      "eval_wer": 0.4481601928499955,
+      "step": 2300
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0002961237037716803,
+      "loss": 0.2331,
+      "step": 2310
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.000296096173258695,
+      "loss": 0.2197,
+      "step": 2320
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0002960686427457098,
+      "loss": 0.2595,
+      "step": 2330
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00029604111223272456,
+      "loss": 0.2571,
+      "step": 2340
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00029601358171973935,
+      "loss": 0.2074,
+      "step": 2350
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00029598605120675415,
+      "loss": 0.2587,
+      "step": 2360
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0002959585206937689,
+      "loss": 0.246,
+      "step": 2370
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0002959309901807837,
+      "loss": 0.2964,
+      "step": 2380
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00029590345966779843,
+      "loss": 0.2453,
+      "step": 2390
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0002958759291548132,
+      "loss": 0.2721,
+      "step": 2400
+    },
+    {
+      "epoch": 0.66,
+      "eval_loss": 0.17422164976596832,
+      "eval_runtime": 1036.0048,
+      "eval_samples_per_second": 11.277,
+      "eval_steps_per_second": 1.41,
+      "eval_wer": 0.44059856272173203,
+      "step": 2400
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.000295848398641828,
+      "loss": 0.2198,
+      "step": 2410
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00029582086812884276,
+      "loss": 0.2405,
+      "step": 2420
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00029579333761585756,
+      "loss": 0.2334,
+      "step": 2430
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0002957658071028723,
+      "loss": 0.2894,
+      "step": 2440
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0002957382765898871,
+      "loss": 0.2534,
+      "step": 2450
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0002957107460769019,
+      "loss": 0.2769,
+      "step": 2460
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00029568321556391663,
+      "loss": 0.284,
+      "step": 2470
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00029565568505093143,
+      "loss": 0.2619,
+      "step": 2480
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00029562815453794617,
+      "loss": 0.4885,
+      "step": 2490
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00029560062402496097,
+      "loss": 0.2807,
+      "step": 2500
+    },
+    {
+      "epoch": 0.69,
+      "eval_loss": 0.1716814488172531,
+      "eval_runtime": 1034.3139,
+      "eval_samples_per_second": 11.295,
+      "eval_steps_per_second": 1.413,
+      "eval_wer": 0.4375966524151733,
+      "step": 2500
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00029557309351197576,
+      "loss": 0.264,
+      "step": 2510
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0002955455629989905,
+      "loss": 0.2517,
+      "step": 2520
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0002955180324860053,
+      "loss": 0.2351,
+      "step": 2530
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00029549050197302004,
+      "loss": 0.2626,
+      "step": 2540
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00029546297146003484,
+      "loss": 0.2721,
+      "step": 2550
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00029543544094704963,
+      "loss": 0.2779,
+      "step": 2560
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0002954079104340644,
+      "loss": 0.2684,
+      "step": 2570
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00029538037992107917,
+      "loss": 0.2444,
+      "step": 2580
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00029535284940809397,
+      "loss": 0.2643,
+      "step": 2590
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0002953253188951087,
+      "loss": 0.2397,
+      "step": 2600
+    },
+    {
+      "epoch": 0.71,
+      "eval_loss": 0.178777813911438,
+      "eval_runtime": 1033.0369,
+      "eval_samples_per_second": 11.309,
+      "eval_steps_per_second": 1.414,
+      "eval_wer": 0.4444191758391704,
+      "step": 2600
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0002952977883821235,
+      "loss": 0.2436,
+      "step": 2610
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0002952702578691383,
+      "loss": 0.2767,
+      "step": 2620
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00029524272735615304,
+      "loss": 0.2188,
+      "step": 2630
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00029521519684316784,
+      "loss": 0.2281,
+      "step": 2640
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0002951876663301826,
+      "loss": 0.265,
+      "step": 2650
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00029516013581719737,
+      "loss": 0.27,
+      "step": 2660
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00029513260530421217,
+      "loss": 0.2213,
+      "step": 2670
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0002951050747912269,
+      "loss": 0.2646,
+      "step": 2680
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0002950775442782417,
+      "loss": 0.2378,
+      "step": 2690
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00029505001376525645,
+      "loss": 0.2672,
+      "step": 2700
+    },
+    {
+      "epoch": 0.74,
+      "eval_loss": 0.15344244241714478,
+      "eval_runtime": 1033.7172,
+      "eval_samples_per_second": 11.302,
+      "eval_steps_per_second": 1.413,
+      "eval_wer": 0.4126262166833439,
+      "step": 2700
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00029502248325227124,
+      "loss": 0.2552,
+      "step": 2710
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00029499495273928604,
+      "loss": 0.2254,
+      "step": 2720
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0002949674222263008,
+      "loss": 0.252,
+      "step": 2730
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0002949398917133156,
+      "loss": 0.2361,
+      "step": 2740
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0002949123612003303,
+      "loss": 0.2479,
+      "step": 2750
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0002948848306873451,
+      "loss": 0.2512,
+      "step": 2760
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0002948573001743599,
+      "loss": 0.2675,
+      "step": 2770
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00029482976966137465,
+      "loss": 0.2157,
+      "step": 2780
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00029480223914838945,
+      "loss": 0.2625,
+      "step": 2790
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.0002947747086354042,
+      "loss": 0.1954,
+      "step": 2800
+    },
+    {
+      "epoch": 0.77,
+      "eval_loss": 0.1593170017004013,
+      "eval_runtime": 1038.0053,
+      "eval_samples_per_second": 11.255,
+      "eval_steps_per_second": 1.408,
+      "eval_wer": 0.40848721913945235,
+      "step": 2800
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.000294747178122419,
+      "loss": 0.23,
+      "step": 2810
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.0002947196476094338,
+      "loss": 0.2252,
+      "step": 2820
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.0002946921170964485,
+      "loss": 0.2564,
+      "step": 2830
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.0002946645865834633,
+      "loss": 0.2578,
+      "step": 2840
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00029463705607047806,
+      "loss": 0.2118,
+      "step": 2850
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00029460952555749285,
+      "loss": 0.2366,
+      "step": 2860
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00029458199504450765,
+      "loss": 0.2228,
+      "step": 2870
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.0002945544645315224,
+      "loss": 0.252,
+      "step": 2880
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.0002945269340185372,
+      "loss": 0.2554,
+      "step": 2890
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00029449940350555193,
+      "loss": 0.2753,
+      "step": 2900
+    },
+    {
+      "epoch": 0.79,
+      "eval_loss": 0.1575159728527069,
+      "eval_runtime": 1033.311,
+      "eval_samples_per_second": 11.306,
+      "eval_steps_per_second": 1.414,
+      "eval_wer": 0.4128081506413172,
+      "step": 2900
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.0002944718729925667,
+      "loss": 0.2448,
+      "step": 2910
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.0002944443424795815,
+      "loss": 0.2356,
+      "step": 2920
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00029441681196659626,
+      "loss": 0.2523,
+      "step": 2930
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.00029438928145361106,
+      "loss": 0.2386,
+      "step": 2940
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.0002943617509406258,
+      "loss": 0.2327,
+      "step": 2950
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.0002943342204276406,
+      "loss": 0.2543,
+      "step": 2960
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.0002943066899146554,
+      "loss": 0.2409,
+      "step": 2970
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 0.0002942791594016702,
+      "loss": 0.263,
+      "step": 2980
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 0.00029425162888868493,
+      "loss": 0.231,
+      "step": 2990
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 0.0002942240983756997,
+      "loss": 0.1975,
+      "step": 3000
+    },
+    {
+      "epoch": 0.82,
+      "eval_loss": 0.14351560175418854,
+      "eval_runtime": 1046.2091,
+      "eval_samples_per_second": 11.167,
+      "eval_steps_per_second": 1.396,
+      "eval_wer": 0.39260211043391247,
+      "step": 3000
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 0.00029419656786271447,
+      "loss": 0.2252,
+      "step": 3010
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.00029416903734972926,
+      "loss": 0.2248,
+      "step": 3020
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.00029414150683674406,
+      "loss": 0.2285,
+      "step": 3030
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.0002941139763237588,
+      "loss": 0.2458,
+      "step": 3040
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.0002940864458107736,
+      "loss": 0.2482,
+      "step": 3050
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.00029405891529778834,
+      "loss": 0.2798,
+      "step": 3060
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.00029403138478480313,
+      "loss": 0.2338,
+      "step": 3070
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.00029400385427181793,
+      "loss": 0.2369,
+      "step": 3080
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.00029397632375883267,
+      "loss": 0.2208,
+      "step": 3090
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.00029394879324584747,
+      "loss": 0.2282,
+      "step": 3100
+    },
+    {
+      "epoch": 0.85,
+      "eval_loss": 0.17016960680484772,
+      "eval_runtime": 1038.5865,
+      "eval_samples_per_second": 11.249,
+      "eval_steps_per_second": 1.407,
+      "eval_wer": 0.4270672245974711,
+      "step": 3100
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.0002939212627328622,
+      "loss": 0.2531,
+      "step": 3110
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.000293893732219877,
+      "loss": 0.3018,
+      "step": 3120
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 0.0002938662017068918,
+      "loss": 0.2405,
+      "step": 3130
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 0.00029383867119390654,
+      "loss": 0.2549,
+      "step": 3140
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 0.00029381114068092134,
+      "loss": 0.2245,
+      "step": 3150
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.0002937836101679361,
+      "loss": 0.2354,
+      "step": 3160
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.00029375607965495087,
+      "loss": 0.2367,
+      "step": 3170
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.00029372854914196567,
+      "loss": 0.2452,
+      "step": 3180
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.0002937010186289804,
+      "loss": 0.2382,
+      "step": 3190
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.0002936734881159952,
+      "loss": 0.2372,
+      "step": 3200
+    },
+    {
+      "epoch": 0.88,
+      "eval_loss": 0.1416802853345871,
+      "eval_runtime": 1032.3105,
+      "eval_samples_per_second": 11.317,
+      "eval_steps_per_second": 1.415,
+      "eval_wer": 0.3964682070408442,
+      "step": 3200
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.00029364595760300995,
+      "loss": 0.2478,
+      "step": 3210
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.00029361842709002474,
+      "loss": 0.2127,
+      "step": 3220
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.00029359089657703954,
+      "loss": 0.2021,
+      "step": 3230
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.0002935633660640543,
+      "loss": 0.2422,
+      "step": 3240
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.0002935358355510691,
+      "loss": 0.2399,
+      "step": 3250
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.0002935083050380838,
+      "loss": 0.2405,
+      "step": 3260
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.0002934807745250986,
+      "loss": 0.1949,
+      "step": 3270
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.0002934532440121134,
+      "loss": 0.2234,
+      "step": 3280
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.0002934257134991282,
+      "loss": 0.2403,
+      "step": 3290
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.00029339818298614295,
+      "loss": 0.229,
+      "step": 3300
+    },
+    {
+      "epoch": 0.9,
+      "eval_loss": 0.13848544657230377,
+      "eval_runtime": 1033.0662,
+      "eval_samples_per_second": 11.309,
+      "eval_steps_per_second": 1.414,
+      "eval_wer": 0.38596152096788866,
+      "step": 3300
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.0002933706524731577,
+      "loss": 0.2132,
+      "step": 3310
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.0002933431219601725,
+      "loss": 0.5408,
+      "step": 3320
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.0002933155914471873,
+      "loss": 0.2344,
+      "step": 3330
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.0002932880609342021,
+      "loss": 0.2454,
+      "step": 3340
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.0002932605304212168,
+      "loss": 0.2326,
+      "step": 3350
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.0002932329999082316,
+      "loss": 0.2076,
+      "step": 3360
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.00029320546939524635,
+      "loss": 0.3867,
+      "step": 3370
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.00029317793888226115,
+      "loss": 0.2038,
+      "step": 3380
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.00029315040836927595,
+      "loss": 0.2371,
+      "step": 3390
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.0002931228778562907,
+      "loss": 0.2332,
+      "step": 3400
+    },
+    {
+      "epoch": 0.93,
+      "eval_loss": 0.13254162669181824,
+      "eval_runtime": 1032.4462,
+      "eval_samples_per_second": 11.316,
+      "eval_steps_per_second": 1.415,
+      "eval_wer": 0.380503502228691,
+      "step": 3400
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.0002930953473433055,
+      "loss": 0.3243,
+      "step": 3410
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.0002930678168303202,
+      "loss": 0.2216,
+      "step": 3420
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.000293040286317335,
+      "loss": 0.2173,
+      "step": 3430
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.0002930127558043498,
+      "loss": 0.2095,
+      "step": 3440
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.00029298522529136456,
+      "loss": 0.2069,
+      "step": 3450
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.00029295769477837935,
+      "loss": 0.2326,
+      "step": 3460
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.0002929301642653941,
+      "loss": 0.2067,
+      "step": 3470
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.0002929026337524089,
+      "loss": 0.2236,
+      "step": 3480
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.0002928751032394237,
+      "loss": 0.229,
+      "step": 3490
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.00029284757272643843,
+      "loss": 0.2074,
+      "step": 3500
+    },
+    {
+      "epoch": 0.96,
+      "eval_loss": 0.13085076212882996,
+      "eval_runtime": 1031.5246,
+      "eval_samples_per_second": 11.326,
+      "eval_steps_per_second": 1.416,
+      "eval_wer": 0.375,
+      "step": 3500
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.0002928200422134532,
+      "loss": 0.2062,
+      "step": 3510
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.00029279251170046797,
+      "loss": 0.3621,
+      "step": 3520
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 0.00029276498118748276,
+      "loss": 0.2127,
+      "step": 3530
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 0.00029273745067449756,
+      "loss": 0.2343,
+      "step": 3540
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 0.0002927099201615123,
+      "loss": 0.1566,
+      "step": 3550
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.0002926823896485271,
+      "loss": 0.2281,
+      "step": 3560
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.00029265485913554184,
+      "loss": 0.2268,
+      "step": 3570
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.00029262732862255663,
+      "loss": 0.1878,
+      "step": 3580
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.00029259979810957143,
+      "loss": 0.1967,
+      "step": 3590
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.00029257226759658617,
+      "loss": 0.1937,
+      "step": 3600
+    },
+    {
+      "epoch": 0.99,
+      "eval_loss": 0.13056674599647522,
+      "eval_runtime": 1033.4241,
+      "eval_samples_per_second": 11.305,
+      "eval_steps_per_second": 1.414,
+      "eval_wer": 0.37841126171199857,
+      "step": 3600
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.00029254473708360097,
+      "loss": 0.204,
+      "step": 3610
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.0002925172065706157,
+      "loss": 0.2036,
+      "step": 3620
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.0002924896760576305,
+      "loss": 0.2268,
+      "step": 3630
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 0.0002924621455446453,
+      "loss": 0.2303,
+      "step": 3640
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 0.0002924346150316601,
+      "loss": 0.2061,
+      "step": 3650
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 0.00029240708451867484,
+      "loss": 0.222,
+      "step": 3660
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 0.0002923795540056896,
+      "loss": 0.1853,
+      "step": 3670
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 0.00029235202349270437,
+      "loss": 0.2252,
+      "step": 3680
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 0.00029232449297971917,
+      "loss": 0.2318,
+      "step": 3690
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 0.00029229696246673396,
+      "loss": 0.2211,
+      "step": 3700
+    },
+    {
+      "epoch": 1.01,
+      "eval_loss": 0.14021088182926178,
+      "eval_runtime": 1031.5194,
+      "eval_samples_per_second": 11.326,
+      "eval_steps_per_second": 1.416,
+      "eval_wer": 0.383619121258983,
+      "step": 3700
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 0.0002922694319537487,
+      "loss": 0.1938,
+      "step": 3710
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 0.00029224190144076345,
+      "loss": 0.2073,
+      "step": 3720
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 0.00029221437092777824,
+      "loss": 0.2023,
+      "step": 3730
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 0.00029218684041479304,
+      "loss": 0.2171,
+      "step": 3740
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 0.00029215930990180783,
+      "loss": 0.1932,
+      "step": 3750
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 0.0002921317793888226,
+      "loss": 0.1921,
+      "step": 3760
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 0.00029210424887583737,
+      "loss": 0.229,
+      "step": 3770
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 0.0002920767183628521,
+      "loss": 0.1971,
+      "step": 3780
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 0.0002920491878498669,
+      "loss": 0.2161,
+      "step": 3790
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 0.0002920216573368817,
+      "loss": 0.1928,
+      "step": 3800
+    },
+    {
+      "epoch": 1.04,
+      "eval_loss": 0.13391314446926117,
+      "eval_runtime": 1039.3763,
+      "eval_samples_per_second": 11.24,
+      "eval_steps_per_second": 1.406,
+      "eval_wer": 0.3714409169471482,
+      "step": 3800
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 0.00029199412682389645,
+      "loss": 0.227,
+      "step": 3810
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 0.00029196659631091124,
+      "loss": 0.1845,
+      "step": 3820
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 0.000291939065797926,
+      "loss": 0.2082,
+      "step": 3830
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 0.0002919115352849408,
+      "loss": 0.2131,
+      "step": 3840
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 0.0002918840047719556,
+      "loss": 0.1836,
+      "step": 3850
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 0.0002918564742589703,
+      "loss": 0.2096,
+      "step": 3860
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 0.0002918289437459851,
+      "loss": 0.198,
+      "step": 3870
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 0.00029180141323299985,
+      "loss": 0.194,
+      "step": 3880
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 0.00029177388272001465,
+      "loss": 0.2517,
+      "step": 3890
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 0.00029174635220702945,
+      "loss": 0.176,
+      "step": 3900
+    },
+    {
+      "epoch": 1.07,
+      "eval_loss": 0.13185542821884155,
+      "eval_runtime": 1034.2276,
+      "eval_samples_per_second": 11.296,
+      "eval_steps_per_second": 1.413,
+      "eval_wer": 0.37426089329573364,
+      "step": 3900
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 0.0002917188216940442,
+      "loss": 0.2084,
+      "step": 3910
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 0.000291691291181059,
+      "loss": 0.1942,
+      "step": 3920
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 0.0002916637606680737,
+      "loss": 0.2091,
+      "step": 3930
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 0.0002916362301550885,
+      "loss": 0.2117,
+      "step": 3940
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 0.0002916086996421033,
+      "loss": 0.1762,
+      "step": 3950
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 0.0002915811691291181,
+      "loss": 0.2032,
+      "step": 3960
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 0.00029155363861613285,
+      "loss": 0.2233,
+      "step": 3970
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 0.0002915261081031476,
+      "loss": 0.1855,
+      "step": 3980
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 0.0002914985775901624,
+      "loss": 0.1992,
+      "step": 3990
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 0.0002914710470771772,
+      "loss": 0.1818,
+      "step": 4000
+    },
+    {
+      "epoch": 1.1,
+      "eval_loss": 0.1326366364955902,
+      "eval_runtime": 1036.1093,
+      "eval_samples_per_second": 11.276,
+      "eval_steps_per_second": 1.41,
+      "eval_wer": 0.3759892658964796,
+      "step": 4000
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 0.000291443516564192,
+      "loss": 0.1836,
+      "step": 4010
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 0.0002914159860512067,
+      "loss": 0.2213,
+      "step": 4020
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 0.00029138845553822147,
+      "loss": 0.1852,
+      "step": 4030
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 0.00029136092502523626,
+      "loss": 0.2357,
+      "step": 4040
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 0.00029133339451225106,
+      "loss": 0.1824,
+      "step": 4050
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 0.00029130586399926585,
+      "loss": 0.1947,
+      "step": 4060
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 0.0002912783334862806,
+      "loss": 0.2006,
+      "step": 4070
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 0.00029125080297329534,
+      "loss": 0.1958,
+      "step": 4080
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 0.00029122327246031013,
+      "loss": 0.2205,
+      "step": 4090
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 0.00029119574194732493,
+      "loss": 0.2191,
+      "step": 4100
+    },
+    {
+      "epoch": 1.12,
+      "eval_loss": 0.13117532432079315,
+      "eval_runtime": 1035.211,
+      "eval_samples_per_second": 11.286,
+      "eval_steps_per_second": 1.411,
+      "eval_wer": 0.37676248521786593,
+      "step": 4100
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 0.0002911682114343397,
+      "loss": 0.1991,
+      "step": 4110
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 0.00029114068092135447,
+      "loss": 0.2014,
+      "step": 4120
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 0.00029111315040836926,
+      "loss": 0.2254,
+      "step": 4130
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 0.000291085619895384,
+      "loss": 0.2056,
+      "step": 4140
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 0.0002910580893823988,
+      "loss": 0.1819,
+      "step": 4150
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 0.0002910305588694136,
+      "loss": 0.2095,
+      "step": 4160
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 0.00029100302835642834,
+      "loss": 0.2117,
+      "step": 4170
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 0.00029097549784344313,
+      "loss": 0.1921,
+      "step": 4180
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 0.0002909479673304579,
+      "loss": 0.1961,
+      "step": 4190
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 0.00029092043681747267,
+      "loss": 0.1848,
+      "step": 4200
+    },
+    {
+      "epoch": 1.15,
+      "eval_loss": 0.12278643250465393,
+      "eval_runtime": 1034.1658,
+      "eval_samples_per_second": 11.297,
+      "eval_steps_per_second": 1.413,
+      "eval_wer": 0.35989948148821976,
+      "step": 4200
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 0.00029089290630448746,
+      "loss": 0.1991,
+      "step": 4210
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 0.0002908653757915022,
+      "loss": 0.1814,
+      "step": 4220
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 0.000290837845278517,
+      "loss": 0.1836,
+      "step": 4230
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 0.00029081031476553174,
+      "loss": 0.1921,
+      "step": 4240
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 0.00029078278425254654,
+      "loss": 0.1855,
+      "step": 4250
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 0.00029075525373956133,
+      "loss": 0.2016,
+      "step": 4260
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 0.00029072772322657613,
+      "loss": 0.205,
+      "step": 4270
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 0.00029070019271359087,
+      "loss": 0.1881,
+      "step": 4280
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 0.0002906726622006056,
+      "loss": 0.2192,
+      "step": 4290
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 0.0002906451316876204,
+      "loss": 0.1671,
+      "step": 4300
+    },
+    {
+      "epoch": 1.18,
+      "eval_loss": 0.12893985211849213,
+      "eval_runtime": 1038.4465,
+      "eval_samples_per_second": 11.25,
+      "eval_steps_per_second": 1.407,
+      "eval_wer": 0.37314654780314743,
+      "step": 4300
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 0.0002906176011746352,
+      "loss": 0.2075,
+      "step": 4310
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 0.00029059007066165,
+      "loss": 0.225,
+      "step": 4320
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 0.00029056254014866474,
+      "loss": 0.2183,
+      "step": 4330
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 0.0002905350096356795,
+      "loss": 0.2136,
+      "step": 4340
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 0.0002905074791226943,
+      "loss": 0.198,
+      "step": 4350
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 0.0002904799486097091,
+      "loss": 0.2346,
+      "step": 4360
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 0.00029045241809672387,
+      "loss": 0.2207,
+      "step": 4370
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 0.0002904248875837386,
+      "loss": 0.1996,
+      "step": 4380
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 0.00029039735707075335,
+      "loss": 0.1872,
+      "step": 4390
+    },
+    {
+      "epoch": 1.21,
+      "learning_rate": 0.00029036982655776815,
+      "loss": 0.1759,
+      "step": 4400
+    },
+    {
+      "epoch": 1.21,
+      "eval_loss": 0.1245460957288742,
+      "eval_runtime": 1036.0049,
+      "eval_samples_per_second": 11.277,
+      "eval_steps_per_second": 1.41,
+      "eval_wer": 0.36324251796597834,
+      "step": 4400
+    },
+    {
+      "epoch": 1.21,
+      "learning_rate": 0.00029034229604478295,
+      "loss": 0.1972,
+      "step": 4410
+    },
+    {
+      "epoch": 1.21,
+      "learning_rate": 0.00029031476553179774,
+      "loss": 0.2011,
+      "step": 4420
+    },
+    {
+      "epoch": 1.21,
+      "learning_rate": 0.0002902872350188125,
+      "loss": 0.1978,
+      "step": 4430
+    },
+    {
+      "epoch": 1.22,
+      "learning_rate": 0.0002902597045058272,
+      "loss": 0.1881,
+      "step": 4440
+    },
+    {
+      "epoch": 1.22,
+      "learning_rate": 0.000290232173992842,
+      "loss": 0.1781,
+      "step": 4450
+    },
+    {
+      "epoch": 1.22,
+      "learning_rate": 0.0002902046434798568,
+      "loss": 0.2026,
+      "step": 4460
+    },
+    {
+      "epoch": 1.22,
+      "learning_rate": 0.0002901771129668716,
+      "loss": 0.2026,
+      "step": 4470
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 0.00029014958245388635,
+      "loss": 0.2007,
+      "step": 4480
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 0.00029012205194090115,
+      "loss": 0.4222,
+      "step": 4490
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 0.0002900945214279159,
+      "loss": 0.1783,
+      "step": 4500
+    },
+    {
+      "epoch": 1.23,
+      "eval_loss": 0.12668462097644806,
+      "eval_runtime": 1032.5946,
+      "eval_samples_per_second": 11.314,
+      "eval_steps_per_second": 1.415,
+      "eval_wer": 0.35453242972800875,
+      "step": 4500
+    },
+    {
+      "epoch": 1.24,
+      "learning_rate": 0.0002900669909149307,
+      "loss": 0.1804,
+      "step": 4510
+    },
+    {
+      "epoch": 1.24,
+      "learning_rate": 0.0002900394604019455,
+      "loss": 0.1893,
+      "step": 4520
+    },
+    {
+      "epoch": 1.24,
+      "learning_rate": 0.0002900119298889602,
+      "loss": 0.1818,
+      "step": 4530
+    },
+    {
+      "epoch": 1.24,
+      "learning_rate": 0.000289984399375975,
+      "loss": 0.1815,
+      "step": 4540
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 0.00028995686886298976,
+      "loss": 0.2069,
+      "step": 4550
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 0.00028992933835000456,
+      "loss": 0.2155,
+      "step": 4560
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 0.00028990180783701935,
+      "loss": 0.1816,
+      "step": 4570
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 0.0002898742773240341,
+      "loss": 0.1832,
+      "step": 4580
+    },
+    {
+      "epoch": 1.26,
+      "learning_rate": 0.0002898467468110489,
+      "loss": 0.1775,
+      "step": 4590
+    },
+    {
+      "epoch": 1.26,
+      "learning_rate": 0.00028981921629806363,
+      "loss": 0.1576,
+      "step": 4600
+    },
+    {
+      "epoch": 1.26,
+      "eval_loss": 0.1280120611190796,
+      "eval_runtime": 1036.0204,
+      "eval_samples_per_second": 11.277,
+      "eval_steps_per_second": 1.41,
+      "eval_wer": 0.3581028836532339,
+      "step": 4600
+    },
+    {
+      "epoch": 1.26,
+      "learning_rate": 0.00028979168578507843,
+      "loss": 0.1803,
+      "step": 4610
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 0.0002897641552720932,
+      "loss": 0.1562,
+      "step": 4620
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 0.000289736624759108,
+      "loss": 0.1841,
+      "step": 4630
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 0.00028970909424612276,
+      "loss": 0.1993,
+      "step": 4640
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 0.0002896815637331375,
+      "loss": 0.1856,
+      "step": 4650
+    },
+    {
+      "epoch": 1.28,
+      "learning_rate": 0.0002896540332201523,
+      "loss": 0.1863,
+      "step": 4660
+    },
+    {
+      "epoch": 1.28,
+      "learning_rate": 0.0002896265027071671,
+      "loss": 0.197,
+      "step": 4670
+    },
+    {
+      "epoch": 1.28,
+      "learning_rate": 0.0002895989721941819,
+      "loss": 0.1745,
+      "step": 4680
+    },
+    {
+      "epoch": 1.29,
+      "learning_rate": 0.00028957144168119663,
+      "loss": 0.1993,
+      "step": 4690
+    },
+    {
+      "epoch": 1.29,
+      "learning_rate": 0.0002895439111682114,
+      "loss": 0.1726,
+      "step": 4700
+    },
+    {
+      "epoch": 1.29,
+      "eval_loss": 0.12151551246643066,
+      "eval_runtime": 1035.4623,
+      "eval_samples_per_second": 11.283,
+      "eval_steps_per_second": 1.411,
+      "eval_wer": 0.35406622396070225,
+      "step": 4700
+    },
+    {
+      "epoch": 1.29,
+      "learning_rate": 0.00028951638065522617,
+      "loss": 0.1771,
+      "step": 4710
+    },
+    {
+      "epoch": 1.29,
+      "learning_rate": 0.00028948885014224096,
+      "loss": 0.177,
+      "step": 4720
+    },
+    {
+      "epoch": 1.3,
+      "learning_rate": 0.00028946131962925576,
+      "loss": 0.1863,
+      "step": 4730
+    },
+    {
+      "epoch": 1.3,
+      "learning_rate": 0.0002894337891162705,
+      "loss": 0.1879,
+      "step": 4740
+    },
+    {
+      "epoch": 1.3,
+      "learning_rate": 0.00028940625860328524,
+      "loss": 0.1581,
+      "step": 4750
+    },
+    {
+      "epoch": 1.3,
+      "learning_rate": 0.00028937872809030004,
+      "loss": 0.1907,
+      "step": 4760
+    },
+    {
+      "epoch": 1.31,
+      "learning_rate": 0.00028935119757731483,
+      "loss": 0.1756,
+      "step": 4770
+    },
+    {
+      "epoch": 1.31,
+      "learning_rate": 0.00028932366706432963,
+      "loss": 0.1793,
+      "step": 4780
+    },
+    {
+      "epoch": 1.31,
+      "learning_rate": 0.00028929613655134437,
+      "loss": 0.183,
+      "step": 4790
+    },
+    {
+      "epoch": 1.32,
+      "learning_rate": 0.0002892686060383591,
+      "loss": 0.1702,
+      "step": 4800
+    },
+    {
+      "epoch": 1.32,
+      "eval_loss": 0.13416741788387299,
+      "eval_runtime": 1029.9454,
+      "eval_samples_per_second": 11.343,
+      "eval_steps_per_second": 1.419,
+      "eval_wer": 0.3690075502592559,
+      "step": 4800
+    },
+    {
+      "epoch": 1.32,
+      "learning_rate": 0.0002892410755253739,
+      "loss": 0.1963,
+      "step": 4810
+    },
+    {
+      "epoch": 1.32,
+      "learning_rate": 0.0002892135450123887,
+      "loss": 0.1787,
+      "step": 4820
+    },
+    {
+      "epoch": 1.32,
+      "learning_rate": 0.0002891860144994035,
+      "loss": 0.1877,
+      "step": 4830
+    },
+    {
+      "epoch": 1.33,
+      "learning_rate": 0.00028915848398641824,
+      "loss": 0.1848,
+      "step": 4840
+    },
+    {
+      "epoch": 1.33,
+      "learning_rate": 0.00028913095347343304,
+      "loss": 0.1763,
+      "step": 4850
+    },
+    {
+      "epoch": 1.33,
+      "learning_rate": 0.0002891034229604478,
+      "loss": 0.2131,
+      "step": 4860
+    },
+    {
+      "epoch": 1.33,
+      "learning_rate": 0.0002890758924474626,
+      "loss": 0.2014,
+      "step": 4870
+    },
+    {
+      "epoch": 1.34,
+      "learning_rate": 0.00028904836193447737,
+      "loss": 0.1825,
+      "step": 4880
+    },
+    {
+      "epoch": 1.34,
+      "learning_rate": 0.0002890208314214921,
+      "loss": 0.1761,
+      "step": 4890
+    },
+    {
+      "epoch": 1.34,
+      "learning_rate": 0.0002889933009085069,
+      "loss": 0.2667,
+      "step": 4900
+    },
+    {
+      "epoch": 1.34,
+      "eval_loss": 0.12561985850334167,
+      "eval_runtime": 1032.5452,
+      "eval_samples_per_second": 11.315,
+      "eval_steps_per_second": 1.415,
+      "eval_wer": 0.3635381606476849,
+      "step": 4900
+    },
+    {
+      "epoch": 1.35,
+      "learning_rate": 0.00028896577039552165,
+      "loss": 0.1961,
+      "step": 4910
+    },
+    {
+      "epoch": 1.35,
+      "learning_rate": 0.00028893823988253645,
+      "loss": 0.4355,
+      "step": 4920
+    },
+    {
+      "epoch": 1.35,
+      "learning_rate": 0.00028891070936955124,
+      "loss": 0.1876,
+      "step": 4930
+    },
+    {
+      "epoch": 1.35,
+      "learning_rate": 0.00028888317885656604,
+      "loss": 0.182,
+      "step": 4940
+    },
+    {
+      "epoch": 1.36,
+      "learning_rate": 0.0002888556483435808,
+      "loss": 0.1972,
+      "step": 4950
+    },
+    {
+      "epoch": 1.36,
+      "learning_rate": 0.0002888281178305955,
+      "loss": 0.1987,
+      "step": 4960
+    },
+    {
+      "epoch": 1.36,
+      "learning_rate": 0.0002888005873176103,
+      "loss": 0.1898,
+      "step": 4970
+    },
+    {
+      "epoch": 1.36,
+      "learning_rate": 0.0002887730568046251,
+      "loss": 0.1877,
+      "step": 4980
+    },
+    {
+      "epoch": 1.37,
+      "learning_rate": 0.0002887455262916399,
+      "loss": 0.1746,
+      "step": 4990
+    },
+    {
+      "epoch": 1.37,
+      "learning_rate": 0.00028871799577865465,
+      "loss": 0.232,
+      "step": 5000
+    },
+    {
+      "epoch": 1.37,
+      "eval_loss": 0.12278987467288971,
+      "eval_runtime": 1033.2314,
+      "eval_samples_per_second": 11.307,
+      "eval_steps_per_second": 1.414,
+      "eval_wer": 0.3580005458018739,
+      "step": 5000
+    },
+    {
+      "epoch": 1.37,
+      "learning_rate": 0.0002886904652656694,
+      "loss": 0.1778,
+      "step": 5010
+    },
+    {
+      "epoch": 1.38,
+      "learning_rate": 0.0002886629347526842,
+      "loss": 0.1862,
+      "step": 5020
+    },
+    {
+      "epoch": 1.38,
+      "learning_rate": 0.000288635404239699,
+      "loss": 0.1933,
+      "step": 5030
+    },
+    {
+      "epoch": 1.38,
+      "learning_rate": 0.0002886078737267138,
+      "loss": 0.1667,
+      "step": 5040
+    },
+    {
+      "epoch": 1.38,
+      "learning_rate": 0.0002885803432137285,
+      "loss": 0.1822,
+      "step": 5050
+    },
+    {
+      "epoch": 1.39,
+      "learning_rate": 0.00028855281270074326,
+      "loss": 0.1534,
+      "step": 5060
+    },
+    {
+      "epoch": 1.39,
+      "learning_rate": 0.00028852528218775806,
+      "loss": 0.3339,
+      "step": 5070
+    },
+    {
+      "epoch": 1.39,
+      "learning_rate": 0.00028849775167477285,
+      "loss": 0.179,
+      "step": 5080
+    },
+    {
+      "epoch": 1.39,
+      "learning_rate": 0.00028847022116178765,
+      "loss": 0.2182,
+      "step": 5090
+    },
+    {
+      "epoch": 1.4,
+      "learning_rate": 0.0002884426906488024,
+      "loss": 0.2012,
+      "step": 5100
+    },
+    {
+      "epoch": 1.4,
+      "eval_loss": 0.11825747042894363,
+      "eval_runtime": 1037.9251,
+      "eval_samples_per_second": 11.256,
+      "eval_steps_per_second": 1.408,
+      "eval_wer": 0.34755071409078503,
+      "step": 5100
+    },
+    {
+      "epoch": 1.4,
+      "learning_rate": 0.00028841516013581713,
+      "loss": 0.1791,
+      "step": 5110
+    },
+    {
+      "epoch": 1.4,
+      "learning_rate": 0.00028838762962283193,
+      "loss": 0.1971,
+      "step": 5120
+    },
+    {
+      "epoch": 1.41,
+      "learning_rate": 0.0002883600991098467,
+      "loss": 0.1989,
+      "step": 5130
+    },
+    {
+      "epoch": 1.41,
+      "learning_rate": 0.0002883325685968615,
+      "loss": 0.2292,
+      "step": 5140
+    },
+    {
+      "epoch": 1.41,
+      "learning_rate": 0.00028830503808387626,
+      "loss": 0.166,
+      "step": 5150
+    },
+    {
+      "epoch": 1.41,
+      "learning_rate": 0.00028827750757089106,
+      "loss": 0.1829,
+      "step": 5160
+    },
+    {
+      "epoch": 1.42,
+      "learning_rate": 0.0002882499770579058,
+      "loss": 0.1925,
+      "step": 5170
+    },
+    {
+      "epoch": 1.42,
+      "learning_rate": 0.0002882224465449206,
+      "loss": 0.1866,
+      "step": 5180
+    },
+    {
+      "epoch": 1.42,
+      "learning_rate": 0.0002881949160319354,
+      "loss": 0.2301,
+      "step": 5190
+    },
+    {
+      "epoch": 1.42,
+      "learning_rate": 0.00028816738551895013,
+      "loss": 0.1862,
+      "step": 5200
+    },
+    {
+      "epoch": 1.42,
+      "eval_loss": 0.12098130583763123,
+      "eval_runtime": 1035.7561,
+      "eval_samples_per_second": 11.28,
+      "eval_steps_per_second": 1.411,
+      "eval_wer": 0.3535317929591558,
+      "step": 5200
+    },
+    {
+      "epoch": 1.43,
+      "learning_rate": 0.00028813985500596493,
+      "loss": 0.1951,
+      "step": 5210
+    },
+    {
+      "epoch": 1.43,
+      "learning_rate": 0.00028811232449297967,
+      "loss": 0.1684,
+      "step": 5220
+    },
+    {
+      "epoch": 1.43,
+      "learning_rate": 0.00028808479397999446,
+      "loss": 0.1903,
+      "step": 5230
+    },
+    {
+      "epoch": 1.44,
+      "learning_rate": 0.00028805726346700926,
+      "loss": 0.1911,
+      "step": 5240
+    },
+    {
+      "epoch": 1.44,
+      "learning_rate": 0.000288029732954024,
+      "loss": 0.1821,
+      "step": 5250
+    },
+    {
+      "epoch": 1.44,
+      "learning_rate": 0.0002880022024410388,
+      "loss": 0.2087,
+      "step": 5260
+    },
+    {
+      "epoch": 1.44,
+      "learning_rate": 0.00028797467192805354,
+      "loss": 0.1847,
+      "step": 5270
+    },
+    {
+      "epoch": 1.45,
+      "learning_rate": 0.00028794714141506833,
+      "loss": 0.2127,
+      "step": 5280
+    },
+    {
+      "epoch": 1.45,
+      "learning_rate": 0.00028791961090208313,
+      "loss": 0.1968,
+      "step": 5290
+    },
+    {
+      "epoch": 1.45,
+      "learning_rate": 0.0002878920803890979,
+      "loss": 0.1947,
+      "step": 5300
+    },
+    {
+      "epoch": 1.45,
+      "eval_loss": 0.12408991158008575,
+      "eval_runtime": 1033.7998,
+      "eval_samples_per_second": 11.301,
+      "eval_steps_per_second": 1.413,
+      "eval_wer": 0.3513940689529701,
+      "step": 5300
+    },
+    {
+      "epoch": 1.45,
+      "learning_rate": 0.00028786454987611267,
+      "loss": 0.1778,
+      "step": 5310
+    },
+    {
+      "epoch": 1.46,
+      "learning_rate": 0.0002878370193631274,
+      "loss": 0.2122,
+      "step": 5320
+    },
+    {
+      "epoch": 1.46,
+      "learning_rate": 0.0002878094888501422,
+      "loss": 0.1956,
+      "step": 5330
+    },
+    {
+      "epoch": 1.46,
+      "learning_rate": 0.000287781958337157,
+      "loss": 0.1801,
+      "step": 5340
+    },
+    {
+      "epoch": 1.47,
+      "learning_rate": 0.0002877544278241718,
+      "loss": 0.1707,
+      "step": 5350
+    },
+    {
+      "epoch": 1.47,
+      "learning_rate": 0.00028772689731118654,
+      "loss": 0.1918,
+      "step": 5360
+    },
+    {
+      "epoch": 1.47,
+      "learning_rate": 0.0002876993667982013,
+      "loss": 0.1845,
+      "step": 5370
+    },
+    {
+      "epoch": 1.47,
+      "learning_rate": 0.0002876718362852161,
+      "loss": 0.1996,
+      "step": 5380
+    },
+    {
+      "epoch": 1.48,
+      "learning_rate": 0.00028764430577223087,
+      "loss": 0.1877,
+      "step": 5390
+    },
+    {
+      "epoch": 1.48,
+      "learning_rate": 0.00028761677525924567,
+      "loss": 0.1806,
+      "step": 5400
+    },
+    {
+      "epoch": 1.48,
+      "eval_loss": 0.11507146060466766,
+      "eval_runtime": 1039.021,
+      "eval_samples_per_second": 11.244,
+      "eval_steps_per_second": 1.406,
+      "eval_wer": 0.33993222960065494,
+      "step": 5400
+    },
+    {
+      "epoch": 1.48,
+      "learning_rate": 0.0002875892447462604,
+      "loss": 0.1864,
+      "step": 5410
+    },
+    {
+      "epoch": 1.49,
+      "learning_rate": 0.00028756171423327515,
+      "loss": 0.1771,
+      "step": 5420
+    },
+    {
+      "epoch": 1.49,
+      "learning_rate": 0.00028753418372028995,
+      "loss": 0.1856,
+      "step": 5430
+    },
+    {
+      "epoch": 1.49,
+      "learning_rate": 0.00028750665320730474,
+      "loss": 0.2536,
+      "step": 5440
+    },
+    {
+      "epoch": 1.49,
+      "learning_rate": 0.00028747912269431954,
+      "loss": 0.1733,
+      "step": 5450
+    },
+    {
+      "epoch": 1.5,
+      "learning_rate": 0.0002874515921813343,
+      "loss": 0.1679,
+      "step": 5460
+    },
+    {
+      "epoch": 1.5,
+      "learning_rate": 0.0002874240616683491,
+      "loss": 0.1943,
+      "step": 5470
+    },
+    {
+      "epoch": 1.5,
+      "learning_rate": 0.0002873965311553638,
+      "loss": 0.1878,
+      "step": 5480
+    },
+    {
+      "epoch": 1.5,
+      "learning_rate": 0.0002873690006423786,
+      "loss": 0.1777,
+      "step": 5490
+    },
+    {
+      "epoch": 1.51,
+      "learning_rate": 0.0002873414701293934,
+      "loss": 0.1779,
+      "step": 5500
+    },
+    {
+      "epoch": 1.51,
+      "eval_loss": 0.11272893100976944,
+      "eval_runtime": 1051.3836,
+      "eval_samples_per_second": 11.112,
+      "eval_steps_per_second": 1.39,
+      "eval_wer": 0.3426157554807605,
+      "step": 5500
+    },
+    {
+      "epoch": 1.51,
+      "learning_rate": 0.00028731393961640815,
+      "loss": 0.1695,
+      "step": 5510
+    },
+    {
+      "epoch": 1.51,
+      "learning_rate": 0.00028728640910342295,
+      "loss": 0.1919,
+      "step": 5520
+    },
+    {
+      "epoch": 1.52,
+      "learning_rate": 0.0002872588785904377,
+      "loss": 0.1786,
+      "step": 5530
+    },
+    {
+      "epoch": 1.52,
+      "learning_rate": 0.0002872313480774525,
+      "loss": 0.186,
+      "step": 5540
+    },
+    {
+      "epoch": 1.52,
+      "learning_rate": 0.0002872038175644673,
+      "loss": 0.1565,
+      "step": 5550
+    },
+    {
+      "epoch": 1.52,
+      "learning_rate": 0.000287176287051482,
+      "loss": 0.2021,
+      "step": 5560
+    },
+    {
+      "epoch": 1.53,
+      "learning_rate": 0.0002871487565384968,
+      "loss": 0.2016,
+      "step": 5570
+    },
+    {
+      "epoch": 1.53,
+      "learning_rate": 0.00028712122602551156,
+      "loss": 0.1918,
+      "step": 5580
+    },
+    {
+      "epoch": 1.53,
+      "learning_rate": 0.00028709369551252635,
+      "loss": 0.2019,
+      "step": 5590
+    },
+    {
+      "epoch": 1.53,
+      "learning_rate": 0.00028706616499954115,
+      "loss": 0.176,
+      "step": 5600
+    },
+    {
+      "epoch": 1.53,
+      "eval_loss": 0.11052325367927551,
+      "eval_runtime": 1047.3933,
+      "eval_samples_per_second": 11.154,
+      "eval_steps_per_second": 1.395,
+      "eval_wer": 0.34463977076321295,
+      "step": 5600
+    },
+    {
+      "epoch": 1.54,
+      "learning_rate": 0.00028703863448655594,
+      "loss": 0.1895,
+      "step": 5610
+    },
+    {
+      "epoch": 1.54,
+      "learning_rate": 0.0002870111039735707,
+      "loss": 0.1773,
+      "step": 5620
+    },
+    {
+      "epoch": 1.54,
+      "learning_rate": 0.00028698357346058543,
+      "loss": 0.1707,
+      "step": 5630
+    },
+    {
+      "epoch": 1.55,
+      "learning_rate": 0.0002869560429476002,
+      "loss": 0.1953,
+      "step": 5640
+    },
+    {
+      "epoch": 1.55,
+      "learning_rate": 0.000286928512434615,
+      "loss": 0.1857,
+      "step": 5650
+    },
+    {
+      "epoch": 1.55,
+      "learning_rate": 0.0002869009819216298,
+      "loss": 0.1813,
+      "step": 5660
+    },
+    {
+      "epoch": 1.55,
+      "learning_rate": 0.00028687345140864456,
+      "loss": 0.3278,
+      "step": 5670
+    },
+    {
+      "epoch": 1.56,
+      "learning_rate": 0.0002868459208956593,
+      "loss": 0.1934,
+      "step": 5680
+    },
+    {
+      "epoch": 1.56,
+      "learning_rate": 0.0002868183903826741,
+      "loss": 0.1889,
+      "step": 5690
+    },
+    {
+      "epoch": 1.56,
+      "learning_rate": 0.0002867908598696889,
+      "loss": 0.1387,
+      "step": 5700
+    },
+    {
+      "epoch": 1.56,
+      "eval_loss": 0.10185012966394424,
+      "eval_runtime": 1037.8682,
+      "eval_samples_per_second": 11.257,
+      "eval_steps_per_second": 1.408,
+      "eval_wer": 0.33014190848721914,
+      "step": 5700
+    },
+    {
+      "epoch": 1.56,
+      "learning_rate": 0.0002867633293567037,
+      "loss": 0.1744,
+      "step": 5710
+    },
+    {
+      "epoch": 1.57,
+      "learning_rate": 0.00028673579884371843,
+      "loss": 0.1746,
+      "step": 5720
+    },
+    {
+      "epoch": 1.57,
+      "learning_rate": 0.00028670826833073317,
+      "loss": 0.1832,
+      "step": 5730
+    },
+    {
+      "epoch": 1.57,
+      "learning_rate": 0.00028668073781774796,
+      "loss": 0.1929,
+      "step": 5740
+    },
+    {
+      "epoch": 1.58,
+      "learning_rate": 0.00028665320730476276,
+      "loss": 0.1672,
+      "step": 5750
+    },
+    {
+      "epoch": 1.58,
+      "learning_rate": 0.00028662567679177756,
+      "loss": 0.1748,
+      "step": 5760
+    },
+    {
+      "epoch": 1.58,
+      "learning_rate": 0.0002865981462787923,
+      "loss": 0.1814,
+      "step": 5770
+    },
+    {
+      "epoch": 1.58,
+      "learning_rate": 0.00028657061576580704,
+      "loss": 0.1771,
+      "step": 5780
+    },
+    {
+      "epoch": 1.59,
+      "learning_rate": 0.00028654308525282183,
+      "loss": 0.1724,
+      "step": 5790
+    },
+    {
+      "epoch": 1.59,
+      "learning_rate": 0.00028651555473983663,
+      "loss": 0.1621,
+      "step": 5800
+    },
+    {
+      "epoch": 1.59,
+      "eval_loss": 0.10375826060771942,
+      "eval_runtime": 1035.0749,
+      "eval_samples_per_second": 11.287,
+      "eval_steps_per_second": 1.411,
+      "eval_wer": 0.33085827344673885,
+      "step": 5800
+    },
+    {
+      "epoch": 1.59,
+      "learning_rate": 0.0002864880242268514,
+      "loss": 0.1839,
+      "step": 5810
+    },
+    {
+      "epoch": 1.59,
+      "learning_rate": 0.00028646049371386617,
+      "loss": 0.1961,
+      "step": 5820
+    },
+    {
+      "epoch": 1.6,
+      "learning_rate": 0.00028643296320088096,
+      "loss": 0.1632,
+      "step": 5830
+    },
+    {
+      "epoch": 1.6,
+      "learning_rate": 0.0002864054326878957,
+      "loss": 0.1617,
+      "step": 5840
+    },
+    {
+      "epoch": 1.6,
+      "learning_rate": 0.0002863779021749105,
+      "loss": 0.1951,
+      "step": 5850
+    },
+    {
+      "epoch": 1.61,
+      "learning_rate": 0.0002863503716619253,
+      "loss": 0.166,
+      "step": 5860
+    },
+    {
+      "epoch": 1.61,
+      "learning_rate": 0.00028632284114894004,
+      "loss": 0.1689,
+      "step": 5870
+    },
+    {
+      "epoch": 1.61,
+      "learning_rate": 0.00028629531063595483,
+      "loss": 0.1943,
+      "step": 5880
+    },
+    {
+      "epoch": 1.61,
+      "learning_rate": 0.0002862677801229696,
+      "loss": 0.1775,
+      "step": 5890
+    },
+    {
+      "epoch": 1.62,
+      "learning_rate": 0.00028624024960998437,
+      "loss": 0.1527,
+      "step": 5900
+    },
+    {
+      "epoch": 1.62,
+      "eval_loss": 0.11036413908004761,
+      "eval_runtime": 1034.9926,
+      "eval_samples_per_second": 11.288,
+      "eval_steps_per_second": 1.412,
+      "eval_wer": 0.33977303738742837,
+      "step": 5900
+    },
+    {
+      "epoch": 1.62,
+      "learning_rate": 0.00028621271909699917,
+      "loss": 0.2039,
+      "step": 5910
+    },
+    {
+      "epoch": 1.62,
+      "learning_rate": 0.0002861851885840139,
+      "loss": 0.1963,
+      "step": 5920
+    },
+    {
+      "epoch": 1.62,
+      "learning_rate": 0.0002861576580710287,
+      "loss": 0.1781,
+      "step": 5930
+    },
+    {
+      "epoch": 1.63,
+      "learning_rate": 0.00028613012755804345,
+      "loss": 0.1761,
+      "step": 5940
+    },
+    {
+      "epoch": 1.63,
+      "learning_rate": 0.00028610259704505824,
+      "loss": 0.1508,
+      "step": 5950
+    },
+    {
+      "epoch": 1.63,
+      "learning_rate": 0.00028607506653207304,
+      "loss": 0.2437,
+      "step": 5960
+    },
+    {
+      "epoch": 1.64,
+      "learning_rate": 0.00028604753601908783,
+      "loss": 0.1787,
+      "step": 5970
+    },
+    {
+      "epoch": 1.64,
+      "learning_rate": 0.0002860200055061026,
+      "loss": 0.1921,
+      "step": 5980
+    },
+    {
+      "epoch": 1.64,
+      "learning_rate": 0.0002859924749931173,
+      "loss": 0.1984,
+      "step": 5990
+    },
+    {
+      "epoch": 1.64,
+      "learning_rate": 0.0002859649444801321,
+      "loss": 0.1748,
+      "step": 6000
+    },
+    {
+      "epoch": 1.64,
+      "eval_loss": 0.10706117004156113,
+      "eval_runtime": 1037.1354,
+      "eval_samples_per_second": 11.265,
+      "eval_steps_per_second": 1.409,
+      "eval_wer": 0.3348267079050305,
+      "step": 6000
+    },
+    {
+      "epoch": 1.65,
+      "learning_rate": 0.0002859374139671469,
+      "loss": 0.2553,
+      "step": 6010
+    },
+    {
+      "epoch": 1.65,
+      "learning_rate": 0.0002859098834541617,
+      "loss": 0.1911,
+      "step": 6020
+    },
+    {
+      "epoch": 1.65,
+      "learning_rate": 0.00028588235294117645,
+      "loss": 0.1553,
+      "step": 6030
+    },
+    {
+      "epoch": 1.66,
+      "learning_rate": 0.0002858548224281912,
+      "loss": 0.1948,
+      "step": 6040
+    },
+    {
+      "epoch": 1.66,
+      "learning_rate": 0.000285827291915206,
+      "loss": 0.2145,
+      "step": 6050
+    },
+    {
+      "epoch": 1.66,
+      "learning_rate": 0.0002857997614022208,
+      "loss": 0.1577,
+      "step": 6060
+    },
+    {
+      "epoch": 1.66,
+      "learning_rate": 0.0002857722308892356,
+      "loss": 0.1695,
+      "step": 6070
+    },
+    {
+      "epoch": 1.67,
+      "learning_rate": 0.0002857447003762503,
+      "loss": 0.2078,
+      "step": 6080
+    },
+    {
+      "epoch": 1.67,
+      "learning_rate": 0.00028571716986326506,
+      "loss": 0.2078,
+      "step": 6090
+    },
+    {
+      "epoch": 1.67,
+      "learning_rate": 0.00028568963935027985,
+      "loss": 0.165,
+      "step": 6100
+    },
+    {
+      "epoch": 1.67,
+      "eval_loss": 0.10916991531848907,
+      "eval_runtime": 1042.7945,
+      "eval_samples_per_second": 11.204,
+      "eval_steps_per_second": 1.401,
+      "eval_wer": 0.33797643955244244,
+      "step": 6100
+    },
+    {
+      "epoch": 1.67,
+      "learning_rate": 0.00028566210883729465,
+      "loss": 0.1749,
+      "step": 6110
+    },
+    {
+      "epoch": 1.68,
+      "learning_rate": 0.00028563457832430944,
+      "loss": 0.165,
+      "step": 6120
+    },
+    {
+      "epoch": 1.68,
+      "learning_rate": 0.0002856070478113242,
+      "loss": 0.1728,
+      "step": 6130
+    },
+    {
+      "epoch": 1.68,
+      "learning_rate": 0.000285579517298339,
+      "loss": 0.1732,
+      "step": 6140
+    },
+    {
+      "epoch": 1.69,
+      "learning_rate": 0.0002855519867853537,
+      "loss": 0.1662,
+      "step": 6150
+    },
+    {
+      "epoch": 1.69,
+      "learning_rate": 0.0002855244562723685,
+      "loss": 0.1689,
+      "step": 6160
+    },
+    {
+      "epoch": 1.69,
+      "learning_rate": 0.0002854969257593833,
+      "loss": 0.1774,
+      "step": 6170
+    },
+    {
+      "epoch": 1.69,
+      "learning_rate": 0.00028546939524639806,
+      "loss": 0.1691,
+      "step": 6180
+    },
+    {
+      "epoch": 1.7,
+      "learning_rate": 0.00028544186473341285,
+      "loss": 0.176,
+      "step": 6190
+    },
+    {
+      "epoch": 1.7,
+      "learning_rate": 0.0002854143342204276,
+      "loss": 0.1384,
+      "step": 6200
+    },
+    {
+      "epoch": 1.7,
+      "eval_loss": 0.10628872364759445,
+      "eval_runtime": 1039.2196,
+      "eval_samples_per_second": 11.242,
+      "eval_steps_per_second": 1.406,
+      "eval_wer": 0.3349404166287638,
+      "step": 6200
+    },
+    {
+      "epoch": 1.7,
+      "learning_rate": 0.0002853868037074424,
+      "loss": 0.2081,
+      "step": 6210
+    },
+    {
+      "epoch": 1.7,
+      "learning_rate": 0.0002853592731944572,
+      "loss": 0.1934,
+      "step": 6220
+    },
+    {
+      "epoch": 1.71,
+      "learning_rate": 0.00028533174268147193,
+      "loss": 0.1613,
+      "step": 6230
+    },
+    {
+      "epoch": 1.71,
+      "learning_rate": 0.0002853042121684867,
+      "loss": 0.1717,
+      "step": 6240
+    },
+    {
+      "epoch": 1.71,
+      "learning_rate": 0.00028527668165550146,
+      "loss": 0.1603,
+      "step": 6250
+    },
+    {
+      "epoch": 1.72,
+      "learning_rate": 0.00028524915114251626,
+      "loss": 0.1749,
+      "step": 6260
+    },
+    {
+      "epoch": 1.72,
+      "learning_rate": 0.00028522162062953106,
+      "loss": 0.1565,
+      "step": 6270
+    },
+    {
+      "epoch": 1.72,
+      "learning_rate": 0.00028519409011654585,
+      "loss": 0.1862,
+      "step": 6280
+    },
+    {
+      "epoch": 1.72,
+      "learning_rate": 0.0002851665596035606,
+      "loss": 0.164,
+      "step": 6290
+    },
+    {
+      "epoch": 1.73,
+      "learning_rate": 0.00028513902909057534,
+      "loss": 0.1859,
+      "step": 6300
+    },
+    {
+      "epoch": 1.73,
+      "eval_loss": 0.10580063611268997,
+      "eval_runtime": 1053.9327,
+      "eval_samples_per_second": 11.085,
+      "eval_steps_per_second": 1.386,
+      "eval_wer": 0.32495679068498134,
+      "step": 6300
+    },
+    {
+      "epoch": 1.73,
+      "learning_rate": 0.00028511149857759013,
+      "loss": 0.1775,
+      "step": 6310
+    },
+    {
+      "epoch": 1.73,
+      "learning_rate": 0.0002850839680646049,
+      "loss": 0.1876,
+      "step": 6320
+    },
+    {
+      "epoch": 1.73,
+      "learning_rate": 0.0002850564375516197,
+      "loss": 0.1628,
+      "step": 6330
+    },
+    {
+      "epoch": 1.74,
+      "learning_rate": 0.00028502890703863446,
+      "loss": 0.1733,
+      "step": 6340
+    },
+    {
+      "epoch": 1.74,
+      "learning_rate": 0.0002850013765256492,
+      "loss": 0.1906,
+      "step": 6350
+    },
+    {
+      "epoch": 1.74,
+      "learning_rate": 0.000284973846012664,
+      "loss": 0.1974,
+      "step": 6360
+    },
+    {
+      "epoch": 1.75,
+      "learning_rate": 0.0002849463154996788,
+      "loss": 0.1655,
+      "step": 6370
+    },
+    {
+      "epoch": 1.75,
+      "learning_rate": 0.0002849187849866936,
+      "loss": 0.1646,
+      "step": 6380
+    },
+    {
+      "epoch": 1.75,
+      "learning_rate": 0.00028489125447370833,
+      "loss": 0.1941,
+      "step": 6390
+    },
+    {
+      "epoch": 1.75,
+      "learning_rate": 0.0002848637239607231,
+      "loss": 0.1919,
+      "step": 6400
+    },
+    {
+      "epoch": 1.75,
+      "eval_loss": 0.1080482006072998,
+      "eval_runtime": 1063.6328,
+      "eval_samples_per_second": 10.984,
+      "eval_steps_per_second": 1.374,
+      "eval_wer": 0.34133084690257437,
+      "step": 6400
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 0.00028483619344773787,
+      "loss": 0.1714,
+      "step": 6410
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 0.00028480866293475267,
+      "loss": 0.1509,
+      "step": 6420
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 0.00028478113242176746,
+      "loss": 0.1878,
+      "step": 6430
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 0.0002847536019087822,
+      "loss": 0.1855,
+      "step": 6440
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 0.00028472607139579695,
+      "loss": 0.1743,
+      "step": 6450
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 0.00028469854088281174,
+      "loss": 0.1612,
+      "step": 6460
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 0.00028467101036982654,
+      "loss": 0.1659,
+      "step": 6470
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 0.00028464347985684133,
+      "loss": 0.1824,
+      "step": 6480
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 0.0002846159493438561,
+      "loss": 0.1656,
+      "step": 6490
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 0.00028458841883087087,
+      "loss": 0.1425,
+      "step": 6500
+    },
+    {
+      "epoch": 1.78,
+      "eval_loss": 0.10503587871789932,
+      "eval_runtime": 1052.5362,
+      "eval_samples_per_second": 11.1,
+      "eval_steps_per_second": 1.388,
+      "eval_wer": 0.3284135358864732,
+      "step": 6500
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 0.0002845608883178856,
+      "loss": 0.1625,
+      "step": 6510
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 0.0002845333578049004,
+      "loss": 0.1626,
+      "step": 6520
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 0.0002845058272919152,
+      "loss": 0.1669,
+      "step": 6530
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 0.00028447829677892995,
+      "loss": 0.1951,
+      "step": 6540
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 0.00028445076626594474,
+      "loss": 0.1674,
+      "step": 6550
+    },
+    {
+      "epoch": 1.8,
+      "learning_rate": 0.0002844232357529595,
+      "loss": 0.1615,
+      "step": 6560
+    },
+    {
+      "epoch": 1.8,
+      "learning_rate": 0.0002843957052399743,
+      "loss": 0.1559,
+      "step": 6570
+    },
+    {
+      "epoch": 1.8,
+      "learning_rate": 0.0002843681747269891,
+      "loss": 0.1697,
+      "step": 6580
+    },
+    {
+      "epoch": 1.81,
+      "learning_rate": 0.00028434064421400387,
+      "loss": 0.186,
+      "step": 6590
+    },
+    {
+      "epoch": 1.81,
+      "learning_rate": 0.0002843131137010186,
+      "loss": 0.1571,
+      "step": 6600
+    },
+    {
+      "epoch": 1.81,
+      "eval_loss": 0.10015376657247543,
+      "eval_runtime": 1050.3185,
+      "eval_samples_per_second": 11.123,
+      "eval_steps_per_second": 1.391,
+      "eval_wer": 0.32436550532156827,
+      "step": 6600
+    },
+    {
+      "epoch": 1.81,
+      "learning_rate": 0.00028428558318803335,
+      "loss": 0.1643,
+      "step": 6610
+    },
+    {
+      "epoch": 1.81,
+      "learning_rate": 0.00028425805267504815,
+      "loss": 0.1636,
+      "step": 6620
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 0.00028423052216206294,
+      "loss": 0.1646,
+      "step": 6630
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 0.00028420299164907774,
+      "loss": 0.1628,
+      "step": 6640
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 0.0002841754611360925,
+      "loss": 0.1288,
+      "step": 6650
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 0.0002841479306231072,
+      "loss": 0.1523,
+      "step": 6660
+    },
+    {
+      "epoch": 1.83,
+      "learning_rate": 0.000284120400110122,
+      "loss": 0.1646,
+      "step": 6670
+    },
+    {
+      "epoch": 1.83,
+      "learning_rate": 0.0002840928695971368,
+      "loss": 0.1748,
+      "step": 6680
+    },
+    {
+      "epoch": 1.83,
+      "learning_rate": 0.0002840653390841516,
+      "loss": 0.164,
+      "step": 6690
+    },
+    {
+      "epoch": 1.84,
+      "learning_rate": 0.00028403780857116635,
+      "loss": 0.1591,
+      "step": 6700
+    },
+    {
+      "epoch": 1.84,
+      "eval_loss": 0.09989971667528152,
+      "eval_runtime": 1047.7519,
+      "eval_samples_per_second": 11.151,
+      "eval_steps_per_second": 1.394,
+      "eval_wer": 0.3237742199581552,
+      "step": 6700
+    },
+    {
+      "epoch": 1.84,
+      "learning_rate": 0.0002840102780581811,
+      "loss": 0.1732,
+      "step": 6710
+    },
+    {
+      "epoch": 1.84,
+      "learning_rate": 0.0002839827475451959,
+      "loss": 0.1573,
+      "step": 6720
+    },
+    {
+      "epoch": 1.84,
+      "learning_rate": 0.0002839552170322107,
+      "loss": 0.1639,
+      "step": 6730
+    },
+    {
+      "epoch": 1.85,
+      "learning_rate": 0.0002839276865192255,
+      "loss": 0.1808,
+      "step": 6740
+    },
+    {
+      "epoch": 1.85,
+      "learning_rate": 0.0002839001560062402,
+      "loss": 0.1593,
+      "step": 6750
+    },
+    {
+      "epoch": 1.85,
+      "learning_rate": 0.00028387262549325496,
+      "loss": 0.168,
+      "step": 6760
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 0.00028384509498026976,
+      "loss": 0.1756,
+      "step": 6770
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 0.00028381756446728456,
+      "loss": 0.163,
+      "step": 6780
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 0.00028379003395429935,
+      "loss": 0.1771,
+      "step": 6790
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 0.00028376525649261264,
+      "loss": 0.1615,
+      "step": 6800
+    },
+    {
+      "epoch": 1.86,
+      "eval_loss": 0.09614775329828262,
+      "eval_runtime": 1045.8287,
+      "eval_samples_per_second": 11.171,
+      "eval_steps_per_second": 1.397,
+      "eval_wer": 0.3156781588283453,
+      "step": 6800
+    },
+    {
+      "epoch": 1.87,
+      "learning_rate": 0.0002837377259796274,
+      "loss": 0.1792,
+      "step": 6810
+    },
+    {
+      "epoch": 1.87,
+      "learning_rate": 0.0002837101954666422,
+      "loss": 0.1612,
+      "step": 6820
+    },
+    {
+      "epoch": 1.87,
+      "learning_rate": 0.00028368266495365697,
+      "loss": 0.1664,
+      "step": 6830
+    },
+    {
+      "epoch": 1.87,
+      "learning_rate": 0.0002836551344406717,
+      "loss": 0.1894,
+      "step": 6840
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 0.0002836276039276865,
+      "loss": 0.1345,
+      "step": 6850
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 0.00028360007341470125,
+      "loss": 0.1623,
+      "step": 6860
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 0.00028357254290171604,
+      "loss": 0.1799,
+      "step": 6870
+    },
+    {
+      "epoch": 1.89,
+      "learning_rate": 0.00028354501238873084,
+      "loss": 0.1759,
+      "step": 6880
+    },
+    {
+      "epoch": 1.89,
+      "learning_rate": 0.0002835174818757456,
+      "loss": 0.1627,
+      "step": 6890
+    },
+    {
+      "epoch": 1.89,
+      "learning_rate": 0.0002834899513627604,
+      "loss": 0.148,
+      "step": 6900
+    },
+    {
+      "epoch": 1.89,
+      "eval_loss": 0.09930101782083511,
+      "eval_runtime": 1069.3227,
+      "eval_samples_per_second": 10.926,
+      "eval_steps_per_second": 1.366,
+      "eval_wer": 0.3197489311379969,
+      "step": 6900
+    },
+    {
+      "epoch": 1.89,
+      "learning_rate": 0.0002834624208497751,
+      "loss": 0.209,
+      "step": 6910
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 0.0002834348903367899,
+      "loss": 0.1594,
+      "step": 6920
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 0.0002834073598238047,
+      "loss": 0.1598,
+      "step": 6930
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 0.00028337982931081945,
+      "loss": 0.1674,
+      "step": 6940
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 0.00028335229879783425,
+      "loss": 0.1648,
+      "step": 6950
+    },
+    {
+      "epoch": 1.91,
+      "learning_rate": 0.000283324768284849,
+      "loss": 0.1673,
+      "step": 6960
+    },
+    {
+      "epoch": 1.91,
+      "learning_rate": 0.0002832972377718638,
+      "loss": 0.1777,
+      "step": 6970
+    },
+    {
+      "epoch": 1.91,
+      "learning_rate": 0.0002832697072588786,
+      "loss": 0.1568,
+      "step": 6980
+    },
+    {
+      "epoch": 1.92,
+      "learning_rate": 0.0002832421767458934,
+      "loss": 0.1883,
+      "step": 6990
+    },
+    {
+      "epoch": 1.92,
+      "learning_rate": 0.0002832146462329081,
+      "loss": 0.1893,
+      "step": 7000
+    },
+    {
+      "epoch": 1.92,
+      "eval_loss": 0.09632071852684021,
+      "eval_runtime": 1046.0027,
+      "eval_samples_per_second": 11.169,
+      "eval_steps_per_second": 1.397,
+      "eval_wer": 0.32194350950604933,
+      "step": 7000
+    },
+    {
+      "epoch": 1.92,
+      "learning_rate": 0.00028318711571992286,
+      "loss": 0.1766,
+      "step": 7010
+    },
+    {
+      "epoch": 1.92,
+      "learning_rate": 0.00028315958520693765,
+      "loss": 0.1697,
+      "step": 7020
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 0.00028313205469395245,
+      "loss": 0.1723,
+      "step": 7030
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 0.00028310452418096725,
+      "loss": 0.1574,
+      "step": 7040
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 0.000283076993667982,
+      "loss": 0.1456,
+      "step": 7050
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 0.0002830494631549968,
+      "loss": 0.1686,
+      "step": 7060
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 0.0002830219326420115,
+      "loss": 0.1706,
+      "step": 7070
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 0.0002829944021290263,
+      "loss": 0.1645,
+      "step": 7080
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 0.0002829668716160411,
+      "loss": 0.1866,
+      "step": 7090
+    },
+    {
+      "epoch": 1.95,
+      "learning_rate": 0.00028293934110305586,
+      "loss": 0.1586,
+      "step": 7100
+    },
+    {
+      "epoch": 1.95,
+      "eval_loss": 0.0969810038805008,
+      "eval_runtime": 1048.7511,
+      "eval_samples_per_second": 11.14,
+      "eval_steps_per_second": 1.393,
+      "eval_wer": 0.3227394705721823,
+      "step": 7100
+    },
+    {
+      "epoch": 1.95,
+      "learning_rate": 0.00028291181059007065,
+      "loss": 0.1577,
+      "step": 7110
+    },
+    {
+      "epoch": 1.95,
+      "learning_rate": 0.0002828842800770854,
+      "loss": 0.1716,
+      "step": 7120
+    },
+    {
+      "epoch": 1.95,
+      "learning_rate": 0.0002828567495641002,
+      "loss": 0.1957,
+      "step": 7130
+    },
+    {
+      "epoch": 1.96,
+      "learning_rate": 0.000282829219051115,
+      "loss": 0.1704,
+      "step": 7140
+    },
+    {
+      "epoch": 1.96,
+      "learning_rate": 0.00028280168853812973,
+      "loss": 0.1522,
+      "step": 7150
+    },
+    {
+      "epoch": 1.96,
+      "learning_rate": 0.0002827741580251445,
+      "loss": 0.1692,
+      "step": 7160
+    },
+    {
+      "epoch": 1.96,
+      "learning_rate": 0.00028274662751215927,
+      "loss": 0.2043,
+      "step": 7170
+    },
+    {
+      "epoch": 1.97,
+      "learning_rate": 0.00028271909699917406,
+      "loss": 0.1745,
+      "step": 7180
+    },
+    {
+      "epoch": 1.97,
+      "learning_rate": 0.00028269156648618886,
+      "loss": 0.1994,
+      "step": 7190
+    },
+    {
+      "epoch": 1.97,
+      "learning_rate": 0.0002826640359732036,
+      "loss": 0.1932,
+      "step": 7200
+    },
+    {
+      "epoch": 1.97,
+      "eval_loss": 0.09549184143543243,
+      "eval_runtime": 1063.1244,
+      "eval_samples_per_second": 10.989,
+      "eval_steps_per_second": 1.374,
+      "eval_wer": 0.3200104612025835,
+      "step": 7200
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 0.0002826365054602184,
+      "loss": 0.1741,
+      "step": 7210
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 0.00028260897494723314,
+      "loss": 0.1612,
+      "step": 7220
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 0.00028258144443424793,
+      "loss": 0.1632,
+      "step": 7230
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 0.00028255391392126273,
+      "loss": 0.1843,
+      "step": 7240
+    },
+    {
+      "epoch": 1.99,
+      "learning_rate": 0.00028252638340827747,
+      "loss": 0.1714,
+      "step": 7250
+    },
+    {
+      "epoch": 1.99,
+      "learning_rate": 0.00028249885289529227,
+      "loss": 0.1662,
+      "step": 7260
+    },
+    {
+      "epoch": 1.99,
+      "learning_rate": 0.000282471322382307,
+      "loss": 0.1622,
+      "step": 7270
+    },
+    {
+      "epoch": 1.99,
+      "learning_rate": 0.0002824437918693218,
+      "loss": 0.1642,
+      "step": 7280
+    },
+    {
+      "epoch": 2.0,
+      "learning_rate": 0.0002824162613563366,
+      "loss": 0.2103,
+      "step": 7290
+    },
+    {
+      "epoch": 2.0,
+      "learning_rate": 0.00028238873084335134,
+      "loss": 0.172,
+      "step": 7300
+    },
+    {
+      "epoch": 2.0,
+      "eval_loss": 0.09298280626535416,
+      "eval_runtime": 1067.22,
+      "eval_samples_per_second": 10.947,
+      "eval_steps_per_second": 1.369,
+      "eval_wer": 0.3159169471481852,
+      "step": 7300
+    },
+    {
+      "epoch": 2.0,
+      "learning_rate": 0.00028236120033036614,
+      "loss": 0.1545,
+      "step": 7310
+    },
+    {
+      "epoch": 2.01,
+      "learning_rate": 0.0002823336698173809,
+      "loss": 0.1518,
+      "step": 7320
+    },
+    {
+      "epoch": 2.01,
+      "learning_rate": 0.0002823061393043957,
+      "loss": 0.1638,
+      "step": 7330
+    },
+    {
+      "epoch": 2.01,
+      "learning_rate": 0.00028227860879141047,
+      "loss": 0.3504,
+      "step": 7340
+    },
+    {
+      "epoch": 2.01,
+      "learning_rate": 0.0002822510782784252,
+      "loss": 0.1497,
+      "step": 7350
+    },
+    {
+      "epoch": 2.02,
+      "learning_rate": 0.00028222354776544,
+      "loss": 0.204,
+      "step": 7360
+    },
+    {
+      "epoch": 2.02,
+      "learning_rate": 0.0002821960172524548,
+      "loss": 0.1577,
+      "step": 7370
+    },
+    {
+      "epoch": 2.02,
+      "learning_rate": 0.00028216848673946954,
+      "loss": 0.1677,
+      "step": 7380
+    },
+    {
+      "epoch": 2.02,
+      "learning_rate": 0.00028214095622648434,
+      "loss": 0.1653,
+      "step": 7390
+    },
+    {
+      "epoch": 2.03,
+      "learning_rate": 0.00028211342571349913,
+      "loss": 0.1617,
+      "step": 7400
+    },
+    {
+      "epoch": 2.03,
+      "eval_loss": 0.09724044799804688,
+      "eval_runtime": 1067.6814,
+      "eval_samples_per_second": 10.942,
+      "eval_steps_per_second": 1.368,
+      "eval_wer": 0.3169630674065314,
+      "step": 7400
+    },
+    {
+      "epoch": 2.03,
+      "learning_rate": 0.0002820858952005139,
+      "loss": 0.1405,
+      "step": 7410
+    },
+    {
+      "epoch": 2.03,
+      "learning_rate": 0.00028205836468752867,
+      "loss": 0.1483,
+      "step": 7420
+    },
+    {
+      "epoch": 2.04,
+      "learning_rate": 0.0002820308341745434,
+      "loss": 0.1764,
+      "step": 7430
+    },
+    {
+      "epoch": 2.04,
+      "learning_rate": 0.0002820033036615582,
+      "loss": 0.1624,
+      "step": 7440
+    },
+    {
+      "epoch": 2.04,
+      "learning_rate": 0.000281975773148573,
+      "loss": 0.1766,
+      "step": 7450
+    },
+    {
+      "epoch": 2.04,
+      "learning_rate": 0.00028194824263558775,
+      "loss": 0.1594,
+      "step": 7460
+    },
+    {
+      "epoch": 2.05,
+      "learning_rate": 0.00028192071212260254,
+      "loss": 0.1473,
+      "step": 7470
+    },
+    {
+      "epoch": 2.05,
+      "learning_rate": 0.0002818931816096173,
+      "loss": 0.1403,
+      "step": 7480
+    },
+    {
+      "epoch": 2.05,
+      "learning_rate": 0.0002818656510966321,
+      "loss": 0.1618,
+      "step": 7490
+    },
+    {
+      "epoch": 2.06,
+      "learning_rate": 0.0002818381205836469,
+      "loss": 0.1644,
+      "step": 7500
+    },
+    {
+      "epoch": 2.06,
+      "eval_loss": 0.0922645702958107,
+      "eval_runtime": 1046.6265,
+      "eval_samples_per_second": 11.163,
+      "eval_steps_per_second": 1.396,
+      "eval_wer": 0.3116983534976803,
+      "step": 7500
+    },
+    {
+      "epoch": 2.06,
+      "learning_rate": 0.0002818105900706616,
+      "loss": 0.1947,
+      "step": 7510
+    },
+    {
+      "epoch": 2.06,
+      "learning_rate": 0.0002817830595576764,
+      "loss": 0.4217,
+      "step": 7520
+    },
+    {
+      "epoch": 2.06,
+      "learning_rate": 0.00028175552904469115,
+      "loss": 0.1694,
+      "step": 7530
+    },
+    {
+      "epoch": 2.07,
+      "learning_rate": 0.00028172799853170595,
+      "loss": 0.1781,
+      "step": 7540
+    },
+    {
+      "epoch": 2.07,
+      "learning_rate": 0.00028170046801872075,
+      "loss": 0.1502,
+      "step": 7550
+    },
+    {
+      "epoch": 2.07,
+      "learning_rate": 0.0002816729375057355,
+      "loss": 0.1733,
+      "step": 7560
+    },
+    {
+      "epoch": 2.07,
+      "learning_rate": 0.0002816454069927503,
+      "loss": 0.1506,
+      "step": 7570
+    },
+    {
+      "epoch": 2.08,
+      "learning_rate": 0.000281617876479765,
+      "loss": 0.1675,
+      "step": 7580
+    },
+    {
+      "epoch": 2.08,
+      "learning_rate": 0.0002815903459667798,
+      "loss": 0.1532,
+      "step": 7590
+    },
+    {
+      "epoch": 2.08,
+      "learning_rate": 0.0002815628154537946,
+      "loss": 0.1098,
+      "step": 7600
+    },
+    {
+      "epoch": 2.08,
+      "eval_loss": 0.10008783638477325,
+      "eval_runtime": 1043.3579,
+      "eval_samples_per_second": 11.197,
+      "eval_steps_per_second": 1.4,
+      "eval_wer": 0.31861184390066405,
+      "step": 7600
+    },
+    {
+      "epoch": 2.09,
+      "learning_rate": 0.00028153528494080936,
+      "loss": 0.1547,
+      "step": 7610
+    },
+    {
+      "epoch": 2.09,
+      "learning_rate": 0.00028150775442782415,
+      "loss": 0.1447,
+      "step": 7620
+    },
+    {
+      "epoch": 2.09,
+      "learning_rate": 0.0002814802239148389,
+      "loss": 0.1553,
+      "step": 7630
+    },
+    {
+      "epoch": 2.09,
+      "learning_rate": 0.0002814526934018537,
+      "loss": 0.156,
+      "step": 7640
+    },
+    {
+      "epoch": 2.1,
+      "learning_rate": 0.0002814251628888685,
+      "loss": 0.1499,
+      "step": 7650
+    },
+    {
+      "epoch": 2.1,
+      "learning_rate": 0.00028139763237588323,
+      "loss": 0.2971,
+      "step": 7660
+    },
+    {
+      "epoch": 2.1,
+      "learning_rate": 0.000281370101862898,
+      "loss": 0.1428,
+      "step": 7670
+    },
+    {
+      "epoch": 2.1,
+      "learning_rate": 0.0002813425713499128,
+      "loss": 0.1695,
+      "step": 7680
+    },
+    {
+      "epoch": 2.11,
+      "learning_rate": 0.00028131504083692756,
+      "loss": 0.1704,
+      "step": 7690
+    },
+    {
+      "epoch": 2.11,
+      "learning_rate": 0.00028128751032394236,
+      "loss": 0.1456,
+      "step": 7700
+    },
+    {
+      "epoch": 2.11,
+      "eval_loss": 0.10283832997083664,
+      "eval_runtime": 1053.1465,
+      "eval_samples_per_second": 11.093,
+      "eval_steps_per_second": 1.387,
+      "eval_wer": 0.32621895751842084,
+      "step": 7700
+    },
+    {
+      "epoch": 2.11,
+      "learning_rate": 0.0002812599798109571,
+      "loss": 0.1464,
+      "step": 7710
+    },
+    {
+      "epoch": 2.12,
+      "learning_rate": 0.0002812324492979719,
+      "loss": 0.1493,
+      "step": 7720
+    },
+    {
+      "epoch": 2.12,
+      "learning_rate": 0.0002812049187849867,
+      "loss": 0.161,
+      "step": 7730
+    },
+    {
+      "epoch": 2.12,
+      "learning_rate": 0.00028117738827200143,
+      "loss": 0.1637,
+      "step": 7740
+    },
+    {
+      "epoch": 2.12,
+      "learning_rate": 0.00028114985775901623,
+      "loss": 0.146,
+      "step": 7750
+    },
+    {
+      "epoch": 2.13,
+      "learning_rate": 0.000281122327246031,
+      "loss": 0.1448,
+      "step": 7760
+    },
+    {
+      "epoch": 2.13,
+      "learning_rate": 0.00028109479673304577,
+      "loss": 0.1378,
+      "step": 7770
+    },
+    {
+      "epoch": 2.13,
+      "learning_rate": 0.00028106726622006056,
+      "loss": 0.164,
+      "step": 7780
+    },
+    {
+      "epoch": 2.13,
+      "learning_rate": 0.0002810397357070753,
+      "loss": 0.1681,
+      "step": 7790
+    },
+    {
+      "epoch": 2.14,
+      "learning_rate": 0.0002810122051940901,
+      "loss": 0.1518,
+      "step": 7800
+    },
+    {
+      "epoch": 2.14,
+      "eval_loss": 0.10179836302995682,
+      "eval_runtime": 1062.7252,
+      "eval_samples_per_second": 10.993,
+      "eval_steps_per_second": 1.375,
+      "eval_wer": 0.3176680614936778,
+      "step": 7800
+    },
+    {
+      "epoch": 2.14,
+      "learning_rate": 0.0002809846746811049,
+      "loss": 0.1422,
+      "step": 7810
+    },
+    {
+      "epoch": 2.14,
+      "learning_rate": 0.00028095714416811964,
+      "loss": 0.1459,
+      "step": 7820
+    },
+    {
+      "epoch": 2.15,
+      "learning_rate": 0.00028092961365513443,
+      "loss": 0.1555,
+      "step": 7830
+    },
+    {
+      "epoch": 2.15,
+      "learning_rate": 0.0002809020831421492,
+      "loss": 0.1484,
+      "step": 7840
+    },
+    {
+      "epoch": 2.15,
+      "learning_rate": 0.00028087730568046246,
+      "loss": 0.1217,
+      "step": 7850
+    },
+    {
+      "epoch": 2.15,
+      "learning_rate": 0.00028084977516747725,
+      "loss": 0.1407,
+      "step": 7860
+    },
+    {
+      "epoch": 2.16,
+      "learning_rate": 0.00028082224465449205,
+      "loss": 0.1494,
+      "step": 7870
+    },
+    {
+      "epoch": 2.16,
+      "learning_rate": 0.0002807947141415068,
+      "loss": 0.1643,
+      "step": 7880
+    },
+    {
+      "epoch": 2.16,
+      "learning_rate": 0.0002807671836285216,
+      "loss": 0.1696,
+      "step": 7890
+    },
+    {
+      "epoch": 2.16,
+      "learning_rate": 0.0002807396531155364,
+      "loss": 0.1485,
+      "step": 7900
+    },
+    {
+      "epoch": 2.16,
+      "eval_loss": 0.09863564372062683,
+      "eval_runtime": 1050.6891,
+      "eval_samples_per_second": 11.119,
+      "eval_steps_per_second": 1.391,
+      "eval_wer": 0.3153256617847721,
+      "step": 7900
+    },
+    {
+      "epoch": 2.17,
+      "learning_rate": 0.0002807121226025511,
+      "loss": 0.1462,
+      "step": 7910
+    },
+    {
+      "epoch": 2.17,
+      "learning_rate": 0.0002806845920895659,
+      "loss": 0.1486,
+      "step": 7920
+    },
+    {
+      "epoch": 2.17,
+      "learning_rate": 0.0002806570615765807,
+      "loss": 0.172,
+      "step": 7930
+    },
+    {
+      "epoch": 2.18,
+      "learning_rate": 0.00028062953106359546,
+      "loss": 0.1719,
+      "step": 7940
+    },
+    {
+      "epoch": 2.18,
+      "learning_rate": 0.00028060200055061025,
+      "loss": 0.1477,
+      "step": 7950
+    },
+    {
+      "epoch": 2.18,
+      "learning_rate": 0.000280574470037625,
+      "loss": 0.1478,
+      "step": 7960
+    },
+    {
+      "epoch": 2.18,
+      "learning_rate": 0.0002805469395246398,
+      "loss": 0.1693,
+      "step": 7970
+    },
+    {
+      "epoch": 2.19,
+      "learning_rate": 0.0002805194090116546,
+      "loss": 0.1479,
+      "step": 7980
+    },
+    {
+      "epoch": 2.19,
+      "learning_rate": 0.0002804918784986693,
+      "loss": 0.1835,
+      "step": 7990
+    },
+    {
+      "epoch": 2.19,
+      "learning_rate": 0.0002804643479856841,
+      "loss": 0.1411,
+      "step": 8000
+    },
+    {
+      "epoch": 2.19,
+      "eval_loss": 0.09377647936344147,
+      "eval_runtime": 1071.9906,
+      "eval_samples_per_second": 10.898,
+      "eval_steps_per_second": 1.363,
+      "eval_wer": 0.31165287000818703,
+      "step": 8000
+    },
+    {
+      "epoch": 2.19,
+      "learning_rate": 0.00028043681747269886,
+      "loss": 0.1486,
+      "step": 8010
+    },
+    {
+      "epoch": 2.2,
+      "learning_rate": 0.00028040928695971366,
+      "loss": 0.1452,
+      "step": 8020
+    },
+    {
+      "epoch": 2.2,
+      "learning_rate": 0.00028038175644672846,
+      "loss": 0.1612,
+      "step": 8030
+    },
+    {
+      "epoch": 2.2,
+      "learning_rate": 0.0002803542259337432,
+      "loss": 0.1725,
+      "step": 8040
+    },
+    {
+      "epoch": 2.21,
+      "learning_rate": 0.000280326695420758,
+      "loss": 0.1436,
+      "step": 8050
+    },
+    {
+      "epoch": 2.21,
+      "learning_rate": 0.00028029916490777273,
+      "loss": 0.1561,
+      "step": 8060
+    },
+    {
+      "epoch": 2.21,
+      "learning_rate": 0.00028027163439478753,
+      "loss": 0.1562,
+      "step": 8070
+    },
+    {
+      "epoch": 2.21,
+      "learning_rate": 0.0002802441038818023,
+      "loss": 0.1799,
+      "step": 8080
+    },
+    {
+      "epoch": 2.22,
+      "learning_rate": 0.00028021657336881707,
+      "loss": 0.1425,
+      "step": 8090
+    },
+    {
+      "epoch": 2.22,
+      "learning_rate": 0.00028018904285583186,
+      "loss": 0.1439,
+      "step": 8100
+    },
+    {
+      "epoch": 2.22,
+      "eval_loss": 0.09374161064624786,
+      "eval_runtime": 1057.5136,
+      "eval_samples_per_second": 11.048,
+      "eval_steps_per_second": 1.382,
+      "eval_wer": 0.31238060584008004,
+      "step": 8100
+    },
+    {
+      "epoch": 2.22,
+      "learning_rate": 0.0002801615123428466,
+      "loss": 0.161,
+      "step": 8110
+    },
+    {
+      "epoch": 2.22,
+      "learning_rate": 0.0002801339818298614,
+      "loss": 0.1245,
+      "step": 8120
+    },
+    {
+      "epoch": 2.23,
+      "learning_rate": 0.0002801064513168762,
+      "loss": 0.1461,
+      "step": 8130
+    },
+    {
+      "epoch": 2.23,
+      "learning_rate": 0.00028007892080389094,
+      "loss": 0.1509,
+      "step": 8140
+    },
+    {
+      "epoch": 2.23,
+      "learning_rate": 0.00028005139029090573,
+      "loss": 0.1579,
+      "step": 8150
+    },
+    {
+      "epoch": 2.24,
+      "learning_rate": 0.0002800238597779205,
+      "loss": 0.1496,
+      "step": 8160
+    },
+    {
+      "epoch": 2.24,
+      "learning_rate": 0.00027999632926493527,
+      "loss": 0.1488,
+      "step": 8170
+    },
+    {
+      "epoch": 2.24,
+      "learning_rate": 0.00027996879875195007,
+      "loss": 0.1413,
+      "step": 8180
+    },
+    {
+      "epoch": 2.24,
+      "learning_rate": 0.0002799412682389648,
+      "loss": 0.1775,
+      "step": 8190
+    },
+    {
+      "epoch": 2.25,
+      "learning_rate": 0.0002799137377259796,
+      "loss": 0.1469,
+      "step": 8200
+    },
+    {
+      "epoch": 2.25,
+      "eval_loss": 0.0919284000992775,
+      "eval_runtime": 1040.1265,
+      "eval_samples_per_second": 11.232,
+      "eval_steps_per_second": 1.405,
+      "eval_wer": 0.3096402255981079,
+      "step": 8200
+    },
+    {
+      "epoch": 2.25,
+      "learning_rate": 0.00027988620721299435,
+      "loss": 0.1613,
+      "step": 8210
+    },
+    {
+      "epoch": 2.25,
+      "learning_rate": 0.00027985867670000914,
+      "loss": 0.1405,
+      "step": 8220
+    },
+    {
+      "epoch": 2.26,
+      "learning_rate": 0.00027983114618702394,
+      "loss": 0.1403,
+      "step": 8230
+    },
+    {
+      "epoch": 2.26,
+      "learning_rate": 0.00027980361567403873,
+      "loss": 0.1688,
+      "step": 8240
+    },
+    {
+      "epoch": 2.26,
+      "learning_rate": 0.0002797760851610535,
+      "loss": 0.1495,
+      "step": 8250
+    },
+    {
+      "epoch": 2.26,
+      "learning_rate": 0.0002797485546480682,
+      "loss": 0.146,
+      "step": 8260
+    },
+    {
+      "epoch": 2.27,
+      "learning_rate": 0.000279721024135083,
+      "loss": 0.169,
+      "step": 8270
+    },
+    {
+      "epoch": 2.27,
+      "learning_rate": 0.0002796934936220978,
+      "loss": 0.1583,
+      "step": 8280
+    },
+    {
+      "epoch": 2.27,
+      "learning_rate": 0.0002796659631091126,
+      "loss": 0.1608,
+      "step": 8290
+    },
+    {
+      "epoch": 2.27,
+      "learning_rate": 0.00027963843259612734,
+      "loss": 0.1476,
+      "step": 8300
+    },
+    {
+      "epoch": 2.27,
+      "eval_loss": 0.09394428133964539,
+      "eval_runtime": 1047.8186,
+      "eval_samples_per_second": 11.15,
+      "eval_steps_per_second": 1.394,
+      "eval_wer": 0.30986764304557446,
+      "step": 8300
+    },
+    {
+      "epoch": 2.28,
+      "learning_rate": 0.00027961090208314214,
+      "loss": 0.154,
+      "step": 8310
+    },
+    {
+      "epoch": 2.28,
+      "learning_rate": 0.0002795833715701569,
+      "loss": 0.1406,
+      "step": 8320
+    },
+    {
+      "epoch": 2.28,
+      "learning_rate": 0.0002795558410571717,
+      "loss": 0.1667,
+      "step": 8330
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 0.0002795283105441865,
+      "loss": 0.1622,
+      "step": 8340
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 0.0002795007800312012,
+      "loss": 0.161,
+      "step": 8350
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 0.000279473249518216,
+      "loss": 0.1515,
+      "step": 8360
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 0.00027944571900523075,
+      "loss": 0.1702,
+      "step": 8370
+    },
+    {
+      "epoch": 2.3,
+      "learning_rate": 0.00027941818849224555,
+      "loss": 0.1435,
+      "step": 8380
+    },
+    {
+      "epoch": 2.3,
+      "learning_rate": 0.00027939065797926034,
+      "loss": 0.138,
+      "step": 8390
+    },
+    {
+      "epoch": 2.3,
+      "learning_rate": 0.0002793631274662751,
+      "loss": 0.1678,
+      "step": 8400
+    },
+    {
+      "epoch": 2.3,
+      "eval_loss": 0.0926092192530632,
+      "eval_runtime": 1050.1998,
+      "eval_samples_per_second": 11.125,
+      "eval_steps_per_second": 1.391,
+      "eval_wer": 0.30953788774674795,
+      "step": 8400
+    },
+    {
+      "epoch": 2.3,
+      "learning_rate": 0.0002793355969532899,
+      "loss": 0.1456,
+      "step": 8410
+    },
+    {
+      "epoch": 2.31,
+      "learning_rate": 0.0002793080664403046,
+      "loss": 0.1874,
+      "step": 8420
+    },
+    {
+      "epoch": 2.31,
+      "learning_rate": 0.0002792805359273194,
+      "loss": 0.1674,
+      "step": 8430
+    },
+    {
+      "epoch": 2.31,
+      "learning_rate": 0.0002792530054143342,
+      "loss": 0.1929,
+      "step": 8440
+    },
+    {
+      "epoch": 2.32,
+      "learning_rate": 0.00027922547490134896,
+      "loss": 0.1529,
+      "step": 8450
+    },
+    {
+      "epoch": 2.32,
+      "learning_rate": 0.00027919794438836375,
+      "loss": 0.1721,
+      "step": 8460
+    },
+    {
+      "epoch": 2.32,
+      "learning_rate": 0.0002791704138753785,
+      "loss": 0.1678,
+      "step": 8470
+    },
+    {
+      "epoch": 2.32,
+      "learning_rate": 0.0002791428833623933,
+      "loss": 0.1474,
+      "step": 8480
+    },
+    {
+      "epoch": 2.33,
+      "learning_rate": 0.0002791153528494081,
+      "loss": 0.1838,
+      "step": 8490
+    },
+    {
+      "epoch": 2.33,
+      "learning_rate": 0.0002790878223364228,
+      "loss": 0.1705,
+      "step": 8500
+    },
+    {
+      "epoch": 2.33,
+      "eval_loss": 0.0972880944609642,
+      "eval_runtime": 1054.6585,
+      "eval_samples_per_second": 11.078,
+      "eval_steps_per_second": 1.385,
+      "eval_wer": 0.3167129082143182,
+      "step": 8500
+    },
+    {
+      "epoch": 2.33,
+      "learning_rate": 0.0002790602918234376,
+      "loss": 0.1696,
+      "step": 8510
+    },
+    {
+      "epoch": 2.33,
+      "learning_rate": 0.00027903276131045236,
+      "loss": 0.145,
+      "step": 8520
+    },
+    {
+      "epoch": 2.34,
+      "learning_rate": 0.00027900523079746716,
+      "loss": 0.1273,
+      "step": 8530
+    },
+    {
+      "epoch": 2.34,
+      "learning_rate": 0.00027897770028448196,
+      "loss": 0.1459,
+      "step": 8540
+    },
+    {
+      "epoch": 2.34,
+      "learning_rate": 0.0002789501697714967,
+      "loss": 0.1447,
+      "step": 8550
+    },
+    {
+      "epoch": 2.35,
+      "learning_rate": 0.0002789226392585115,
+      "loss": 0.1587,
+      "step": 8560
+    },
+    {
+      "epoch": 2.35,
+      "learning_rate": 0.00027889510874552623,
+      "loss": 0.1653,
+      "step": 8570
+    },
+    {
+      "epoch": 2.35,
+      "learning_rate": 0.00027886757823254103,
+      "loss": 0.1427,
+      "step": 8580
+    },
+    {
+      "epoch": 2.35,
+      "learning_rate": 0.0002788400477195558,
+      "loss": 0.1642,
+      "step": 8590
+    },
+    {
+      "epoch": 2.36,
+      "learning_rate": 0.0002788125172065706,
+      "loss": 0.1323,
+      "step": 8600
+    },
+    {
+      "epoch": 2.36,
+      "eval_loss": 0.09097806364297867,
+      "eval_runtime": 1042.0814,
+      "eval_samples_per_second": 11.211,
+      "eval_steps_per_second": 1.402,
+      "eval_wer": 0.30551259892658966,
+      "step": 8600
+    },
+    {
+      "epoch": 2.36,
+      "learning_rate": 0.00027878498669358536,
+      "loss": 0.185,
+      "step": 8610
+    },
+    {
+      "epoch": 2.36,
+      "learning_rate": 0.0002787574561806001,
+      "loss": 0.1429,
+      "step": 8620
+    },
+    {
+      "epoch": 2.36,
+      "learning_rate": 0.0002787299256676149,
+      "loss": 0.1356,
+      "step": 8630
+    },
+    {
+      "epoch": 2.37,
+      "learning_rate": 0.0002787023951546297,
+      "loss": 0.1454,
+      "step": 8640
+    },
+    {
+      "epoch": 2.37,
+      "learning_rate": 0.0002786748646416445,
+      "loss": 0.1389,
+      "step": 8650
+    },
+    {
+      "epoch": 2.37,
+      "learning_rate": 0.00027864733412865923,
+      "loss": 0.1469,
+      "step": 8660
+    },
+    {
+      "epoch": 2.38,
+      "learning_rate": 0.000278619803615674,
+      "loss": 0.149,
+      "step": 8670
+    },
+    {
+      "epoch": 2.38,
+      "learning_rate": 0.00027859227310268877,
+      "loss": 0.1541,
+      "step": 8680
+    },
+    {
+      "epoch": 2.38,
+      "learning_rate": 0.00027856474258970357,
+      "loss": 0.1547,
+      "step": 8690
+    },
+    {
+      "epoch": 2.38,
+      "learning_rate": 0.00027853721207671836,
+      "loss": 0.1258,
+      "step": 8700
+    },
+    {
+      "epoch": 2.38,
+      "eval_loss": 0.0872187614440918,
+      "eval_runtime": 1047.8054,
+      "eval_samples_per_second": 11.15,
+      "eval_steps_per_second": 1.394,
+      "eval_wer": 0.30171472755389794,
+      "step": 8700
+    },
+    {
+      "epoch": 2.39,
+      "learning_rate": 0.0002785096815637331,
+      "loss": 0.1555,
+      "step": 8710
+    },
+    {
+      "epoch": 2.39,
+      "learning_rate": 0.0002784821510507479,
+      "loss": 0.1583,
+      "step": 8720
+    },
+    {
+      "epoch": 2.39,
+      "learning_rate": 0.00027845462053776264,
+      "loss": 0.1569,
+      "step": 8730
+    },
+    {
+      "epoch": 2.39,
+      "learning_rate": 0.00027842709002477744,
+      "loss": 0.1695,
+      "step": 8740
+    },
+    {
+      "epoch": 2.4,
+      "learning_rate": 0.00027839955951179223,
+      "loss": 0.1258,
+      "step": 8750
+    },
+    {
+      "epoch": 2.4,
+      "learning_rate": 0.000278372028998807,
+      "loss": 0.1616,
+      "step": 8760
+    },
+    {
+      "epoch": 2.4,
+      "learning_rate": 0.00027834449848582177,
+      "loss": 0.1586,
+      "step": 8770
+    },
+    {
+      "epoch": 2.41,
+      "learning_rate": 0.0002783169679728365,
+      "loss": 0.1733,
+      "step": 8780
+    },
+    {
+      "epoch": 2.41,
+      "learning_rate": 0.0002782894374598513,
+      "loss": 0.1467,
+      "step": 8790
+    },
+    {
+      "epoch": 2.41,
+      "learning_rate": 0.0002782619069468661,
+      "loss": 0.1536,
+      "step": 8800
+    },
+    {
+      "epoch": 2.41,
+      "eval_loss": 0.09626557677984238,
+      "eval_runtime": 1048.2253,
+      "eval_samples_per_second": 11.146,
+      "eval_steps_per_second": 1.394,
+      "eval_wer": 0.3102542527062676,
+      "step": 8800
+    },
+    {
+      "epoch": 2.41,
+      "learning_rate": 0.00027823437643388084,
+      "loss": 0.1618,
+      "step": 8810
+    },
+    {
+      "epoch": 2.42,
+      "learning_rate": 0.00027820684592089564,
+      "loss": 0.1546,
+      "step": 8820
+    },
+    {
+      "epoch": 2.42,
+      "learning_rate": 0.0002781793154079104,
+      "loss": 0.1415,
+      "step": 8830
+    },
+    {
+      "epoch": 2.42,
+      "learning_rate": 0.0002781517848949252,
+      "loss": 0.1769,
+      "step": 8840
+    },
+    {
+      "epoch": 2.42,
+      "learning_rate": 0.00027812425438194,
+      "loss": 0.1657,
+      "step": 8850
+    },
+    {
+      "epoch": 2.43,
+      "learning_rate": 0.0002780967238689547,
+      "loss": 0.1708,
+      "step": 8860
+    },
+    {
+      "epoch": 2.43,
+      "learning_rate": 0.0002780691933559695,
+      "loss": 0.1486,
+      "step": 8870
+    },
+    {
+      "epoch": 2.43,
+      "learning_rate": 0.00027804166284298425,
+      "loss": 0.1504,
+      "step": 8880
+    },
+    {
+      "epoch": 2.44,
+      "learning_rate": 0.00027801413232999905,
+      "loss": 0.1773,
+      "step": 8890
+    },
+    {
+      "epoch": 2.44,
+      "learning_rate": 0.00027798660181701384,
+      "loss": 0.1628,
+      "step": 8900
+    },
+    {
+      "epoch": 2.44,
+      "eval_loss": 0.08714758604764938,
+      "eval_runtime": 1049.6925,
+      "eval_samples_per_second": 11.13,
+      "eval_steps_per_second": 1.392,
+      "eval_wer": 0.30454607477485673,
+      "step": 8900
+    },
+    {
+      "epoch": 2.44,
+      "learning_rate": 0.00027795907130402864,
+      "loss": 0.1854,
+      "step": 8910
+    },
+    {
+      "epoch": 2.44,
+      "learning_rate": 0.0002779315407910434,
+      "loss": 0.1482,
+      "step": 8920
+    },
+    {
+      "epoch": 2.45,
+      "learning_rate": 0.0002779040102780581,
+      "loss": 0.1622,
+      "step": 8930
+    },
+    {
+      "epoch": 2.45,
+      "learning_rate": 0.0002778764797650729,
+      "loss": 0.1618,
+      "step": 8940
+    },
+    {
+      "epoch": 2.45,
+      "learning_rate": 0.0002778489492520877,
+      "loss": 0.1535,
+      "step": 8950
+    },
+    {
+      "epoch": 2.46,
+      "learning_rate": 0.0002778214187391025,
+      "loss": 0.1579,
+      "step": 8960
+    },
+    {
+      "epoch": 2.46,
+      "learning_rate": 0.00027779388822611725,
+      "loss": 0.144,
+      "step": 8970
+    },
+    {
+      "epoch": 2.46,
+      "learning_rate": 0.000277766357713132,
+      "loss": 0.1676,
+      "step": 8980
+    },
+    {
+      "epoch": 2.46,
+      "learning_rate": 0.0002777388272001468,
+      "loss": 0.1622,
+      "step": 8990
+    },
+    {
+      "epoch": 2.47,
+      "learning_rate": 0.0002777112966871616,
+      "loss": 0.1504,
+      "step": 9000
+    },
+    {
+      "epoch": 2.47,
+      "eval_loss": 0.08976973593235016,
+      "eval_runtime": 1046.6105,
+      "eval_samples_per_second": 11.163,
+      "eval_steps_per_second": 1.396,
+      "eval_wer": 0.3038069680705904,
+      "step": 9000
+    },
+    {
+      "epoch": 2.47,
+      "learning_rate": 0.0002776837661741764,
+      "loss": 0.1543,
+      "step": 9010
+    },
+    {
+      "epoch": 2.47,
+      "learning_rate": 0.0002776562356611911,
+      "loss": 0.1569,
+      "step": 9020
+    },
+    {
+      "epoch": 2.47,
+      "learning_rate": 0.00027762870514820586,
+      "loss": 0.1346,
+      "step": 9030
+    },
+    {
+      "epoch": 2.48,
+      "learning_rate": 0.00027760117463522066,
+      "loss": 0.1653,
+      "step": 9040
+    },
+    {
+      "epoch": 2.48,
+      "learning_rate": 0.00027757364412223546,
+      "loss": 0.1364,
+      "step": 9050
+    },
+    {
+      "epoch": 2.48,
+      "learning_rate": 0.00027754611360925025,
+      "loss": 0.1346,
+      "step": 9060
+    },
+    {
+      "epoch": 2.49,
+      "learning_rate": 0.000277518583096265,
+      "loss": 0.1355,
+      "step": 9070
+    },
+    {
+      "epoch": 2.49,
+      "learning_rate": 0.0002774910525832798,
+      "loss": 0.147,
+      "step": 9080
+    },
+    {
+      "epoch": 2.49,
+      "learning_rate": 0.00027746352207029453,
+      "loss": 0.1582,
+      "step": 9090
+    },
+    {
+      "epoch": 2.49,
+      "learning_rate": 0.0002774359915573093,
+      "loss": 0.1301,
+      "step": 9100
+    },
+    {
+      "epoch": 2.49,
+      "eval_loss": 0.08558912575244904,
+      "eval_runtime": 1045.4399,
+      "eval_samples_per_second": 11.175,
+      "eval_steps_per_second": 1.397,
+      "eval_wer": 0.29659783498590014,
+      "step": 9100
+    },
+    {
+      "epoch": 2.5,
+      "learning_rate": 0.0002774084610443241,
+      "loss": 0.1486,
+      "step": 9110
+    },
+    {
+      "epoch": 2.5,
+      "learning_rate": 0.00027738093053133886,
+      "loss": 0.1662,
+      "step": 9120
+    },
+    {
+      "epoch": 2.5,
+      "learning_rate": 0.00027735340001835366,
+      "loss": 0.1518,
+      "step": 9130
+    },
+    {
+      "epoch": 2.5,
+      "learning_rate": 0.0002773258695053684,
+      "loss": 0.1664,
+      "step": 9140
+    },
+    {
+      "epoch": 2.51,
+      "learning_rate": 0.0002772983389923832,
+      "loss": 0.1394,
+      "step": 9150
+    },
+    {
+      "epoch": 2.51,
+      "learning_rate": 0.000277270808479398,
+      "loss": 0.1721,
+      "step": 9160
+    },
+    {
+      "epoch": 2.51,
+      "learning_rate": 0.00027724327796641273,
+      "loss": 0.1455,
+      "step": 9170
+    },
+    {
+      "epoch": 2.52,
+      "learning_rate": 0.00027721574745342753,
+      "loss": 0.1723,
+      "step": 9180
+    },
+    {
+      "epoch": 2.52,
+      "learning_rate": 0.00027718821694044227,
+      "loss": 0.1507,
+      "step": 9190
+    },
+    {
+      "epoch": 2.52,
+      "learning_rate": 0.00027716068642745707,
+      "loss": 0.1488,
+      "step": 9200
+    },
+    {
+      "epoch": 2.52,
+      "eval_loss": 0.08458111435174942,
+      "eval_runtime": 1050.2942,
+      "eval_samples_per_second": 11.124,
+      "eval_steps_per_second": 1.391,
+      "eval_wer": 0.2977804057127263,
+      "step": 9200
+    },
+    {
+      "epoch": 2.52,
+      "learning_rate": 0.00027713315591447186,
+      "loss": 0.1467,
+      "step": 9210
+    },
+    {
+      "epoch": 2.53,
+      "learning_rate": 0.0002771056254014866,
+      "loss": 0.1567,
+      "step": 9220
+    },
+    {
+      "epoch": 2.53,
+      "learning_rate": 0.0002770780948885014,
+      "loss": 0.1866,
+      "step": 9230
+    },
+    {
+      "epoch": 2.53,
+      "learning_rate": 0.00027705056437551614,
+      "loss": 0.167,
+      "step": 9240
+    },
+    {
+      "epoch": 2.53,
+      "learning_rate": 0.00027702303386253094,
+      "loss": 0.1382,
+      "step": 9250
+    },
+    {
+      "epoch": 2.54,
+      "learning_rate": 0.00027699550334954573,
+      "loss": 0.176,
+      "step": 9260
+    },
+    {
+      "epoch": 2.54,
+      "learning_rate": 0.00027696797283656053,
+      "loss": 0.15,
+      "step": 9270
+    },
+    {
+      "epoch": 2.54,
+      "learning_rate": 0.00027694044232357527,
+      "loss": 0.1464,
+      "step": 9280
+    },
+    {
+      "epoch": 2.55,
+      "learning_rate": 0.00027691291181059,
+      "loss": 0.1787,
+      "step": 9290
+    },
+    {
+      "epoch": 2.55,
+      "learning_rate": 0.0002768853812976048,
+      "loss": 0.1621,
+      "step": 9300
+    },
+    {
+      "epoch": 2.55,
+      "eval_loss": 0.08554688096046448,
+      "eval_runtime": 1061.0591,
+      "eval_samples_per_second": 11.011,
+      "eval_steps_per_second": 1.377,
+      "eval_wer": 0.30240835076867095,
+      "step": 9300
+    },
+    {
+      "epoch": 2.55,
+      "learning_rate": 0.0002768578507846196,
+      "loss": 0.1544,
+      "step": 9310
+    },
+    {
+      "epoch": 2.55,
+      "learning_rate": 0.0002768303202716344,
+      "loss": 0.3587,
+      "step": 9320
+    },
+    {
+      "epoch": 2.56,
+      "learning_rate": 0.00027680278975864914,
+      "loss": 0.169,
+      "step": 9330
+    },
+    {
+      "epoch": 2.56,
+      "learning_rate": 0.0002767752592456639,
+      "loss": 0.1487,
+      "step": 9340
+    },
+    {
+      "epoch": 2.56,
+      "learning_rate": 0.0002767477287326787,
+      "loss": 0.1448,
+      "step": 9350
+    },
+    {
+      "epoch": 2.56,
+      "learning_rate": 0.0002767201982196935,
+      "loss": 0.1492,
+      "step": 9360
+    },
+    {
+      "epoch": 2.57,
+      "learning_rate": 0.00027669266770670827,
+      "loss": 0.1484,
+      "step": 9370
+    },
+    {
+      "epoch": 2.57,
+      "learning_rate": 0.000276665137193723,
+      "loss": 0.142,
+      "step": 9380
+    },
+    {
+      "epoch": 2.57,
+      "learning_rate": 0.00027663760668073775,
+      "loss": 0.1614,
+      "step": 9390
+    },
+    {
+      "epoch": 2.58,
+      "learning_rate": 0.00027661007616775255,
+      "loss": 0.1453,
+      "step": 9400
+    },
+    {
+      "epoch": 2.58,
+      "eval_loss": 0.08304944634437561,
+      "eval_runtime": 1053.8093,
+      "eval_samples_per_second": 11.086,
+      "eval_steps_per_second": 1.386,
+      "eval_wer": 0.2940393887019012,
+      "step": 9400
+    },
+    {
+      "epoch": 2.58,
+      "learning_rate": 0.00027658254565476734,
+      "loss": 0.1516,
+      "step": 9410
+    },
+    {
+      "epoch": 2.58,
+      "learning_rate": 0.00027655501514178214,
+      "loss": 0.1336,
+      "step": 9420
+    },
+    {
+      "epoch": 2.58,
+      "learning_rate": 0.0002765274846287969,
+      "loss": 0.1627,
+      "step": 9430
+    },
+    {
+      "epoch": 2.59,
+      "learning_rate": 0.0002764999541158117,
+      "loss": 0.1617,
+      "step": 9440
+    },
+    {
+      "epoch": 2.59,
+      "learning_rate": 0.0002764724236028264,
+      "loss": 0.1532,
+      "step": 9450
+    },
+    {
+      "epoch": 2.59,
+      "learning_rate": 0.0002764448930898412,
+      "loss": 0.138,
+      "step": 9460
+    },
+    {
+      "epoch": 2.59,
+      "learning_rate": 0.000276417362576856,
+      "loss": 0.1381,
+      "step": 9470
+    },
+    {
+      "epoch": 2.6,
+      "learning_rate": 0.00027638983206387075,
+      "loss": 0.1609,
+      "step": 9480
+    },
+    {
+      "epoch": 2.6,
+      "learning_rate": 0.00027636230155088555,
+      "loss": 0.1559,
+      "step": 9490
+    },
+    {
+      "epoch": 2.6,
+      "learning_rate": 0.0002763347710379003,
+      "loss": 0.13,
+      "step": 9500
+    },
+    {
+      "epoch": 2.6,
+      "eval_loss": 0.08238458633422852,
+      "eval_runtime": 1057.6848,
+      "eval_samples_per_second": 11.046,
+      "eval_steps_per_second": 1.381,
+      "eval_wer": 0.2919585190575821,
+      "step": 9500
+    },
+    {
+      "epoch": 2.61,
+      "learning_rate": 0.0002763072405249151,
+      "loss": 0.1517,
+      "step": 9510
+    },
+    {
+      "epoch": 2.61,
+      "learning_rate": 0.0002762797100119299,
+      "loss": 0.1294,
+      "step": 9520
+    },
+    {
+      "epoch": 2.61,
+      "learning_rate": 0.0002762521794989446,
+      "loss": 0.1423,
+      "step": 9530
+    },
+    {
+      "epoch": 2.61,
+      "learning_rate": 0.0002762246489859594,
+      "loss": 0.1446,
+      "step": 9540
+    },
+    {
+      "epoch": 2.62,
+      "learning_rate": 0.00027619711847297416,
+      "loss": 0.1273,
+      "step": 9550
+    },
+    {
+      "epoch": 2.62,
+      "learning_rate": 0.00027616958795998896,
+      "loss": 0.1656,
+      "step": 9560
+    },
+    {
+      "epoch": 2.62,
+      "learning_rate": 0.00027614205744700375,
+      "loss": 0.1377,
+      "step": 9570
+    },
+    {
+      "epoch": 2.63,
+      "learning_rate": 0.00027611452693401855,
+      "loss": 0.146,
+      "step": 9580
+    },
+    {
+      "epoch": 2.63,
+      "learning_rate": 0.0002760869964210333,
+      "loss": 0.1363,
+      "step": 9590
+    },
+    {
+      "epoch": 2.63,
+      "learning_rate": 0.00027605946590804803,
+      "loss": 0.1225,
+      "step": 9600
+    },
+    {
+      "epoch": 2.63,
+      "eval_loss": 0.08304128050804138,
+      "eval_runtime": 1050.786,
+      "eval_samples_per_second": 11.118,
+      "eval_steps_per_second": 1.39,
+      "eval_wer": 0.29202674429182207,
+      "step": 9600
+    },
+    {
+      "epoch": 2.63,
+      "learning_rate": 0.0002760319353950628,
+      "loss": 0.146,
+      "step": 9610
+    },
+    {
+      "epoch": 2.64,
+      "learning_rate": 0.0002760044048820776,
+      "loss": 0.1588,
+      "step": 9620
+    },
+    {
+      "epoch": 2.64,
+      "learning_rate": 0.0002759768743690924,
+      "loss": 0.1451,
+      "step": 9630
+    },
+    {
+      "epoch": 2.64,
+      "learning_rate": 0.00027594934385610716,
+      "loss": 0.2022,
+      "step": 9640
+    },
+    {
+      "epoch": 2.64,
+      "learning_rate": 0.0002759218133431219,
+      "loss": 0.1589,
+      "step": 9650
+    },
+    {
+      "epoch": 2.65,
+      "learning_rate": 0.0002758942828301367,
+      "loss": 0.1438,
+      "step": 9660
+    },
+    {
+      "epoch": 2.65,
+      "learning_rate": 0.0002758667523171515,
+      "loss": 0.1744,
+      "step": 9670
+    },
+    {
+      "epoch": 2.65,
+      "learning_rate": 0.0002758392218041663,
+      "loss": 0.1413,
+      "step": 9680
+    },
+    {
+      "epoch": 2.66,
+      "learning_rate": 0.00027581169129118103,
+      "loss": 0.143,
+      "step": 9690
+    },
+    {
+      "epoch": 2.66,
+      "learning_rate": 0.00027578416077819577,
+      "loss": 0.1337,
+      "step": 9700
+    },
+    {
+      "epoch": 2.66,
+      "eval_loss": 0.08384636044502258,
+      "eval_runtime": 1047.8124,
+      "eval_samples_per_second": 11.15,
+      "eval_steps_per_second": 1.394,
+      "eval_wer": 0.2913558628217957,
+      "step": 9700
+    },
+    {
+      "epoch": 2.66,
+      "learning_rate": 0.00027575663026521057,
+      "loss": 0.1633,
+      "step": 9710
+    },
+    {
+      "epoch": 2.66,
+      "learning_rate": 0.00027572909975222536,
+      "loss": 0.1326,
+      "step": 9720
+    },
+    {
+      "epoch": 2.67,
+      "learning_rate": 0.00027570156923924016,
+      "loss": 0.1437,
+      "step": 9730
+    },
+    {
+      "epoch": 2.67,
+      "learning_rate": 0.0002756740387262549,
+      "loss": 0.1832,
+      "step": 9740
+    },
+    {
+      "epoch": 2.67,
+      "learning_rate": 0.00027564650821326964,
+      "loss": 0.1174,
+      "step": 9750
+    },
+    {
+      "epoch": 2.67,
+      "learning_rate": 0.00027561897770028444,
+      "loss": 0.1536,
+      "step": 9760
+    },
+    {
+      "epoch": 2.68,
+      "learning_rate": 0.00027559144718729923,
+      "loss": 0.1433,
+      "step": 9770
+    },
+    {
+      "epoch": 2.68,
+      "learning_rate": 0.00027556391667431403,
+      "loss": 0.1524,
+      "step": 9780
+    },
+    {
+      "epoch": 2.68,
+      "learning_rate": 0.00027553638616132877,
+      "loss": 0.1646,
+      "step": 9790
+    },
+    {
+      "epoch": 2.69,
+      "learning_rate": 0.00027550885564834357,
+      "loss": 0.1192,
+      "step": 9800
+    },
+    {
+      "epoch": 2.69,
+      "eval_loss": 0.0845814049243927,
+      "eval_runtime": 1048.3361,
+      "eval_samples_per_second": 11.144,
+      "eval_steps_per_second": 1.394,
+      "eval_wer": 0.29916765214227237,
+      "step": 9800
+    },
+    {
+      "epoch": 2.69,
+      "learning_rate": 0.0002754813251353583,
+      "loss": 0.1384,
+      "step": 9810
+    },
+    {
+      "epoch": 2.69,
+      "learning_rate": 0.0002754537946223731,
+      "loss": 0.1412,
+      "step": 9820
+    },
+    {
+      "epoch": 2.69,
+      "learning_rate": 0.0002754262641093879,
+      "loss": 0.149,
+      "step": 9830
+    },
+    {
+      "epoch": 2.7,
+      "learning_rate": 0.00027539873359640264,
+      "loss": 0.1552,
+      "step": 9840
+    },
+    {
+      "epoch": 2.7,
+      "learning_rate": 0.00027537120308341744,
+      "loss": 0.1293,
+      "step": 9850
+    },
+    {
+      "epoch": 2.7,
+      "learning_rate": 0.0002753436725704322,
+      "loss": 0.1591,
+      "step": 9860
+    },
+    {
+      "epoch": 2.7,
+      "learning_rate": 0.000275316142057447,
+      "loss": 0.1333,
+      "step": 9870
+    },
+    {
+      "epoch": 2.71,
+      "learning_rate": 0.00027528861154446177,
+      "loss": 0.144,
+      "step": 9880
+    },
+    {
+      "epoch": 2.71,
+      "learning_rate": 0.0002752610810314765,
+      "loss": 0.1652,
+      "step": 9890
+    },
+    {
+      "epoch": 2.71,
+      "learning_rate": 0.0002752335505184913,
+      "loss": 0.1478,
+      "step": 9900
+    },
+    {
+      "epoch": 2.71,
+      "eval_loss": 0.07936351001262665,
+      "eval_runtime": 1059.5079,
+      "eval_samples_per_second": 11.027,
+      "eval_steps_per_second": 1.379,
+      "eval_wer": 0.292129082143182,
+      "step": 9900
+    },
+    {
+      "epoch": 2.72,
+      "learning_rate": 0.00027520602000550605,
+      "loss": 0.1519,
+      "step": 9910
+    },
+    {
+      "epoch": 2.72,
+      "learning_rate": 0.00027517848949252084,
+      "loss": 0.1477,
+      "step": 9920
+    },
+    {
+      "epoch": 2.72,
+      "learning_rate": 0.00027515095897953564,
+      "loss": 0.1609,
+      "step": 9930
+    },
+    {
+      "epoch": 2.72,
+      "learning_rate": 0.00027512342846655044,
+      "loss": 0.1521,
+      "step": 9940
+    },
+    {
+      "epoch": 2.73,
+      "learning_rate": 0.0002750958979535652,
+      "loss": 0.1191,
+      "step": 9950
+    },
+    {
+      "epoch": 2.73,
+      "learning_rate": 0.0002750683674405799,
+      "loss": 0.131,
+      "step": 9960
+    },
+    {
+      "epoch": 2.73,
+      "learning_rate": 0.0002750408369275947,
+      "loss": 0.1323,
+      "step": 9970
+    },
+    {
+      "epoch": 2.73,
+      "learning_rate": 0.0002750133064146095,
+      "loss": 0.1182,
+      "step": 9980
+    },
+    {
+      "epoch": 2.74,
+      "learning_rate": 0.0002749857759016243,
+      "loss": 0.1493,
+      "step": 9990
+    },
+    {
+      "epoch": 2.74,
+      "learning_rate": 0.00027495824538863905,
+      "loss": 0.1188,
+      "step": 10000
+    },
+    {
+      "epoch": 2.74,
+      "eval_loss": 0.0771343931555748,
+      "eval_runtime": 1057.7326,
+      "eval_samples_per_second": 11.045,
+      "eval_steps_per_second": 1.381,
+      "eval_wer": 0.28751250795961064,
+      "step": 10000
+    },
+    {
+      "epoch": 2.74,
+      "learning_rate": 0.0002749307148756538,
+      "loss": 0.14,
+      "step": 10010
+    },
+    {
+      "epoch": 2.75,
+      "learning_rate": 0.0002749031843626686,
+      "loss": 0.1393,
+      "step": 10020
+    },
+    {
+      "epoch": 2.75,
+      "learning_rate": 0.0002748756538496834,
+      "loss": 0.1269,
+      "step": 10030
+    },
+    {
+      "epoch": 2.75,
+      "learning_rate": 0.0002748481233366982,
+      "loss": 0.152,
+      "step": 10040
+    },
+    {
+      "epoch": 2.75,
+      "learning_rate": 0.0002748205928237129,
+      "loss": 0.1153,
+      "step": 10050
+    },
+    {
+      "epoch": 2.76,
+      "learning_rate": 0.00027479306231072766,
+      "loss": 0.1319,
+      "step": 10060
+    },
+    {
+      "epoch": 2.76,
+      "learning_rate": 0.00027476553179774246,
+      "loss": 0.1534,
+      "step": 10070
+    },
+    {
+      "epoch": 2.76,
+      "learning_rate": 0.00027473800128475725,
+      "loss": 0.1336,
+      "step": 10080
+    },
+    {
+      "epoch": 2.76,
+      "learning_rate": 0.00027471047077177205,
+      "loss": 0.1335,
+      "step": 10090
+    },
+    {
+      "epoch": 2.77,
+      "learning_rate": 0.0002746829402587868,
+      "loss": 0.2217,
+      "step": 10100
+    },
+    {
+      "epoch": 2.77,
+      "eval_loss": 0.08399680256843567,
+      "eval_runtime": 1055.6916,
+      "eval_samples_per_second": 11.067,
+      "eval_steps_per_second": 1.384,
+      "eval_wer": 0.30015691803875194,
+      "step": 10100
+    },
+    {
+      "epoch": 2.77,
+      "learning_rate": 0.0002746554097458016,
+      "loss": 0.1605,
+      "step": 10110
+    },
+    {
+      "epoch": 2.77,
+      "learning_rate": 0.0002746278792328163,
+      "loss": 0.1481,
+      "step": 10120
+    },
+    {
+      "epoch": 2.78,
+      "learning_rate": 0.0002746003487198311,
+      "loss": 0.1511,
+      "step": 10130
+    },
+    {
+      "epoch": 2.78,
+      "learning_rate": 0.0002745728182068459,
+      "loss": 0.1261,
+      "step": 10140
+    },
+    {
+      "epoch": 2.78,
+      "learning_rate": 0.00027454528769386066,
+      "loss": 0.1192,
+      "step": 10150
+    },
+    {
+      "epoch": 2.78,
+      "learning_rate": 0.00027451775718087545,
+      "loss": 0.1568,
+      "step": 10160
+    },
+    {
+      "epoch": 2.79,
+      "learning_rate": 0.0002744902266678902,
+      "loss": 0.1355,
+      "step": 10170
+    },
+    {
+      "epoch": 2.79,
+      "learning_rate": 0.000274462696154905,
+      "loss": 0.1283,
+      "step": 10180
+    },
+    {
+      "epoch": 2.79,
+      "learning_rate": 0.0002744351656419198,
+      "loss": 0.1693,
+      "step": 10190
+    },
+    {
+      "epoch": 2.79,
+      "learning_rate": 0.00027440763512893453,
+      "loss": 0.1169,
+      "step": 10200
+    },
+    {
+      "epoch": 2.79,
+      "eval_loss": 0.07694612443447113,
+      "eval_runtime": 1046.644,
+      "eval_samples_per_second": 11.162,
+      "eval_steps_per_second": 1.396,
+      "eval_wer": 0.2876375875557173,
+      "step": 10200
+    },
+    {
+      "epoch": 2.8,
+      "learning_rate": 0.0002743801046159493,
+      "loss": 0.165,
+      "step": 10210
+    },
+    {
+      "epoch": 2.8,
+      "learning_rate": 0.00027435257410296407,
+      "loss": 0.1532,
+      "step": 10220
+    },
+    {
+      "epoch": 2.8,
+      "learning_rate": 0.00027432504358997886,
+      "loss": 0.1442,
+      "step": 10230
+    },
+    {
+      "epoch": 2.81,
+      "learning_rate": 0.00027429751307699366,
+      "loss": 0.1493,
+      "step": 10240
+    },
+    {
+      "epoch": 2.81,
+      "learning_rate": 0.00027426998256400845,
+      "loss": 0.1159,
+      "step": 10250
+    },
+    {
+      "epoch": 2.81,
+      "learning_rate": 0.0002742424520510232,
+      "loss": 0.1357,
+      "step": 10260
+    },
+    {
+      "epoch": 2.81,
+      "learning_rate": 0.00027421492153803794,
+      "loss": 0.1288,
+      "step": 10270
+    },
+    {
+      "epoch": 2.82,
+      "learning_rate": 0.00027418739102505273,
+      "loss": 0.1484,
+      "step": 10280
+    },
+    {
+      "epoch": 2.82,
+      "learning_rate": 0.00027415986051206753,
+      "loss": 0.1471,
+      "step": 10290
+    },
+    {
+      "epoch": 2.82,
+      "learning_rate": 0.0002741323299990823,
+      "loss": 0.1334,
+      "step": 10300
+    },
+    {
+      "epoch": 2.82,
+      "eval_loss": 0.07819453626871109,
+      "eval_runtime": 1048.7903,
+      "eval_samples_per_second": 11.14,
+      "eval_steps_per_second": 1.393,
+      "eval_wer": 0.287455653597744,
+      "step": 10300
+    },
+    {
+      "epoch": 2.83,
+      "learning_rate": 0.00027410479948609707,
+      "loss": 0.1331,
+      "step": 10310
+    },
+    {
+      "epoch": 2.83,
+      "learning_rate": 0.0002740772689731118,
+      "loss": 0.151,
+      "step": 10320
+    },
+    {
+      "epoch": 2.83,
+      "learning_rate": 0.0002740497384601266,
+      "loss": 0.1388,
+      "step": 10330
+    },
+    {
+      "epoch": 2.83,
+      "learning_rate": 0.0002740222079471414,
+      "loss": 0.1825,
+      "step": 10340
+    },
+    {
+      "epoch": 2.84,
+      "learning_rate": 0.0002739946774341562,
+      "loss": 0.1315,
+      "step": 10350
+    },
+    {
+      "epoch": 2.84,
+      "learning_rate": 0.00027396714692117094,
+      "loss": 0.1362,
+      "step": 10360
+    },
+    {
+      "epoch": 2.84,
+      "learning_rate": 0.0002739396164081857,
+      "loss": 0.1836,
+      "step": 10370
+    },
+    {
+      "epoch": 2.84,
+      "learning_rate": 0.0002739120858952005,
+      "loss": 0.1388,
+      "step": 10380
+    },
+    {
+      "epoch": 2.85,
+      "learning_rate": 0.00027388455538221527,
+      "loss": 0.1705,
+      "step": 10390
+    },
+    {
+      "epoch": 2.85,
+      "learning_rate": 0.00027385702486923007,
+      "loss": 0.1623,
+      "step": 10400
+    },
+    {
+      "epoch": 2.85,
+      "eval_loss": 0.08330324292182922,
+      "eval_runtime": 1040.7374,
+      "eval_samples_per_second": 11.226,
+      "eval_steps_per_second": 1.404,
+      "eval_wer": 0.2947557536614209,
+      "step": 10400
+    },
+    {
+      "epoch": 2.85,
+      "learning_rate": 0.0002738294943562448,
+      "loss": 0.1416,
+      "step": 10410
+    },
+    {
+      "epoch": 2.86,
+      "learning_rate": 0.00027380196384325955,
+      "loss": 0.1439,
+      "step": 10420
+    },
+    {
+      "epoch": 2.86,
+      "learning_rate": 0.00027377443333027434,
+      "loss": 0.13,
+      "step": 10430
+    },
+    {
+      "epoch": 2.86,
+      "learning_rate": 0.00027374690281728914,
+      "loss": 0.1491,
+      "step": 10440
+    },
+    {
+      "epoch": 2.86,
+      "learning_rate": 0.00027371937230430394,
+      "loss": 0.1427,
+      "step": 10450
+    },
+    {
+      "epoch": 2.87,
+      "learning_rate": 0.0002736918417913187,
+      "loss": 0.1386,
+      "step": 10460
+    },
+    {
+      "epoch": 2.87,
+      "learning_rate": 0.0002736643112783335,
+      "loss": 0.1472,
+      "step": 10470
+    },
+    {
+      "epoch": 2.87,
+      "learning_rate": 0.0002736367807653482,
+      "loss": 0.1345,
+      "step": 10480
+    },
+    {
+      "epoch": 2.87,
+      "learning_rate": 0.000273609250252363,
+      "loss": 0.1459,
+      "step": 10490
+    },
+    {
+      "epoch": 2.88,
+      "learning_rate": 0.0002735817197393778,
+      "loss": 0.2,
+      "step": 10500
+    },
+    {
+      "epoch": 2.88,
+      "eval_loss": 0.07720213383436203,
+      "eval_runtime": 1051.6199,
+      "eval_samples_per_second": 11.11,
+      "eval_steps_per_second": 1.389,
+      "eval_wer": 0.28559083052851814,
+      "step": 10500
+    },
+    {
+      "epoch": 2.88,
+      "learning_rate": 0.00027355418922639255,
+      "loss": 0.1559,
+      "step": 10510
+    },
+    {
+      "epoch": 2.88,
+      "learning_rate": 0.00027352665871340734,
+      "loss": 0.1678,
+      "step": 10520
+    },
+    {
+      "epoch": 2.89,
+      "learning_rate": 0.0002734991282004221,
+      "loss": 0.1393,
+      "step": 10530
+    },
+    {
+      "epoch": 2.89,
+      "learning_rate": 0.0002734715976874369,
+      "loss": 0.1568,
+      "step": 10540
+    },
+    {
+      "epoch": 2.89,
+      "learning_rate": 0.0002734440671744517,
+      "loss": 0.1266,
+      "step": 10550
+    },
+    {
+      "epoch": 2.89,
+      "learning_rate": 0.00027341653666146647,
+      "loss": 0.1457,
+      "step": 10560
+    },
+    {
+      "epoch": 2.9,
+      "learning_rate": 0.0002733890061484812,
+      "loss": 0.1428,
+      "step": 10570
+    },
+    {
+      "epoch": 2.9,
+      "learning_rate": 0.00027336147563549596,
+      "loss": 0.1418,
+      "step": 10580
+    },
+    {
+      "epoch": 2.9,
+      "learning_rate": 0.00027333394512251075,
+      "loss": 0.1239,
+      "step": 10590
+    },
+    {
+      "epoch": 2.9,
+      "learning_rate": 0.00027330641460952555,
+      "loss": 0.1288,
+      "step": 10600
+    },
+    {
+      "epoch": 2.9,
+      "eval_loss": 0.07732350379228592,
+      "eval_runtime": 1052.8488,
+      "eval_samples_per_second": 11.097,
+      "eval_steps_per_second": 1.388,
+      "eval_wer": 0.28539752569817156,
+      "step": 10600
+    },
+    {
+      "epoch": 2.91,
+      "learning_rate": 0.00027327888409654034,
+      "loss": 0.1591,
+      "step": 10610
+    },
+    {
+      "epoch": 2.91,
+      "learning_rate": 0.0002732513535835551,
+      "loss": 0.179,
+      "step": 10620
+    },
+    {
+      "epoch": 2.91,
+      "learning_rate": 0.0002732238230705698,
+      "loss": 0.1627,
+      "step": 10630
+    },
+    {
+      "epoch": 2.92,
+      "learning_rate": 0.0002731962925575846,
+      "loss": 0.151,
+      "step": 10640
+    },
+    {
+      "epoch": 2.92,
+      "learning_rate": 0.0002731687620445994,
+      "loss": 0.1237,
+      "step": 10650
+    },
+    {
+      "epoch": 2.92,
+      "learning_rate": 0.0002731412315316142,
+      "loss": 0.1334,
+      "step": 10660
+    },
+    {
+      "epoch": 2.92,
+      "learning_rate": 0.00027311370101862895,
+      "loss": 0.1311,
+      "step": 10670
+    },
+    {
+      "epoch": 2.93,
+      "learning_rate": 0.0002730861705056437,
+      "loss": 0.1412,
+      "step": 10680
+    },
+    {
+      "epoch": 2.93,
+      "learning_rate": 0.0002730586399926585,
+      "loss": 0.1705,
+      "step": 10690
+    },
+    {
+      "epoch": 2.93,
+      "learning_rate": 0.0002730311094796733,
+      "loss": 0.1201,
+      "step": 10700
+    },
+    {
+      "epoch": 2.93,
+      "eval_loss": 0.07816141843795776,
+      "eval_runtime": 1068.2164,
+      "eval_samples_per_second": 10.937,
+      "eval_steps_per_second": 1.368,
+      "eval_wer": 0.2885017738560902,
+      "step": 10700
+    },
+    {
+      "epoch": 2.93,
+      "learning_rate": 0.0002730035789666881,
+      "loss": 0.1286,
+      "step": 10710
+    },
+    {
+      "epoch": 2.94,
+      "learning_rate": 0.0002729760484537028,
+      "loss": 0.1341,
+      "step": 10720
+    },
+    {
+      "epoch": 2.94,
+      "learning_rate": 0.00027294851794071757,
+      "loss": 0.1418,
+      "step": 10730
+    },
+    {
+      "epoch": 2.94,
+      "learning_rate": 0.00027292098742773236,
+      "loss": 0.1541,
+      "step": 10740
+    },
+    {
+      "epoch": 2.95,
+      "learning_rate": 0.00027289345691474716,
+      "loss": 0.1817,
+      "step": 10750
+    },
+    {
+      "epoch": 2.95,
+      "learning_rate": 0.00027286592640176195,
+      "loss": 0.1397,
+      "step": 10760
+    },
+    {
+      "epoch": 2.95,
+      "learning_rate": 0.0002728383958887767,
+      "loss": 0.1222,
+      "step": 10770
+    },
+    {
+      "epoch": 2.95,
+      "learning_rate": 0.0002728108653757915,
+      "loss": 0.1289,
+      "step": 10780
+    },
+    {
+      "epoch": 2.96,
+      "learning_rate": 0.00027278333486280623,
+      "loss": 0.141,
+      "step": 10790
+    },
+    {
+      "epoch": 2.96,
+      "learning_rate": 0.00027275580434982103,
+      "loss": 0.1467,
+      "step": 10800
+    },
+    {
+      "epoch": 2.96,
+      "eval_loss": 0.07779738306999207,
+      "eval_runtime": 1050.2055,
+      "eval_samples_per_second": 11.124,
+      "eval_steps_per_second": 1.391,
+      "eval_wer": 0.2863640498499045,
+      "step": 10800
+    },
+    {
+      "epoch": 2.96,
+      "learning_rate": 0.0002727282738368358,
+      "loss": 0.1565,
+      "step": 10810
+    },
+    {
+      "epoch": 2.96,
+      "learning_rate": 0.00027270074332385057,
+      "loss": 0.1536,
+      "step": 10820
+    },
+    {
+      "epoch": 2.97,
+      "learning_rate": 0.00027267321281086536,
+      "loss": 0.1532,
+      "step": 10830
+    },
+    {
+      "epoch": 2.97,
+      "learning_rate": 0.0002726456822978801,
+      "loss": 0.1658,
+      "step": 10840
+    },
+    {
+      "epoch": 2.97,
+      "learning_rate": 0.0002726181517848949,
+      "loss": 0.1091,
+      "step": 10850
+    },
+    {
+      "epoch": 2.98,
+      "learning_rate": 0.0002725906212719097,
+      "loss": 0.158,
+      "step": 10860
+    },
+    {
+      "epoch": 2.98,
+      "learning_rate": 0.00027256309075892444,
+      "loss": 0.1538,
+      "step": 10870
+    },
+    {
+      "epoch": 2.98,
+      "learning_rate": 0.00027253556024593923,
+      "loss": 0.1713,
+      "step": 10880
+    },
+    {
+      "epoch": 2.98,
+      "learning_rate": 0.000272508029732954,
+      "loss": 0.1597,
+      "step": 10890
+    },
+    {
+      "epoch": 2.99,
+      "learning_rate": 0.00027248049921996877,
+      "loss": 0.1452,
+      "step": 10900
+    },
+    {
+      "epoch": 2.99,
+      "eval_loss": 0.0746755301952362,
+      "eval_runtime": 1043.7631,
+      "eval_samples_per_second": 11.193,
+      "eval_steps_per_second": 1.4,
+      "eval_wer": 0.28247521149822613,
+      "step": 10900
+    },
+    {
+      "epoch": 2.99,
+      "learning_rate": 0.00027245296870698357,
+      "loss": 0.1357,
+      "step": 10910
+    },
+    {
+      "epoch": 2.99,
+      "learning_rate": 0.00027242543819399836,
+      "loss": 0.1308,
+      "step": 10920
+    },
+    {
+      "epoch": 2.99,
+      "learning_rate": 0.0002723979076810131,
+      "loss": 0.1342,
+      "step": 10930
+    },
+    {
+      "epoch": 3.0,
+      "learning_rate": 0.00027237037716802784,
+      "loss": 0.1607,
+      "step": 10940
+    },
+    {
+      "epoch": 3.0,
+      "learning_rate": 0.00027234284665504264,
+      "loss": 0.1492,
+      "step": 10950
+    },
+    {
+      "epoch": 3.0,
+      "learning_rate": 0.00027231531614205744,
+      "loss": 0.1423,
+      "step": 10960
+    },
+    {
+      "epoch": 3.01,
+      "learning_rate": 0.00027228778562907223,
+      "loss": 0.1281,
+      "step": 10970
+    },
+    {
+      "epoch": 3.01,
+      "learning_rate": 0.000272260255116087,
+      "loss": 0.1284,
+      "step": 10980
+    },
+    {
+      "epoch": 3.01,
+      "learning_rate": 0.0002722327246031017,
+      "loss": 0.1524,
+      "step": 10990
+    },
+    {
+      "epoch": 3.01,
+      "learning_rate": 0.0002722051940901165,
+      "loss": 0.1182,
+      "step": 11000
+    },
+    {
+      "epoch": 3.01,
+      "eval_loss": 0.07591033726930618,
+      "eval_runtime": 1055.6051,
+      "eval_samples_per_second": 11.068,
+      "eval_steps_per_second": 1.384,
+      "eval_wer": 0.28122441553716,
+      "step": 11000
+    },
+    {
+      "epoch": 3.02,
+      "learning_rate": 0.0002721776635771313,
+      "loss": 0.1261,
+      "step": 11010
+    },
+    {
+      "epoch": 3.02,
+      "learning_rate": 0.0002721501330641461,
+      "loss": 0.1208,
+      "step": 11020
+    },
+    {
+      "epoch": 3.02,
+      "learning_rate": 0.00027212260255116084,
+      "loss": 0.1195,
+      "step": 11030
+    },
+    {
+      "epoch": 3.03,
+      "learning_rate": 0.0002720950720381756,
+      "loss": 0.1374,
+      "step": 11040
+    },
+    {
+      "epoch": 3.03,
+      "learning_rate": 0.0002720675415251904,
+      "loss": 0.1129,
+      "step": 11050
+    },
+    {
+      "epoch": 3.03,
+      "learning_rate": 0.0002720400110122052,
+      "loss": 0.1203,
+      "step": 11060
+    },
+    {
+      "epoch": 3.03,
+      "learning_rate": 0.00027201248049921997,
+      "loss": 0.1218,
+      "step": 11070
+    },
+    {
+      "epoch": 3.04,
+      "learning_rate": 0.0002719849499862347,
+      "loss": 0.1215,
+      "step": 11080
+    },
+    {
+      "epoch": 3.04,
+      "learning_rate": 0.0002719574194732495,
+      "loss": 0.1353,
+      "step": 11090
+    },
+    {
+      "epoch": 3.04,
+      "learning_rate": 0.00027192988896026425,
+      "loss": 0.1078,
+      "step": 11100
+    },
+    {
+      "epoch": 3.04,
+      "eval_loss": 0.07436859607696533,
+      "eval_runtime": 1060.0152,
+      "eval_samples_per_second": 11.022,
+      "eval_steps_per_second": 1.378,
+      "eval_wer": 0.27713090148276176,
+      "step": 11100
+    },
+    {
+      "epoch": 3.04,
+      "learning_rate": 0.00027190235844727905,
+      "loss": 0.122,
+      "step": 11110
+    },
+    {
+      "epoch": 3.05,
+      "learning_rate": 0.00027187482793429384,
+      "loss": 0.1207,
+      "step": 11120
+    },
+    {
+      "epoch": 3.05,
+      "learning_rate": 0.0002718472974213086,
+      "loss": 0.1236,
+      "step": 11130
+    },
+    {
+      "epoch": 3.05,
+      "learning_rate": 0.0002718197669083234,
+      "loss": 0.1423,
+      "step": 11140
+    },
+    {
+      "epoch": 3.06,
+      "learning_rate": 0.0002717922363953381,
+      "loss": 0.1057,
+      "step": 11150
+    },
+    {
+      "epoch": 3.06,
+      "learning_rate": 0.0002717647058823529,
+      "loss": 0.1309,
+      "step": 11160
+    },
+    {
+      "epoch": 3.06,
+      "learning_rate": 0.0002717371753693677,
+      "loss": 0.1566,
+      "step": 11170
+    },
+    {
+      "epoch": 3.06,
+      "learning_rate": 0.00027170964485638245,
+      "loss": 0.1259,
+      "step": 11180
+    },
+    {
+      "epoch": 3.07,
+      "learning_rate": 0.00027168211434339725,
+      "loss": 0.1369,
+      "step": 11190
+    },
+    {
+      "epoch": 3.07,
+      "learning_rate": 0.000271654583830412,
+      "loss": 0.1426,
+      "step": 11200
+    },
+    {
+      "epoch": 3.07,
+      "eval_loss": 0.07973187416791916,
+      "eval_runtime": 1054.5609,
+      "eval_samples_per_second": 11.079,
+      "eval_steps_per_second": 1.385,
+      "eval_wer": 0.28829709815337035,
+      "step": 11200
+    },
+    {
+      "epoch": 3.07,
+      "learning_rate": 0.0002716270533174268,
+      "loss": 0.135,
+      "step": 11210
+    },
+    {
+      "epoch": 3.07,
+      "learning_rate": 0.0002715995228044416,
+      "loss": 0.1249,
+      "step": 11220
+    },
+    {
+      "epoch": 3.08,
+      "learning_rate": 0.0002715719922914564,
+      "loss": 0.1414,
+      "step": 11230
+    },
+    {
+      "epoch": 3.08,
+      "learning_rate": 0.0002715444617784711,
+      "loss": 0.1297,
+      "step": 11240
+    },
+    {
+      "epoch": 3.08,
+      "learning_rate": 0.00027151693126548586,
+      "loss": 0.1273,
+      "step": 11250
+    },
+    {
+      "epoch": 3.09,
+      "learning_rate": 0.00027148940075250066,
+      "loss": 0.1372,
+      "step": 11260
+    },
+    {
+      "epoch": 3.09,
+      "learning_rate": 0.00027146187023951545,
+      "loss": 0.1226,
+      "step": 11270
+    },
+    {
+      "epoch": 3.09,
+      "learning_rate": 0.00027143433972653025,
+      "loss": 0.1658,
+      "step": 11280
+    },
+    {
+      "epoch": 3.09,
+      "learning_rate": 0.000271406809213545,
+      "loss": 0.1384,
+      "step": 11290
+    },
+    {
+      "epoch": 3.1,
+      "learning_rate": 0.00027137927870055973,
+      "loss": 0.1322,
+      "step": 11300
+    },
+    {
+      "epoch": 3.1,
+      "eval_loss": 0.07649017870426178,
+      "eval_runtime": 1041.843,
+      "eval_samples_per_second": 11.214,
+      "eval_steps_per_second": 1.402,
+      "eval_wer": 0.2796552351496407,
+      "step": 11300
+    },
+    {
+      "epoch": 3.1,
+      "learning_rate": 0.00027135174818757453,
+      "loss": 0.1446,
+      "step": 11310
+    },
+    {
+      "epoch": 3.1,
+      "learning_rate": 0.0002713242176745893,
+      "loss": 0.1355,
+      "step": 11320
+    },
+    {
+      "epoch": 3.1,
+      "learning_rate": 0.0002712966871616041,
+      "loss": 0.1253,
+      "step": 11330
+    },
+    {
+      "epoch": 3.11,
+      "learning_rate": 0.00027126915664861886,
+      "loss": 0.1483,
+      "step": 11340
+    },
+    {
+      "epoch": 3.11,
+      "learning_rate": 0.0002712416261356336,
+      "loss": 0.1045,
+      "step": 11350
+    },
+    {
+      "epoch": 3.11,
+      "learning_rate": 0.0002712140956226484,
+      "loss": 0.1288,
+      "step": 11360
+    },
+    {
+      "epoch": 3.12,
+      "learning_rate": 0.0002711865651096632,
+      "loss": 0.1186,
+      "step": 11370
+    },
+    {
+      "epoch": 3.12,
+      "learning_rate": 0.000271159034596678,
+      "loss": 0.1238,
+      "step": 11380
+    },
+    {
+      "epoch": 3.12,
+      "learning_rate": 0.00027113150408369273,
+      "loss": 0.1288,
+      "step": 11390
+    },
+    {
+      "epoch": 3.12,
+      "learning_rate": 0.0002711039735707075,
+      "loss": 0.1655,
+      "step": 11400
+    },
+    {
+      "epoch": 3.12,
+      "eval_loss": 0.07433360815048218,
+      "eval_runtime": 1060.5558,
+      "eval_samples_per_second": 11.016,
+      "eval_steps_per_second": 1.378,
+      "eval_wer": 0.28851314472846357,
+      "step": 11400
+    },
+    {
+      "epoch": 3.13,
+      "learning_rate": 0.00027107644305772227,
+      "loss": 0.1117,
+      "step": 11410
+    },
+    {
+      "epoch": 3.13,
+      "learning_rate": 0.00027104891254473707,
+      "loss": 0.1776,
+      "step": 11420
+    },
+    {
+      "epoch": 3.13,
+      "learning_rate": 0.00027102138203175186,
+      "loss": 0.1274,
+      "step": 11430
+    },
+    {
+      "epoch": 3.13,
+      "learning_rate": 0.0002709938515187666,
+      "loss": 0.1339,
+      "step": 11440
+    },
+    {
+      "epoch": 3.14,
+      "learning_rate": 0.0002709663210057814,
+      "loss": 0.1317,
+      "step": 11450
+    },
+    {
+      "epoch": 3.14,
+      "learning_rate": 0.00027093879049279614,
+      "loss": 0.1253,
+      "step": 11460
+    },
+    {
+      "epoch": 3.14,
+      "learning_rate": 0.00027091125997981094,
+      "loss": 0.1291,
+      "step": 11470
+    },
+    {
+      "epoch": 3.15,
+      "learning_rate": 0.00027088372946682573,
+      "loss": 0.132,
+      "step": 11480
+    },
+    {
+      "epoch": 3.15,
+      "learning_rate": 0.0002708561989538405,
+      "loss": 0.1593,
+      "step": 11490
+    },
+    {
+      "epoch": 3.15,
+      "learning_rate": 0.00027082866844085527,
+      "loss": 0.1243,
+      "step": 11500
+    },
+    {
+      "epoch": 3.15,
+      "eval_loss": 0.07442672550678253,
+      "eval_runtime": 1063.6769,
+      "eval_samples_per_second": 10.984,
+      "eval_steps_per_second": 1.374,
+      "eval_wer": 0.2792231419994542,
+      "step": 11500
+    },
+    {
+      "epoch": 3.15,
+      "learning_rate": 0.00027080113792787,
+      "loss": 0.1272,
+      "step": 11510
+    },
+    {
+      "epoch": 3.16,
+      "learning_rate": 0.0002707736074148848,
+      "loss": 0.1373,
+      "step": 11520
+    },
+    {
+      "epoch": 3.16,
+      "learning_rate": 0.0002707460769018996,
+      "loss": 0.1241,
+      "step": 11530
+    },
+    {
+      "epoch": 3.16,
+      "learning_rate": 0.00027071854638891434,
+      "loss": 0.1344,
+      "step": 11540
+    },
+    {
+      "epoch": 3.16,
+      "learning_rate": 0.00027069101587592914,
+      "loss": 0.1718,
+      "step": 11550
+    },
+    {
+      "epoch": 3.17,
+      "learning_rate": 0.0002706634853629439,
+      "loss": 0.1483,
+      "step": 11560
+    },
+    {
+      "epoch": 3.17,
+      "learning_rate": 0.0002706359548499587,
+      "loss": 0.1443,
+      "step": 11570
+    },
+    {
+      "epoch": 3.17,
+      "learning_rate": 0.00027060842433697347,
+      "loss": 0.143,
+      "step": 11580
+    },
+    {
+      "epoch": 3.18,
+      "learning_rate": 0.00027058089382398827,
+      "loss": 0.1322,
+      "step": 11590
+    },
+    {
+      "epoch": 3.18,
+      "learning_rate": 0.000270553363311003,
+      "loss": 0.1724,
+      "step": 11600
+    },
+    {
+      "epoch": 3.18,
+      "eval_loss": 0.07493375241756439,
+      "eval_runtime": 1056.542,
+      "eval_samples_per_second": 11.058,
+      "eval_steps_per_second": 1.383,
+      "eval_wer": 0.2777676703356681,
+      "step": 11600
+    },
+    {
+      "epoch": 3.18,
+      "learning_rate": 0.00027052583279801775,
+      "loss": 0.1393,
+      "step": 11610
+    },
+    {
+      "epoch": 3.18,
+      "learning_rate": 0.00027049830228503255,
+      "loss": 0.1258,
+      "step": 11620
+    },
+    {
+      "epoch": 3.19,
+      "learning_rate": 0.00027047077177204734,
+      "loss": 0.1246,
+      "step": 11630
+    },
+    {
+      "epoch": 3.19,
+      "learning_rate": 0.00027044324125906214,
+      "loss": 0.1178,
+      "step": 11640
+    },
+    {
+      "epoch": 3.19,
+      "learning_rate": 0.0002704157107460769,
+      "loss": 0.1238,
+      "step": 11650
+    },
+    {
+      "epoch": 3.19,
+      "learning_rate": 0.0002703881802330916,
+      "loss": 0.1403,
+      "step": 11660
+    },
+    {
+      "epoch": 3.2,
+      "learning_rate": 0.0002703606497201064,
+      "loss": 0.147,
+      "step": 11670
+    },
+    {
+      "epoch": 3.2,
+      "learning_rate": 0.0002703331192071212,
+      "loss": 0.139,
+      "step": 11680
+    },
+    {
+      "epoch": 3.2,
+      "learning_rate": 0.000270305588694136,
+      "loss": 0.1418,
+      "step": 11690
+    },
+    {
+      "epoch": 3.21,
+      "learning_rate": 0.00027027805818115075,
+      "loss": 0.1136,
+      "step": 11700
+    },
+    {
+      "epoch": 3.21,
+      "eval_loss": 0.07298364490270615,
+      "eval_runtime": 1068.3256,
+      "eval_samples_per_second": 10.936,
+      "eval_steps_per_second": 1.368,
+      "eval_wer": 0.27636905303374876,
+      "step": 11700
+    },
+    {
+      "epoch": 3.21,
+      "learning_rate": 0.0002702505276681655,
+      "loss": 0.12,
+      "step": 11710
+    },
+    {
+      "epoch": 3.21,
+      "learning_rate": 0.0002702229971551803,
+      "loss": 0.3796,
+      "step": 11720
+    },
+    {
+      "epoch": 3.21,
+      "learning_rate": 0.0002701954666421951,
+      "loss": 0.125,
+      "step": 11730
+    },
+    {
+      "epoch": 3.22,
+      "learning_rate": 0.0002701679361292099,
+      "loss": 0.1521,
+      "step": 11740
+    },
+    {
+      "epoch": 3.22,
+      "learning_rate": 0.0002701404056162246,
+      "loss": 0.1114,
+      "step": 11750
+    },
+    {
+      "epoch": 3.22,
+      "learning_rate": 0.0002701128751032394,
+      "loss": 0.1502,
+      "step": 11760
+    },
+    {
+      "epoch": 3.23,
+      "learning_rate": 0.00027008534459025416,
+      "loss": 0.1327,
+      "step": 11770
+    },
+    {
+      "epoch": 3.23,
+      "learning_rate": 0.00027005781407726895,
+      "loss": 0.1269,
+      "step": 11780
+    },
+    {
+      "epoch": 3.23,
+      "learning_rate": 0.00027003028356428375,
+      "loss": 0.1491,
+      "step": 11790
+    },
+    {
+      "epoch": 3.23,
+      "learning_rate": 0.0002700027530512985,
+      "loss": 0.1428,
+      "step": 11800
+    },
+    {
+      "epoch": 3.23,
+      "eval_loss": 0.07759026437997818,
+      "eval_runtime": 1063.5053,
+      "eval_samples_per_second": 10.985,
+      "eval_steps_per_second": 1.374,
+      "eval_wer": 0.283578186118439,
+      "step": 11800
+    },
+    {
+      "epoch": 3.24,
+      "learning_rate": 0.0002699752225383133,
+      "loss": 0.1508,
+      "step": 11810
+    },
+    {
+      "epoch": 3.24,
+      "learning_rate": 0.00026994769202532803,
+      "loss": 0.1398,
+      "step": 11820
+    },
+    {
+      "epoch": 3.24,
+      "learning_rate": 0.0002699201615123428,
+      "loss": 0.1143,
+      "step": 11830
+    },
+    {
+      "epoch": 3.24,
+      "learning_rate": 0.0002698926309993576,
+      "loss": 0.132,
+      "step": 11840
+    },
+    {
+      "epoch": 3.25,
+      "learning_rate": 0.00026986510048637236,
+      "loss": 0.1306,
+      "step": 11850
+    },
+    {
+      "epoch": 3.25,
+      "learning_rate": 0.00026983756997338716,
+      "loss": 0.1439,
+      "step": 11860
+    },
+    {
+      "epoch": 3.25,
+      "learning_rate": 0.0002698100394604019,
+      "loss": 0.1557,
+      "step": 11870
+    },
+    {
+      "epoch": 3.26,
+      "learning_rate": 0.0002697825089474167,
+      "loss": 0.1394,
+      "step": 11880
+    },
+    {
+      "epoch": 3.26,
+      "learning_rate": 0.0002697549784344315,
+      "loss": 0.1352,
+      "step": 11890
+    },
+    {
+      "epoch": 3.26,
+      "learning_rate": 0.0002697274479214463,
+      "loss": 0.1189,
+      "step": 11900
+    },
+    {
+      "epoch": 3.26,
+      "eval_loss": 0.07876188308000565,
+      "eval_runtime": 1048.6869,
+      "eval_samples_per_second": 11.141,
+      "eval_steps_per_second": 1.393,
+      "eval_wer": 0.2892636223051033,
+      "step": 11900
+    },
+    {
+      "epoch": 3.26,
+      "learning_rate": 0.00026969991740846103,
+      "loss": 0.1477,
+      "step": 11910
+    },
+    {
+      "epoch": 3.27,
+      "learning_rate": 0.00026967238689547577,
+      "loss": 0.1309,
+      "step": 11920
+    },
+    {
+      "epoch": 3.27,
+      "learning_rate": 0.00026964485638249057,
+      "loss": 0.1291,
+      "step": 11930
+    },
+    {
+      "epoch": 3.27,
+      "learning_rate": 0.00026961732586950536,
+      "loss": 0.1233,
+      "step": 11940
+    },
+    {
+      "epoch": 3.27,
+      "learning_rate": 0.00026958979535652016,
+      "loss": 0.1144,
+      "step": 11950
+    },
+    {
+      "epoch": 3.28,
+      "learning_rate": 0.0002695622648435349,
+      "loss": 0.1471,
+      "step": 11960
+    },
+    {
+      "epoch": 3.28,
+      "learning_rate": 0.00026953473433054964,
+      "loss": 0.1418,
+      "step": 11970
+    },
+    {
+      "epoch": 3.28,
+      "learning_rate": 0.00026950720381756444,
+      "loss": 0.1284,
+      "step": 11980
+    },
+    {
+      "epoch": 3.29,
+      "learning_rate": 0.00026947967330457923,
+      "loss": 0.1463,
+      "step": 11990
+    },
+    {
+      "epoch": 3.29,
+      "learning_rate": 0.00026945214279159403,
+      "loss": 0.1065,
+      "step": 12000
+    },
+    {
+      "epoch": 3.29,
+      "eval_loss": 0.0732453316450119,
+      "eval_runtime": 1048.1918,
+      "eval_samples_per_second": 11.146,
+      "eval_steps_per_second": 1.394,
+      "eval_wer": 0.2746179386882562,
+      "step": 12000
+    },
+    {
+      "epoch": 3.29,
+      "learning_rate": 0.00026942461227860877,
+      "loss": 0.1389,
+      "step": 12010
+    },
+    {
+      "epoch": 3.29,
+      "learning_rate": 0.0002693970817656235,
+      "loss": 0.1317,
+      "step": 12020
+    },
+    {
+      "epoch": 3.3,
+      "learning_rate": 0.0002693695512526383,
+      "loss": 0.1402,
+      "step": 12030
+    },
+    {
+      "epoch": 3.3,
+      "learning_rate": 0.0002693420207396531,
+      "loss": 0.1282,
+      "step": 12040
+    },
+    {
+      "epoch": 3.3,
+      "learning_rate": 0.0002693144902266679,
+      "loss": 0.1132,
+      "step": 12050
+    },
+    {
+      "epoch": 3.3,
+      "learning_rate": 0.00026928695971368264,
+      "loss": 0.1618,
+      "step": 12060
+    },
+    {
+      "epoch": 3.31,
+      "learning_rate": 0.0002692594292006974,
+      "loss": 0.1398,
+      "step": 12070
+    },
+    {
+      "epoch": 3.31,
+      "learning_rate": 0.0002692318986877122,
+      "loss": 0.1341,
+      "step": 12080
+    },
+    {
+      "epoch": 3.31,
+      "learning_rate": 0.00026920436817472697,
+      "loss": 0.1472,
+      "step": 12090
+    },
+    {
+      "epoch": 3.32,
+      "learning_rate": 0.00026917683766174177,
+      "loss": 0.14,
+      "step": 12100
+    },
+    {
+      "epoch": 3.32,
+      "eval_loss": 0.07203283160924911,
+      "eval_runtime": 1117.1414,
+      "eval_samples_per_second": 10.458,
+      "eval_steps_per_second": 1.308,
+      "eval_wer": 0.27564131720185575,
+      "step": 12100
+    },
+    {
+      "epoch": 3.32,
+      "learning_rate": 0.00010876454293628807,
+      "loss": 8.3306,
+      "step": 12110
+    },
+    {
+      "epoch": 3.32,
+      "learning_rate": 0.0001085983379501385,
+      "loss": 3.3464,
+      "step": 12120
+    },
+    {
+      "epoch": 3.32,
+      "learning_rate": 0.00010843213296398891,
+      "loss": 3.1357,
+      "step": 12130
+    },
+    {
+      "epoch": 3.33,
+      "learning_rate": 0.00010826592797783933,
+      "loss": 3.1124,
+      "step": 12140
+    },
+    {
+      "epoch": 3.33,
+      "learning_rate": 0.0001081163434903047,
+      "loss": 3.1601,
+      "step": 12150
+    },
+    {
+      "epoch": 3.33,
+      "learning_rate": 0.00010795013850415512,
+      "loss": 3.103,
+      "step": 12160
+    },
+    {
+      "epoch": 3.33,
+      "learning_rate": 0.00010778393351800553,
+      "loss": 3.0144,
+      "step": 12170
+    },
+    {
+      "epoch": 3.34,
+      "learning_rate": 0.00010761772853185595,
+      "loss": 3.0122,
+      "step": 12180
+    },
+    {
+      "epoch": 3.34,
+      "learning_rate": 0.00010745152354570635,
+      "loss": 2.9556,
+      "step": 12190
+    },
+    {
+      "epoch": 3.34,
+      "learning_rate": 0.00010728531855955679,
+      "loss": 2.7696,
+      "step": 12200
+    },
+    {
+      "epoch": 3.34,
+      "eval_loss": 2.8195414543151855,
+      "eval_runtime": 1007.6166,
+      "eval_samples_per_second": 11.595,
+      "eval_steps_per_second": 1.45,
+      "eval_wer": 1.0,
+      "step": 12200
+    },
+    {
+      "epoch": 3.35,
+      "learning_rate": 0.00010711911357340719,
+      "loss": 4.4042,
+      "step": 12210
+    },
+    {
+      "epoch": 3.35,
+      "learning_rate": 0.00010695290858725761,
+      "loss": 3.3022,
+      "step": 12220
+    },
+    {
+      "epoch": 3.35,
+      "learning_rate": 0.00010678670360110802,
+      "loss": 3.0588,
+      "step": 12230
+    },
+    {
+      "epoch": 3.35,
+      "learning_rate": 0.00010662049861495845,
+      "loss": 3.0238,
+      "step": 12240
+    },
+    {
+      "epoch": 3.36,
+      "learning_rate": 0.00010645429362880886,
+      "loss": 2.9607,
+      "step": 12250
+    },
+    {
+      "epoch": 3.36,
+      "learning_rate": 0.00010628808864265928,
+      "loss": 2.9219,
+      "step": 12260
+    },
+    {
+      "epoch": 3.36,
+      "learning_rate": 0.00010612188365650968,
+      "loss": 2.7164,
+      "step": 12270
+    },
+    {
+      "epoch": 3.36,
+      "learning_rate": 0.0001059556786703601,
+      "loss": 2.273,
+      "step": 12280
+    },
+    {
+      "epoch": 3.37,
+      "learning_rate": 0.00010578947368421051,
+      "loss": 1.479,
+      "step": 12290
+    },
+    {
+      "epoch": 3.37,
+      "learning_rate": 0.00010562326869806094,
+      "loss": 0.7388,
+      "step": 12300
+    },
+    {
+      "epoch": 3.37,
+      "eval_loss": 0.42925623059272766,
+      "eval_runtime": 992.4477,
+      "eval_samples_per_second": 11.772,
+      "eval_steps_per_second": 1.472,
+      "eval_wer": 0.7220845083234786,
+      "step": 12300
+    },
+    {
+      "epoch": 3.37,
+      "learning_rate": 0.00010545706371191135,
+      "loss": 0.5404,
+      "step": 12310
+    },
+    {
+      "epoch": 3.38,
+      "learning_rate": 0.00010529085872576177,
+      "loss": 0.3253,
+      "step": 12320
+    },
+    {
+      "epoch": 3.38,
+      "learning_rate": 0.00010512465373961217,
+      "loss": 0.2805,
+      "step": 12330
+    },
+    {
+      "epoch": 3.38,
+      "learning_rate": 0.00010495844875346259,
+      "loss": 0.2625,
+      "step": 12340
+    },
+    {
+      "epoch": 3.38,
+      "learning_rate": 0.00010479224376731301,
+      "loss": 0.2047,
+      "step": 12350
+    },
+    {
+      "epoch": 3.39,
+      "learning_rate": 0.00010462603878116343,
+      "loss": 0.1923,
+      "step": 12360
+    },
+    {
+      "epoch": 3.39,
+      "learning_rate": 0.00010445983379501384,
+      "loss": 0.1929,
+      "step": 12370
+    },
+    {
+      "epoch": 3.39,
+      "learning_rate": 0.00010429362880886426,
+      "loss": 0.1813,
+      "step": 12380
+    },
+    {
+      "epoch": 3.39,
+      "learning_rate": 0.00010412742382271466,
+      "loss": 0.1785,
+      "step": 12390
+    },
+    {
+      "epoch": 3.4,
+      "learning_rate": 0.0001039612188365651,
+      "loss": 0.1829,
+      "step": 12400
+    },
+    {
+      "epoch": 3.4,
+      "eval_loss": 0.09247562289237976,
+      "eval_runtime": 991.3415,
+      "eval_samples_per_second": 11.785,
+      "eval_steps_per_second": 1.474,
+      "eval_wer": 0.29330028199763486,
+      "step": 12400
+    },
+    {
+      "epoch": 3.4,
+      "learning_rate": 0.0001037950138504155,
+      "loss": 0.2012,
+      "step": 12410
+    },
+    {
+      "epoch": 3.4,
+      "learning_rate": 0.00010362880886426592,
+      "loss": 0.151,
+      "step": 12420
+    },
+    {
+      "epoch": 3.41,
+      "learning_rate": 0.00010346260387811633,
+      "loss": 0.1492,
+      "step": 12430
+    },
+    {
+      "epoch": 3.41,
+      "learning_rate": 0.00010329639889196675,
+      "loss": 0.1593,
+      "step": 12440
+    },
+    {
+      "epoch": 3.41,
+      "learning_rate": 0.00010313019390581715,
+      "loss": 0.1294,
+      "step": 12450
+    },
+    {
+      "epoch": 3.41,
+      "learning_rate": 0.00010296398891966759,
+      "loss": 0.1458,
+      "step": 12460
+    },
+    {
+      "epoch": 3.42,
+      "learning_rate": 0.00010279778393351799,
+      "loss": 0.1639,
+      "step": 12470
+    },
+    {
+      "epoch": 3.42,
+      "learning_rate": 0.00010263157894736841,
+      "loss": 0.1313,
+      "step": 12480
+    },
+    {
+      "epoch": 3.42,
+      "learning_rate": 0.00010246537396121882,
+      "loss": 0.1751,
+      "step": 12490
+    },
+    {
+      "epoch": 3.42,
+      "learning_rate": 0.00010229916897506925,
+      "loss": 0.1534,
+      "step": 12500
+    },
+    {
+      "epoch": 3.42,
+      "eval_loss": 0.07944495230913162,
+      "eval_runtime": 989.7561,
+      "eval_samples_per_second": 11.804,
+      "eval_steps_per_second": 1.476,
+      "eval_wer": 0.2762780860547621,
+      "step": 12500
+    },
+    {
+      "epoch": 3.43,
+      "learning_rate": 0.00010213296398891966,
+      "loss": 0.1306,
+      "step": 12510
+    },
+    {
+      "epoch": 3.43,
+      "learning_rate": 0.00010196675900277008,
+      "loss": 0.1483,
+      "step": 12520
+    },
+    {
+      "epoch": 3.43,
+      "learning_rate": 0.00010180055401662048,
+      "loss": 0.1643,
+      "step": 12530
+    },
+    {
+      "epoch": 3.44,
+      "learning_rate": 0.0001016343490304709,
+      "loss": 0.1494,
+      "step": 12540
+    },
+    {
+      "epoch": 3.44,
+      "learning_rate": 0.00010146814404432131,
+      "loss": 0.1435,
+      "step": 12550
+    },
+    {
+      "epoch": 3.44,
+      "learning_rate": 0.00010130193905817174,
+      "loss": 0.135,
+      "step": 12560
+    },
+    {
+      "epoch": 3.44,
+      "learning_rate": 0.00010113573407202215,
+      "loss": 0.1873,
+      "step": 12570
+    },
+    {
+      "epoch": 3.45,
+      "learning_rate": 0.00010096952908587257,
+      "loss": 0.1526,
+      "step": 12580
+    },
+    {
+      "epoch": 3.45,
+      "learning_rate": 0.00010080332409972297,
+      "loss": 0.1725,
+      "step": 12590
+    },
+    {
+      "epoch": 3.45,
+      "learning_rate": 0.0001006371191135734,
+      "loss": 0.134,
+      "step": 12600
+    },
+    {
+      "epoch": 3.45,
+      "eval_loss": 0.07680243253707886,
+      "eval_runtime": 989.8703,
+      "eval_samples_per_second": 11.803,
+      "eval_steps_per_second": 1.476,
+      "eval_wer": 0.2718548167015373,
+      "step": 12600
+    },
+    {
+      "epoch": 3.45,
+      "learning_rate": 0.00010047091412742381,
+      "loss": 0.1714,
+      "step": 12610
+    },
+    {
+      "epoch": 3.46,
+      "learning_rate": 0.00010030470914127423,
+      "loss": 0.1599,
+      "step": 12620
+    },
+    {
+      "epoch": 3.46,
+      "learning_rate": 0.00010013850415512464,
+      "loss": 0.1292,
+      "step": 12630
+    },
+    {
+      "epoch": 3.46,
+      "learning_rate": 9.997229916897506e-05,
+      "loss": 0.1634,
+      "step": 12640
+    },
+    {
+      "epoch": 3.47,
+      "learning_rate": 9.980609418282546e-05,
+      "loss": 0.1341,
+      "step": 12650
+    },
+    {
+      "epoch": 3.47,
+      "learning_rate": 9.96398891966759e-05,
+      "loss": 0.1636,
+      "step": 12660
+    },
+    {
+      "epoch": 3.47,
+      "learning_rate": 9.94736842105263e-05,
+      "loss": 0.1418,
+      "step": 12670
+    },
+    {
+      "epoch": 3.47,
+      "learning_rate": 9.930747922437672e-05,
+      "loss": 0.1301,
+      "step": 12680
+    },
+    {
+      "epoch": 3.48,
+      "learning_rate": 9.914127423822713e-05,
+      "loss": 0.1679,
+      "step": 12690
+    },
+    {
+      "epoch": 3.48,
+      "learning_rate": 9.897506925207755e-05,
+      "loss": 0.1095,
+      "step": 12700
+    },
+    {
+      "epoch": 3.48,
+      "eval_loss": 0.07164399325847626,
+      "eval_runtime": 991.1101,
+      "eval_samples_per_second": 11.788,
+      "eval_steps_per_second": 1.474,
+      "eval_wer": 0.2641339943600473,
+      "step": 12700
+    },
+    {
+      "epoch": 3.48,
+      "learning_rate": 9.880886426592796e-05,
+      "loss": 0.1323,
+      "step": 12710
+    },
+    {
+      "epoch": 3.48,
+      "learning_rate": 9.864265927977839e-05,
+      "loss": 0.1354,
+      "step": 12720
+    },
+    {
+      "epoch": 3.49,
+      "learning_rate": 9.84764542936288e-05,
+      "loss": 0.1336,
+      "step": 12730
+    },
+    {
+      "epoch": 3.49,
+      "learning_rate": 9.831024930747921e-05,
+      "loss": 0.1182,
+      "step": 12740
+    },
+    {
+      "epoch": 3.49,
+      "learning_rate": 9.814404432132962e-05,
+      "loss": 0.1188,
+      "step": 12750
+    },
+    {
+      "epoch": 3.5,
+      "learning_rate": 9.797783933518005e-05,
+      "loss": 0.1438,
+      "step": 12760
+    },
+    {
+      "epoch": 3.5,
+      "learning_rate": 9.781163434903046e-05,
+      "loss": 0.1498,
+      "step": 12770
+    },
+    {
+      "epoch": 3.5,
+      "learning_rate": 9.764542936288088e-05,
+      "loss": 0.1255,
+      "step": 12780
+    },
+    {
+      "epoch": 3.5,
+      "learning_rate": 9.747922437673129e-05,
+      "loss": 0.1412,
+      "step": 12790
+    },
+    {
+      "epoch": 3.51,
+      "learning_rate": 9.73130193905817e-05,
+      "loss": 0.1185,
+      "step": 12800
+    },
+    {
+      "epoch": 3.51,
+      "eval_loss": 0.07018042355775833,
+      "eval_runtime": 991.2624,
+      "eval_samples_per_second": 11.786,
+      "eval_steps_per_second": 1.474,
+      "eval_wer": 0.2602906394978623,
+      "step": 12800
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 0.0001,
+      "loss": 10.4866,
+      "step": 12810
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 0.0002,
+      "loss": 4.7606,
+      "step": 12820
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 0.0003,
+      "loss": 3.1806,
+      "step": 12830
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 0.0004,
+      "loss": 3.1834,
+      "step": 12840
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 0.0005,
+      "loss": 3.1436,
+      "step": 12850
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 0.0006,
+      "loss": 3.0882,
+      "step": 12860
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 0.0007,
+      "loss": 2.8196,
+      "step": 12870
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 0.0008,
+      "loss": 1.7603,
+      "step": 12880
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 0.0009000000000000001,
+      "loss": 0.7548,
+      "step": 12890
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 0.001,
+      "loss": 0.6919,
+      "step": 12900
+    },
+    {
+      "epoch": 1.77,
+      "eval_loss": 0.5212433338165283,
+      "eval_runtime": 1284.8639,
+      "eval_samples_per_second": 9.093,
+      "eval_steps_per_second": 1.137,
+      "eval_wer": 0.7533089238606386,
+      "step": 12900
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 0.0011,
+      "loss": 0.5624,
+      "step": 12910
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 0.0012,
+      "loss": 0.5046,
+      "step": 12920
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 0.0013000000000000002,
+      "loss": 0.5652,
+      "step": 12930
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 0.0014,
+      "loss": 0.6445,
+      "step": 12940
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 0.0015,
+      "loss": 0.6018,
+      "step": 12950
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 0.0016,
+      "loss": 0.6265,
+      "step": 12960
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 0.0017,
+      "loss": 0.5558,
+      "step": 12970
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 0.0018000000000000002,
+      "loss": 0.5364,
+      "step": 12980
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 0.0019,
+      "loss": 0.6144,
+      "step": 12990
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 0.002,
+      "loss": 0.6742,
+      "step": 13000
+    },
+    {
+      "epoch": 1.78,
+      "eval_loss": 0.7235519886016846,
+      "eval_runtime": 1088.3191,
+      "eval_samples_per_second": 10.735,
+      "eval_steps_per_second": 1.342,
+      "eval_wer": 0.8849154007095424,
+      "step": 13000
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 0.0019994489599118336,
+      "loss": 0.7978,
+      "step": 13010
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 0.0019988979198236672,
+      "loss": 0.6329,
+      "step": 13020
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 0.001998346879735501,
+      "loss": 0.6059,
+      "step": 13030
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 0.0019977958396473344,
+      "loss": 0.6688,
+      "step": 13040
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 0.001997244799559168,
+      "loss": 0.7865,
+      "step": 13050
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 0.0019966937594710016,
+      "loss": 0.786,
+      "step": 13060
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 0.001996142719382835,
+      "loss": 0.6873,
+      "step": 13070
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 0.001995591679294669,
+      "loss": 0.644,
+      "step": 13080
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 0.0019950406392065024,
+      "loss": 0.6981,
+      "step": 13090
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 0.001994489599118336,
+      "loss": 0.7929,
+      "step": 13100
+    },
+    {
+      "epoch": 1.79,
+      "eval_loss": 0.6991682648658752,
+      "eval_runtime": 1082.258,
+      "eval_samples_per_second": 10.795,
+      "eval_steps_per_second": 1.35,
+      "eval_wer": 0.8898389884471937,
+      "step": 13100
+    },
+    {
+      "epoch": 1.8,
+      "learning_rate": 0.0019939385590301696,
+      "loss": 0.7901,
+      "step": 13110
+    },
+    {
+      "epoch": 1.8,
+      "learning_rate": 0.001993387518942003,
+      "loss": 0.6219,
+      "step": 13120
+    },
+    {
+      "epoch": 1.8,
+      "learning_rate": 0.001992836478853837,
+      "loss": 0.5593,
+      "step": 13130
+    },
+    {
+      "epoch": 1.8,
+      "learning_rate": 0.00199228543876567,
+      "loss": 0.6273,
+      "step": 13140
+    },
+    {
+      "epoch": 1.8,
+      "learning_rate": 0.001991734398677504,
+      "loss": 0.7598,
+      "step": 13150
+    },
+    {
+      "epoch": 1.8,
+      "learning_rate": 0.0019911833585893376,
+      "loss": 0.7298,
+      "step": 13160
+    },
+    {
+      "epoch": 1.8,
+      "learning_rate": 0.001990632318501171,
+      "loss": 0.5741,
+      "step": 13170
+    },
+    {
+      "epoch": 1.81,
+      "learning_rate": 0.0019900812784130048,
+      "loss": 0.5678,
+      "step": 13180
+    },
+    {
+      "epoch": 1.81,
+      "learning_rate": 0.001989530238324838,
+      "loss": 0.6603,
+      "step": 13190
+    },
+    {
+      "epoch": 1.81,
+      "learning_rate": 0.001988979198236672,
+      "loss": 0.6284,
+      "step": 13200
+    },
+    {
+      "epoch": 1.81,
+      "eval_loss": 0.6220380663871765,
+      "eval_runtime": 1072.8352,
+      "eval_samples_per_second": 10.89,
+      "eval_steps_per_second": 1.362,
+      "eval_wer": 0.8438324388247066,
+      "step": 13200
+    },
+    {
+      "epoch": 1.81,
+      "learning_rate": 0.001988428158148505,
+      "loss": 0.6718,
+      "step": 13210
+    },
+    {
+      "epoch": 1.81,
+      "learning_rate": 0.001987877118060339,
+      "loss": 0.5915,
+      "step": 13220
+    },
+    {
+      "epoch": 1.81,
+      "learning_rate": 0.0019873260779721723,
+      "loss": 0.5858,
+      "step": 13230
+    },
+    {
+      "epoch": 1.81,
+      "learning_rate": 0.0019867750378840063,
+      "loss": 0.5585,
+      "step": 13240
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 0.0019862239977958395,
+      "loss": 0.5648,
+      "step": 13250
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 0.001985672957707673,
+      "loss": 0.6787,
+      "step": 13260
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 0.0019851219176195067,
+      "loss": 0.5563,
+      "step": 13270
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 0.0019845708775313403,
+      "loss": 0.5354,
+      "step": 13280
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 0.001984019837443174,
+      "loss": 0.61,
+      "step": 13290
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 0.0019834687973550075,
+      "loss": 0.5259,
+      "step": 13300
+    },
+    {
+      "epoch": 1.82,
+      "eval_loss": 0.6680784225463867,
+      "eval_runtime": 1087.562,
+      "eval_samples_per_second": 10.742,
+      "eval_steps_per_second": 1.343,
+      "eval_wer": 0.8441508232511599,
+      "step": 13300
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 0.001982917757266841,
+      "loss": 0.6848,
+      "step": 13310
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 0.0019823667171786747,
+      "loss": 0.5889,
+      "step": 13320
+    },
+    {
+      "epoch": 1.83,
+      "learning_rate": 0.0019818156770905083,
+      "loss": 0.576,
+      "step": 13330
+    },
+    {
+      "epoch": 1.83,
+      "learning_rate": 0.001981264637002342,
+      "loss": 0.6211,
+      "step": 13340
+    },
+    {
+      "epoch": 1.83,
+      "learning_rate": 0.0019807135969141755,
+      "loss": 0.7772,
+      "step": 13350
+    },
+    {
+      "epoch": 1.83,
+      "learning_rate": 0.001980162556826009,
+      "loss": 0.779,
+      "step": 13360
+    },
+    {
+      "epoch": 1.83,
+      "learning_rate": 0.0019796115167378427,
+      "loss": 0.6357,
+      "step": 13370
+    },
+    {
+      "epoch": 1.83,
+      "learning_rate": 0.0019790604766496763,
+      "loss": 0.5716,
+      "step": 13380
+    },
+    {
+      "epoch": 1.83,
+      "learning_rate": 0.00197850943656151,
+      "loss": 0.5955,
+      "step": 13390
+    },
+    {
+      "epoch": 1.84,
+      "learning_rate": 0.0019779583964733435,
+      "loss": 0.6102,
+      "step": 13400
+    },
+    {
+      "epoch": 1.84,
+      "eval_loss": 0.6866852641105652,
+      "eval_runtime": 1072.1089,
+      "eval_samples_per_second": 10.897,
+      "eval_steps_per_second": 1.363,
+      "eval_wer": 0.897468843809697,
+      "step": 13400
+    },
+    {
+      "epoch": 1.84,
+      "learning_rate": 0.001977407356385177,
+      "loss": 0.6881,
+      "step": 13410
+    },
+    {
+      "epoch": 1.84,
+      "learning_rate": 0.0019768563162970106,
+      "loss": 0.6098,
+      "step": 13420
+    },
+    {
+      "epoch": 1.84,
+      "learning_rate": 0.0019763052762088442,
+      "loss": 0.5796,
+      "step": 13430
+    },
+    {
+      "epoch": 1.84,
+      "learning_rate": 0.001975754236120678,
+      "loss": 0.5692,
+      "step": 13440
+    },
+    {
+      "epoch": 1.84,
+      "learning_rate": 0.001975258300041328,
+      "loss": 0.7085,
+      "step": 13450
+    },
+    {
+      "epoch": 1.84,
+      "learning_rate": 0.001974707259953162,
+      "loss": 0.7,
+      "step": 13460
+    },
+    {
+      "epoch": 1.85,
+      "learning_rate": 0.001974156219864995,
+      "loss": 0.6483,
+      "step": 13470
+    },
+    {
+      "epoch": 1.85,
+      "learning_rate": 0.0019736051797768286,
+      "loss": 0.5868,
+      "step": 13480
+    },
+    {
+      "epoch": 1.85,
+      "learning_rate": 0.0019730541396886622,
+      "loss": 0.635,
+      "step": 13490
+    },
+    {
+      "epoch": 1.85,
+      "learning_rate": 0.001972503099600496,
+      "loss": 0.6609,
+      "step": 13500
+    },
+    {
+      "epoch": 1.85,
+      "eval_loss": 0.7215743660926819,
+      "eval_runtime": 1069.9974,
+      "eval_samples_per_second": 10.919,
+      "eval_steps_per_second": 1.365,
+      "eval_wer": 0.8681774765760029,
+      "step": 13500
+    },
+    {
+      "epoch": 1.85,
+      "learning_rate": 0.0019719520595123294,
+      "loss": 0.7052,
+      "step": 13510
+    },
+    {
+      "epoch": 1.85,
+      "learning_rate": 0.001971401019424163,
+      "loss": 0.646,
+      "step": 13520
+    },
+    {
+      "epoch": 1.85,
+      "learning_rate": 0.0019708499793359966,
+      "loss": 0.5597,
+      "step": 13530
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 0.0019702989392478302,
+      "loss": 0.5837,
+      "step": 13540
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 0.001969747899159664,
+      "loss": 0.633,
+      "step": 13550
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 0.0019691968590714974,
+      "loss": 0.7105,
+      "step": 13560
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 0.001968645818983331,
+      "loss": 0.5485,
+      "step": 13570
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 0.0019680947788951646,
+      "loss": 0.5213,
+      "step": 13580
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 0.001967543738806998,
+      "loss": 0.6221,
+      "step": 13590
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 0.001966992698718832,
+      "loss": 0.5471,
+      "step": 13600
+    },
+    {
+      "epoch": 1.86,
+      "eval_loss": 0.7448805570602417,
+      "eval_runtime": 1072.2916,
+      "eval_samples_per_second": 10.895,
+      "eval_steps_per_second": 1.363,
+      "eval_wer": 0.9033248430819613,
+      "step": 13600
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 0.0019664416586306654,
+      "loss": 0.6726,
+      "step": 13610
+    },
+    {
+      "epoch": 1.87,
+      "learning_rate": 0.001965890618542499,
+      "loss": 0.5851,
+      "step": 13620
+    },
+    {
+      "epoch": 1.87,
+      "learning_rate": 0.0019653395784543326,
+      "loss": 0.602,
+      "step": 13630
+    },
+    {
+      "epoch": 1.87,
+      "learning_rate": 0.001964788538366166,
+      "loss": 0.5757,
+      "step": 13640
+    },
+    {
+      "epoch": 1.87,
+      "learning_rate": 0.0019642374982779998,
+      "loss": 0.5881,
+      "step": 13650
+    },
+    {
+      "epoch": 1.87,
+      "learning_rate": 0.0019636864581898334,
+      "loss": 0.6891,
+      "step": 13660
+    },
+    {
+      "epoch": 1.87,
+      "learning_rate": 0.001963135418101667,
+      "loss": 0.5304,
+      "step": 13670
+    },
+    {
+      "epoch": 1.87,
+      "learning_rate": 0.0019625843780135006,
+      "loss": 0.5125,
+      "step": 13680
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 0.001962033337925334,
+      "loss": 0.5705,
+      "step": 13690
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 0.0019614822978371678,
+      "loss": 0.5535,
+      "step": 13700
+    },
+    {
+      "epoch": 1.88,
+      "eval_loss": 0.5763468742370605,
+      "eval_runtime": 1071.5287,
+      "eval_samples_per_second": 10.903,
+      "eval_steps_per_second": 1.363,
+      "eval_wer": 0.8305057764031657,
+      "step": 13700
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 0.0019609312577490014,
+      "loss": 0.6177,
+      "step": 13710
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 0.001960380217660835,
+      "loss": 0.564,
+      "step": 13720
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 0.0019598291775726685,
+      "loss": 0.5064,
+      "step": 13730
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 0.001959278137484502,
+      "loss": 0.5962,
+      "step": 13740
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 0.0019587270973963357,
+      "loss": 0.616,
+      "step": 13750
+    },
+    {
+      "epoch": 1.89,
+      "learning_rate": 0.0019581760573081693,
+      "loss": 0.6316,
+      "step": 13760
+    },
+    {
+      "epoch": 1.89,
+      "learning_rate": 0.001957625017220003,
+      "loss": 0.6386,
+      "step": 13770
+    },
+    {
+      "epoch": 1.89,
+      "learning_rate": 0.0019570739771318365,
+      "loss": 0.5872,
+      "step": 13780
+    },
+    {
+      "epoch": 1.89,
+      "learning_rate": 0.0019565229370436697,
+      "loss": 0.5492,
+      "step": 13790
+    },
+    {
+      "epoch": 1.89,
+      "learning_rate": 0.0019559718969555037,
+      "loss": 0.5933,
+      "step": 13800
+    },
+    {
+      "epoch": 1.89,
+      "eval_loss": 0.6062564253807068,
+      "eval_runtime": 1071.465,
+      "eval_samples_per_second": 10.904,
+      "eval_steps_per_second": 1.364,
+      "eval_wer": 0.8677453834258164,
+      "step": 13800
+    },
+    {
+      "epoch": 1.89,
+      "learning_rate": 0.001955420856867337,
+      "loss": 0.7188,
+      "step": 13810
+    },
+    {
+      "epoch": 1.89,
+      "learning_rate": 0.001954869816779171,
+      "loss": 0.6475,
+      "step": 13820
+    },
+    {
+      "epoch": 1.89,
+      "learning_rate": 0.001954318776691004,
+      "loss": 0.554,
+      "step": 13830
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 0.001953767736602838,
+      "loss": 0.6043,
+      "step": 13840
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 0.0019532166965146713,
+      "loss": 0.5779,
+      "step": 13850
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 0.001952665656426505,
+      "loss": 0.6341,
+      "step": 13860
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 0.0019521146163383387,
+      "loss": 0.5432,
+      "step": 13870
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 0.0019515635762501723,
+      "loss": 0.5702,
+      "step": 13880
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 0.0019510125361620059,
+      "loss": 0.6163,
+      "step": 13890
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 0.0019504614960738395,
+      "loss": 0.6006,
+      "step": 13900
+    },
+    {
+      "epoch": 1.9,
+      "eval_loss": 0.547044575214386,
+      "eval_runtime": 1069.3312,
+      "eval_samples_per_second": 10.926,
+      "eval_steps_per_second": 1.366,
+      "eval_wer": 0.8033180205585373,
+      "step": 13900
+    },
+    {
+      "epoch": 1.91,
+      "learning_rate": 0.001949910455985673,
+      "loss": 0.6424,
+      "step": 13910
+    },
+    {
+      "epoch": 1.91,
+      "learning_rate": 0.0019493594158975064,
+      "loss": 0.5653,
+      "step": 13920
+    },
+    {
+      "epoch": 1.91,
+      "learning_rate": 0.0019488083758093403,
+      "loss": 0.6309,
+      "step": 13930
+    },
+    {
+      "epoch": 1.91,
+      "learning_rate": 0.0019482573357211736,
+      "loss": 0.5666,
+      "step": 13940
+    },
+    {
+      "epoch": 1.91,
+      "learning_rate": 0.0019477062956330074,
+      "loss": 0.5931,
+      "step": 13950
+    },
+    {
+      "epoch": 1.91,
+      "learning_rate": 0.0019471552555448408,
+      "loss": 0.6474,
+      "step": 13960
+    },
+    {
+      "epoch": 1.91,
+      "learning_rate": 0.0019466042154566746,
+      "loss": 0.5782,
+      "step": 13970
+    },
+    {
+      "epoch": 1.92,
+      "learning_rate": 0.001946053175368508,
+      "loss": 0.5371,
+      "step": 13980
+    },
+    {
+      "epoch": 1.92,
+      "learning_rate": 0.0019455021352803416,
+      "loss": 0.5849,
+      "step": 13990
+    },
+    {
+      "epoch": 1.92,
+      "learning_rate": 0.0019449510951921752,
+      "loss": 0.6245,
+      "step": 14000
+    },
+    {
+      "epoch": 1.92,
+      "eval_loss": 0.6192805767059326,
+      "eval_runtime": 1070.5576,
+      "eval_samples_per_second": 10.913,
+      "eval_steps_per_second": 1.365,
+      "eval_wer": 0.8423087419266806,
+      "step": 14000
+    },
+    {
+      "epoch": 1.92,
+      "learning_rate": 0.0019444000551040088,
+      "loss": 0.648,
+      "step": 14010
+    },
+    {
+      "epoch": 1.92,
+      "learning_rate": 0.0019438490150158426,
+      "loss": 0.596,
+      "step": 14020
+    },
+    {
+      "epoch": 1.92,
+      "learning_rate": 0.001943297974927676,
+      "loss": 0.5851,
+      "step": 14030
+    },
+    {
+      "epoch": 1.92,
+      "learning_rate": 0.0019427469348395098,
+      "loss": 0.5929,
+      "step": 14040
+    },
+    {
+      "epoch": 1.92,
+      "learning_rate": 0.0019421958947513432,
+      "loss": 0.5405,
+      "step": 14050
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 0.0019416448546631768,
+      "loss": 0.5958,
+      "step": 14060
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 0.0019410938145750104,
+      "loss": 0.5689,
+      "step": 14070
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 0.001940542774486844,
+      "loss": 0.5568,
+      "step": 14080
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 0.0019399917343986776,
+      "loss": 0.5959,
+      "step": 14090
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 0.0019394406943105112,
+      "loss": 0.469,
+      "step": 14100
+    },
+    {
+      "epoch": 1.93,
+      "eval_loss": 0.5840388536453247,
+      "eval_runtime": 1073.4287,
+      "eval_samples_per_second": 10.884,
+      "eval_steps_per_second": 1.361,
+      "eval_wer": 0.8291981260802329,
+      "step": 14100
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 0.0019388896542223445,
+      "loss": 0.6458,
+      "step": 14110
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 0.0019383386141341784,
+      "loss": 0.5768,
+      "step": 14120
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 0.0019377875740460117,
+      "loss": 0.5447,
+      "step": 14130
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 0.0019372365339578456,
+      "loss": 0.5689,
+      "step": 14140
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 0.001936685493869679,
+      "loss": 0.622,
+      "step": 14150
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 0.0019361344537815127,
+      "loss": 0.6431,
+      "step": 14160
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 0.0019355834136933463,
+      "loss": 0.5689,
+      "step": 14170
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 0.0019350323736051797,
+      "loss": 0.6116,
+      "step": 14180
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 0.0019344813335170135,
+      "loss": 0.5665,
+      "step": 14190
+    },
+    {
+      "epoch": 1.95,
+      "learning_rate": 0.001933930293428847,
+      "loss": 0.539,
+      "step": 14200
+    },
+    {
+      "epoch": 1.95,
+      "eval_loss": 0.5749104619026184,
+      "eval_runtime": 1071.5765,
+      "eval_samples_per_second": 10.903,
+      "eval_steps_per_second": 1.363,
+      "eval_wer": 0.8210565814609296,
+      "step": 14200
+    },
+    {
+      "epoch": 1.95,
+      "learning_rate": 0.0019333792533406807,
+      "loss": 0.612,
+      "step": 14210
+    },
+    {
+      "epoch": 1.95,
+      "learning_rate": 0.001932828213252514,
+      "loss": 0.5716,
+      "step": 14220
+    },
+    {
+      "epoch": 1.95,
+      "learning_rate": 0.001932277173164348,
+      "loss": 0.5422,
+      "step": 14230
+    },
+    {
+      "epoch": 1.95,
+      "learning_rate": 0.0019317261330761813,
+      "loss": 0.5669,
+      "step": 14240
+    },
+    {
+      "epoch": 1.95,
+      "learning_rate": 0.001931175092988015,
+      "loss": 0.6551,
+      "step": 14250
+    },
+    {
+      "epoch": 1.95,
+      "learning_rate": 0.0019306240528998485,
+      "loss": 0.6419,
+      "step": 14260
+    },
+    {
+      "epoch": 1.96,
+      "learning_rate": 0.001930073012811682,
+      "loss": 0.6037,
+      "step": 14270
+    },
+    {
+      "epoch": 1.96,
+      "learning_rate": 0.0019295219727235157,
+      "loss": 0.5171,
+      "step": 14280
+    },
+    {
+      "epoch": 1.96,
+      "learning_rate": 0.0019289709326353493,
+      "loss": 0.6136,
+      "step": 14290
+    },
+    {
+      "epoch": 1.96,
+      "learning_rate": 0.0019284198925471827,
+      "loss": 0.6109,
+      "step": 14300
+    },
+    {
+      "epoch": 1.96,
+      "eval_loss": 0.6034438610076904,
+      "eval_runtime": 1072.0017,
+      "eval_samples_per_second": 10.898,
+      "eval_steps_per_second": 1.363,
+      "eval_wer": 0.8474711179841717,
+      "step": 14300
+    },
+    {
+      "epoch": 1.96,
+      "learning_rate": 0.0019278688524590165,
+      "loss": 0.6214,
+      "step": 14310
+    },
+    {
+      "epoch": 1.96,
+      "learning_rate": 0.0019273178123708498,
+      "loss": 0.5964,
+      "step": 14320
+    },
+    {
+      "epoch": 1.96,
+      "learning_rate": 0.0019267667722826837,
+      "loss": 0.6015,
+      "step": 14330
+    },
+    {
+      "epoch": 1.96,
+      "learning_rate": 0.0019262157321945173,
+      "loss": 0.5969,
+      "step": 14340
+    },
+    {
+      "epoch": 1.97,
+      "learning_rate": 0.0019256646921063509,
+      "loss": 0.6378,
+      "step": 14350
+    },
+    {
+      "epoch": 1.97,
+      "learning_rate": 0.0019251136520181845,
+      "loss": 0.6196,
+      "step": 14360
+    },
+    {
+      "epoch": 1.97,
+      "learning_rate": 0.0019245626119300178,
+      "loss": 0.5844,
+      "step": 14370
+    },
+    {
+      "epoch": 1.97,
+      "learning_rate": 0.0019240115718418516,
+      "loss": 0.5747,
+      "step": 14380
+    },
+    {
+      "epoch": 1.97,
+      "learning_rate": 0.001923460531753685,
+      "loss": 0.5918,
+      "step": 14390
+    },
+    {
+      "epoch": 1.97,
+      "learning_rate": 0.0019229094916655188,
+      "loss": 0.4675,
+      "step": 14400
+    },
+    {
+      "epoch": 1.97,
+      "eval_loss": 0.5792332291603088,
+      "eval_runtime": 1073.9201,
+      "eval_samples_per_second": 10.879,
+      "eval_steps_per_second": 1.36,
+      "eval_wer": 0.7990539434185391,
+      "step": 14400
+    },
+    {
+      "epoch": 1.97,
+      "learning_rate": 0.0019223584515773522,
+      "loss": 0.6125,
+      "step": 14410
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 0.001921807411489186,
+      "loss": 0.6493,
+      "step": 14420
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 0.0019212563714010194,
+      "loss": 0.6108,
+      "step": 14430
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 0.001920705331312853,
+      "loss": 0.5471,
+      "step": 14440
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 0.0019201542912246866,
+      "loss": 0.6279,
+      "step": 14450
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 0.0019196032511365202,
+      "loss": 0.6306,
+      "step": 14460
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 0.0019190522110483538,
+      "loss": 0.6038,
+      "step": 14470
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 0.0019185011709601874,
+      "loss": 0.5171,
+      "step": 14480
+    },
+    {
+      "epoch": 1.99,
+      "learning_rate": 0.001917950130872021,
+      "loss": 0.5384,
+      "step": 14490
+    },
+    {
+      "epoch": 1.99,
+      "learning_rate": 0.0019173990907838546,
+      "loss": 0.4737,
+      "step": 14500
+    },
+    {
+      "epoch": 1.99,
+      "eval_loss": 0.5673007965087891,
+      "eval_runtime": 1070.9535,
+      "eval_samples_per_second": 10.909,
+      "eval_steps_per_second": 1.364,
+      "eval_wer": 0.8333030110070044,
+      "step": 14500
+    },
+    {
+      "epoch": 1.99,
+      "learning_rate": 0.0019168480506956882,
+      "loss": 0.6589,
+      "step": 14510
+    },
+    {
+      "epoch": 1.99,
+      "learning_rate": 0.0019162970106075218,
+      "loss": 0.5214,
+      "step": 14520
+    },
+    {
+      "epoch": 1.99,
+      "learning_rate": 0.0019157459705193554,
+      "loss": 0.5929,
+      "step": 14530
+    },
+    {
+      "epoch": 1.99,
+      "learning_rate": 0.001915194930431189,
+      "loss": 0.5388,
+      "step": 14540
+    },
+    {
+      "epoch": 1.99,
+      "learning_rate": 0.0019146438903430226,
+      "loss": 0.546,
+      "step": 14550
+    },
+    {
+      "epoch": 1.99,
+      "learning_rate": 0.001914092850254856,
+      "loss": 0.5886,
+      "step": 14560
+    },
+    {
+      "epoch": 2.0,
+      "learning_rate": 0.0019135418101666898,
+      "loss": 0.4824,
+      "step": 14570
+    },
+    {
+      "epoch": 2.0,
+      "learning_rate": 0.0019129907700785231,
+      "loss": 0.5295,
+      "step": 14580
+    },
+    {
+      "epoch": 2.0,
+      "learning_rate": 0.001912439729990357,
+      "loss": 0.5198,
+      "step": 14590
+    },
+    {
+      "epoch": 2.0,
+      "learning_rate": 0.0019118886899021903,
+      "loss": 0.6289,
+      "step": 14600
+    },
+    {
+      "epoch": 2.0,
+      "eval_loss": 0.5210581421852112,
+      "eval_runtime": 1072.7032,
+      "eval_samples_per_second": 10.891,
+      "eval_steps_per_second": 1.362,
+      "eval_wer": 0.8101064313654144,
+      "step": 14600
+    },
+    {
+      "epoch": 2.0,
+      "learning_rate": 0.001911337649814024,
+      "loss": 0.578,
+      "step": 14610
+    },
+    {
+      "epoch": 2.0,
+      "learning_rate": 0.0019107866097258575,
+      "loss": 0.5589,
+      "step": 14620
+    },
+    {
+      "epoch": 2.0,
+      "learning_rate": 0.0019102355696376911,
+      "loss": 0.5232,
+      "step": 14630
+    },
+    {
+      "epoch": 2.01,
+      "learning_rate": 0.001909684529549525,
+      "loss": 0.5325,
+      "step": 14640
+    },
+    {
+      "epoch": 2.01,
+      "learning_rate": 0.0019091334894613583,
+      "loss": 0.4904,
+      "step": 14650
+    },
+    {
+      "epoch": 2.01,
+      "learning_rate": 0.0019085824493731921,
+      "loss": 0.5774,
+      "step": 14660
+    },
+    {
+      "epoch": 2.01,
+      "learning_rate": 0.0019080314092850255,
+      "loss": 0.4947,
+      "step": 14670
+    },
+    {
+      "epoch": 2.01,
+      "learning_rate": 0.001907480369196859,
+      "loss": 0.5071,
+      "step": 14680
+    },
+    {
+      "epoch": 2.01,
+      "learning_rate": 0.0019069293291086927,
+      "loss": 0.4889,
+      "step": 14690
+    },
+    {
+      "epoch": 2.01,
+      "learning_rate": 0.0019063782890205263,
+      "loss": 0.5712,
+      "step": 14700
+    },
+    {
+      "epoch": 2.01,
+      "eval_loss": 0.49998703598976135,
+      "eval_runtime": 1073.8232,
+      "eval_samples_per_second": 10.88,
+      "eval_steps_per_second": 1.361,
+      "eval_wer": 0.7944828527244611,
+      "step": 14700
+    },
+    {
+      "epoch": 2.02,
+      "learning_rate": 0.0019058272489323599,
+      "loss": 0.5785,
+      "step": 14710
+    },
+    {
+      "epoch": 2.02,
+      "learning_rate": 0.0019052762088441935,
+      "loss": 0.5249,
+      "step": 14720
+    },
+    {
+      "epoch": 2.02,
+      "learning_rate": 0.001904725168756027,
+      "loss": 0.5543,
+      "step": 14730
+    },
+    {
+      "epoch": 2.02,
+      "learning_rate": 0.0019041741286678607,
+      "loss": 0.5281,
+      "step": 14740
+    },
+    {
+      "epoch": 2.02,
+      "learning_rate": 0.001903623088579694,
+      "loss": 0.5796,
+      "step": 14750
+    },
+    {
+      "epoch": 2.02,
+      "learning_rate": 0.0019030720484915279,
+      "loss": 0.6029,
+      "step": 14760
+    },
+    {
+      "epoch": 2.02,
+      "learning_rate": 0.0019025210084033612,
+      "loss": 0.547,
+      "step": 14770
+    },
+    {
+      "epoch": 2.02,
+      "learning_rate": 0.001901969968315195,
+      "loss": 0.5039,
+      "step": 14780
+    },
+    {
+      "epoch": 2.03,
+      "learning_rate": 0.0019014189282270286,
+      "loss": 0.5635,
+      "step": 14790
+    },
+    {
+      "epoch": 2.03,
+      "learning_rate": 0.0019009229921476787,
+      "loss": 0.9085,
+      "step": 14800
+    },
+    {
+      "epoch": 2.03,
+      "eval_loss": 0.5572303533554077,
+      "eval_runtime": 1072.1665,
+      "eval_samples_per_second": 10.897,
+      "eval_steps_per_second": 1.363,
+      "eval_wer": 0.7912990084599291,
+      "step": 14800
+    },
+    {
+      "epoch": 2.03,
+      "learning_rate": 0.0019003719520595125,
+      "loss": 0.6257,
+      "step": 14810
+    },
+    {
+      "epoch": 2.03,
+      "learning_rate": 0.0018998209119713459,
+      "loss": 0.5225,
+      "step": 14820
+    },
+    {
+      "epoch": 2.03,
+      "learning_rate": 0.0018992698718831797,
+      "loss": 0.5498,
+      "step": 14830
+    },
+    {
+      "epoch": 2.03,
+      "learning_rate": 0.001898718831795013,
+      "loss": 0.5065,
+      "step": 14840
+    },
+    {
+      "epoch": 2.03,
+      "learning_rate": 0.0018981677917068466,
+      "loss": 0.5546,
+      "step": 14850
+    },
+    {
+      "epoch": 2.04,
+      "learning_rate": 0.0018976167516186802,
+      "loss": 0.6144,
+      "step": 14860
+    },
+    {
+      "epoch": 2.04,
+      "learning_rate": 0.0018970657115305138,
+      "loss": 0.5732,
+      "step": 14870
+    },
+    {
+      "epoch": 2.04,
+      "learning_rate": 0.0018965146714423477,
+      "loss": 0.544,
+      "step": 14880
+    },
+    {
+      "epoch": 2.04,
+      "learning_rate": 0.001895963631354181,
+      "loss": 0.5961,
+      "step": 14890
+    },
+    {
+      "epoch": 2.04,
+      "learning_rate": 0.0018954125912660146,
+      "loss": 0.5669,
+      "step": 14900
+    },
+    {
+      "epoch": 2.04,
+      "eval_loss": 0.517475426197052,
+      "eval_runtime": 1072.4392,
+      "eval_samples_per_second": 10.894,
+      "eval_steps_per_second": 1.362,
+      "eval_wer": 0.7902983716910761,
+      "step": 14900
+    },
+    {
+      "epoch": 2.04,
+      "learning_rate": 0.0018948615511778482,
+      "loss": 0.9665,
+      "step": 14910
+    },
+    {
+      "epoch": 2.04,
+      "learning_rate": 0.0018943105110896818,
+      "loss": 0.5432,
+      "step": 14920
+    },
+    {
+      "epoch": 2.05,
+      "learning_rate": 0.0018937594710015154,
+      "loss": 0.5726,
+      "step": 14930
+    },
+    {
+      "epoch": 2.05,
+      "learning_rate": 0.001893208430913349,
+      "loss": 0.5485,
+      "step": 14940
+    },
+    {
+      "epoch": 2.05,
+      "learning_rate": 0.0018926573908251826,
+      "loss": 0.5202,
+      "step": 14950
+    },
+    {
+      "epoch": 2.05,
+      "learning_rate": 0.0018921063507370162,
+      "loss": 0.5959,
+      "step": 14960
+    },
+    {
+      "epoch": 2.05,
+      "learning_rate": 0.0018915553106488496,
+      "loss": 0.5994,
+      "step": 14970
+    },
+    {
+      "epoch": 2.05,
+      "learning_rate": 0.0018910042705606834,
+      "loss": 0.5424,
+      "step": 14980
+    },
+    {
+      "epoch": 2.05,
+      "learning_rate": 0.0018904532304725168,
+      "loss": 0.5271,
+      "step": 14990
+    },
+    {
+      "epoch": 2.06,
+      "learning_rate": 0.0018899021903843506,
+      "loss": 0.4907,
+      "step": 15000
+    },
+    {
+      "epoch": 2.06,
+      "eval_loss": 0.540747344493866,
+      "eval_runtime": 1074.2402,
+      "eval_samples_per_second": 10.876,
+      "eval_steps_per_second": 1.36,
+      "eval_wer": 0.7993950695897389,
+      "step": 15000
+    },
+    {
+      "epoch": 2.06,
+      "learning_rate": 0.001889351150296184,
+      "loss": 0.5953,
+      "step": 15010
+    },
+    {
+      "epoch": 2.06,
+      "learning_rate": 0.0018888001102080178,
+      "loss": 0.5417,
+      "step": 15020
+    },
+    {
+      "epoch": 2.06,
+      "learning_rate": 0.0018882490701198514,
+      "loss": 0.5226,
+      "step": 15030
+    },
+    {
+      "epoch": 2.06,
+      "learning_rate": 0.0018876980300316848,
+      "loss": 0.5297,
+      "step": 15040
+    },
+    {
+      "epoch": 2.06,
+      "learning_rate": 0.0018871469899435186,
+      "loss": 0.5586,
+      "step": 15050
+    },
+    {
+      "epoch": 2.06,
+      "learning_rate": 0.001886595949855352,
+      "loss": 0.5749,
+      "step": 15060
+    },
+    {
+      "epoch": 2.06,
+      "learning_rate": 0.0018860449097671858,
+      "loss": 0.5887,
+      "step": 15070
+    },
+    {
+      "epoch": 2.07,
+      "learning_rate": 0.0018854938696790191,
+      "loss": 0.5209,
+      "step": 15080
+    },
+    {
+      "epoch": 2.07,
+      "learning_rate": 0.0018849428295908527,
+      "loss": 0.5515,
+      "step": 15090
+    },
+    {
+      "epoch": 2.07,
+      "learning_rate": 0.0018843917895026863,
+      "loss": 0.6087,
+      "step": 15100
+    },
+    {
+      "epoch": 2.07,
+      "eval_loss": 0.5597626566886902,
+      "eval_runtime": 1074.3814,
+      "eval_samples_per_second": 10.874,
+      "eval_steps_per_second": 1.36,
+      "eval_wer": 0.8134380969707996,
+      "step": 15100
+    },
+    {
+      "epoch": 2.07,
+      "learning_rate": 0.00188384074941452,
+      "loss": 0.5845,
+      "step": 15110
+    },
+    {
+      "epoch": 2.07,
+      "learning_rate": 0.0018832897093263535,
+      "loss": 0.5546,
+      "step": 15120
+    },
+    {
+      "epoch": 2.07,
+      "learning_rate": 0.0018827386692381871,
+      "loss": 0.4896,
+      "step": 15130
+    },
+    {
+      "epoch": 2.07,
+      "learning_rate": 0.0018821876291500207,
+      "loss": 0.569,
+      "step": 15140
+    },
+    {
+      "epoch": 2.08,
+      "learning_rate": 0.0018816365890618543,
+      "loss": 0.6535,
+      "step": 15150
+    },
+    {
+      "epoch": 2.08,
+      "learning_rate": 0.0018810855489736877,
+      "loss": 0.6044,
+      "step": 15160
+    },
+    {
+      "epoch": 2.08,
+      "learning_rate": 0.0018805345088855215,
+      "loss": 0.5353,
+      "step": 15170
+    },
+    {
+      "epoch": 2.08,
+      "learning_rate": 0.001879983468797355,
+      "loss": 0.4729,
+      "step": 15180
+    },
+    {
+      "epoch": 2.08,
+      "learning_rate": 0.0018794324287091887,
+      "loss": 0.584,
+      "step": 15190
+    },
+    {
+      "epoch": 2.08,
+      "learning_rate": 0.0018788813886210223,
+      "loss": 0.6574,
+      "step": 15200
+    },
+    {
+      "epoch": 2.08,
+      "eval_loss": 0.5471270084381104,
+      "eval_runtime": 1074.0657,
+      "eval_samples_per_second": 10.877,
+      "eval_steps_per_second": 1.36,
+      "eval_wer": 0.8014190848721914,
+      "step": 15200
+    },
+    {
+      "epoch": 2.08,
+      "learning_rate": 0.0018783303485328557,
+      "loss": 0.6116,
+      "step": 15210
+    },
+    {
+      "epoch": 2.09,
+      "learning_rate": 0.0018777793084446895,
+      "loss": 0.5345,
+      "step": 15220
+    },
+    {
+      "epoch": 2.09,
+      "learning_rate": 0.0018772282683565229,
+      "loss": 0.5272,
+      "step": 15230
+    },
+    {
+      "epoch": 2.09,
+      "learning_rate": 0.0018766772282683567,
+      "loss": 0.5587,
+      "step": 15240
+    },
+    {
+      "epoch": 2.09,
+      "learning_rate": 0.00187612618818019,
+      "loss": 0.5599,
+      "step": 15250
+    },
+    {
+      "epoch": 2.09,
+      "learning_rate": 0.0018755751480920239,
+      "loss": 0.5387,
+      "step": 15260
+    },
+    {
+      "epoch": 2.09,
+      "learning_rate": 0.0018750241080038573,
+      "loss": 0.51,
+      "step": 15270
+    },
+    {
+      "epoch": 2.09,
+      "learning_rate": 0.0018744730679156908,
+      "loss": 0.5018,
+      "step": 15280
+    },
+    {
+      "epoch": 2.09,
+      "learning_rate": 0.0018739220278275244,
+      "loss": 0.508,
+      "step": 15290
+    },
+    {
+      "epoch": 2.1,
+      "learning_rate": 0.001873370987739358,
+      "loss": 0.5082,
+      "step": 15300
+    },
+    {
+      "epoch": 2.1,
+      "eval_loss": 0.5032139420509338,
+      "eval_runtime": 1072.7091,
+      "eval_samples_per_second": 10.891,
+      "eval_steps_per_second": 1.362,
+      "eval_wer": 0.7929705266988083,
+      "step": 15300
+    },
+    {
+      "epoch": 2.1,
+      "learning_rate": 0.0018728199476511916,
+      "loss": 0.6229,
+      "step": 15310
+    },
+    {
+      "epoch": 2.1,
+      "learning_rate": 0.0018722689075630252,
+      "loss": 0.5258,
+      "step": 15320
+    },
+    {
+      "epoch": 2.1,
+      "learning_rate": 0.0018717178674748588,
+      "loss": 0.5071,
+      "step": 15330
+    },
+    {
+      "epoch": 2.1,
+      "learning_rate": 0.0018711668273866924,
+      "loss": 0.501,
+      "step": 15340
+    },
+    {
+      "epoch": 2.1,
+      "learning_rate": 0.001870615787298526,
+      "loss": 0.5319,
+      "step": 15350
+    },
+    {
+      "epoch": 2.1,
+      "learning_rate": 0.0018700647472103596,
+      "loss": 0.5681,
+      "step": 15360
+    },
+    {
+      "epoch": 2.11,
+      "learning_rate": 0.0018695137071221932,
+      "loss": 0.5663,
+      "step": 15370
+    },
+    {
+      "epoch": 2.11,
+      "learning_rate": 0.0018689626670340268,
+      "loss": 0.5312,
+      "step": 15380
+    },
+    {
+      "epoch": 2.11,
+      "learning_rate": 0.0018684116269458604,
+      "loss": 0.5784,
+      "step": 15390
+    },
+    {
+      "epoch": 2.11,
+      "learning_rate": 0.0018678605868576938,
+      "loss": 0.5103,
+      "step": 15400
+    },
+    {
+      "epoch": 2.11,
+      "eval_loss": 0.5066972374916077,
+      "eval_runtime": 1071.7105,
+      "eval_samples_per_second": 10.901,
+      "eval_steps_per_second": 1.363,
+      "eval_wer": 0.7894114436459565,
+      "step": 15400
+    },
+    {
+      "epoch": 2.11,
+      "learning_rate": 0.0018673095467695276,
+      "loss": 0.605,
+      "step": 15410
+    },
+    {
+      "epoch": 2.11,
+      "learning_rate": 0.001866758506681361,
+      "loss": 0.8112,
+      "step": 15420
+    },
+    {
+      "epoch": 2.11,
+      "learning_rate": 0.0018662074665931948,
+      "loss": 0.5266,
+      "step": 15430
+    },
+    {
+      "epoch": 2.12,
+      "learning_rate": 0.0018656564265050282,
+      "loss": 0.5116,
+      "step": 15440
+    },
+    {
+      "epoch": 2.12,
+      "learning_rate": 0.001865105386416862,
+      "loss": 0.4441,
+      "step": 15450
+    },
+    {
+      "epoch": 2.12,
+      "learning_rate": 0.0018645543463286954,
+      "loss": 0.7495,
+      "step": 15460
+    },
+    {
+      "epoch": 2.12,
+      "learning_rate": 0.001864003306240529,
+      "loss": 0.5345,
+      "step": 15470
+    },
+    {
+      "epoch": 2.12,
+      "learning_rate": 0.0018634522661523626,
+      "loss": 0.5471,
+      "step": 15480
+    },
+    {
+      "epoch": 2.12,
+      "learning_rate": 0.0018629012260641961,
+      "loss": 0.5299,
+      "step": 15490
+    },
+    {
+      "epoch": 2.12,
+      "learning_rate": 0.00186235018597603,
+      "loss": 0.4954,
+      "step": 15500
+    },
+    {
+      "epoch": 2.12,
+      "eval_loss": 0.5026968717575073,
+      "eval_runtime": 1073.1339,
+      "eval_samples_per_second": 10.887,
+      "eval_steps_per_second": 1.361,
+      "eval_wer": 0.7715819157645775,
+      "step": 15500
+    },
+    {
+      "epoch": 2.12,
+      "learning_rate": 0.0018617991458878633,
+      "loss": 0.5507,
+      "step": 15510
+    },
+    {
+      "epoch": 2.13,
+      "learning_rate": 0.0018612481057996972,
+      "loss": 0.5052,
+      "step": 15520
+    },
+    {
+      "epoch": 2.13,
+      "learning_rate": 0.0018606970657115305,
+      "loss": 0.5216,
+      "step": 15530
+    },
+    {
+      "epoch": 2.13,
+      "learning_rate": 0.0018601460256233641,
+      "loss": 0.5266,
+      "step": 15540
+    },
+    {
+      "epoch": 2.13,
+      "learning_rate": 0.0018595949855351977,
+      "loss": 0.5254,
+      "step": 15550
+    },
+    {
+      "epoch": 2.13,
+      "learning_rate": 0.0018590439454470313,
+      "loss": 0.6192,
+      "step": 15560
+    },
+    {
+      "epoch": 2.13,
+      "learning_rate": 0.001858492905358865,
+      "loss": 0.5433,
+      "step": 15570
+    },
+    {
+      "epoch": 2.13,
+      "learning_rate": 0.0018579418652706985,
+      "loss": 0.5089,
+      "step": 15580
+    },
+    {
+      "epoch": 2.14,
+      "learning_rate": 0.001857390825182532,
+      "loss": 0.503,
+      "step": 15590
+    },
+    {
+      "epoch": 2.14,
+      "learning_rate": 0.0018568397850943657,
+      "loss": 0.5438,
+      "step": 15600
+    },
+    {
+      "epoch": 2.14,
+      "eval_loss": 0.5089832544326782,
+      "eval_runtime": 1073.31,
+      "eval_samples_per_second": 10.885,
+      "eval_steps_per_second": 1.361,
+      "eval_wer": 0.7783475848267079,
+      "step": 15600
+    },
+    {
+      "epoch": 2.14,
+      "learning_rate": 0.001856288745006199,
+      "loss": 0.5914,
+      "step": 15610
+    },
+    {
+      "epoch": 2.14,
+      "learning_rate": 0.001855737704918033,
+      "loss": 0.5706,
+      "step": 15620
+    },
+    {
+      "epoch": 2.14,
+      "learning_rate": 0.0018551866648298663,
+      "loss": 0.5428,
+      "step": 15630
+    },
+    {
+      "epoch": 2.14,
+      "learning_rate": 0.0018546356247417,
+      "loss": 0.5178,
+      "step": 15640
+    },
+    {
+      "epoch": 2.14,
+      "learning_rate": 0.0018540845846535337,
+      "loss": 0.538,
+      "step": 15650
+    },
+    {
+      "epoch": 2.15,
+      "learning_rate": 0.001853533544565367,
+      "loss": 0.579,
+      "step": 15660
+    },
+    {
+      "epoch": 2.15,
+      "learning_rate": 0.0018529825044772009,
+      "loss": 0.4937,
+      "step": 15670
+    },
+    {
+      "epoch": 2.15,
+      "learning_rate": 0.0018524314643890343,
+      "loss": 0.5198,
+      "step": 15680
+    },
+    {
+      "epoch": 2.15,
+      "learning_rate": 0.001851880424300868,
+      "loss": 0.4821,
+      "step": 15690
+    },
+    {
+      "epoch": 2.15,
+      "learning_rate": 0.0018513293842127014,
+      "loss": 0.4469,
+      "step": 15700
+    },
+    {
+      "epoch": 2.15,
+      "eval_loss": 0.4352567791938782,
+      "eval_runtime": 1075.4658,
+      "eval_samples_per_second": 10.863,
+      "eval_steps_per_second": 1.358,
+      "eval_wer": 0.7355476212134995,
+      "step": 15700
+    },
+    {
+      "epoch": 2.15,
+      "learning_rate": 0.0018507783441245353,
+      "loss": 0.5919,
+      "step": 15710
+    },
+    {
+      "epoch": 2.15,
+      "learning_rate": 0.0018502273040363686,
+      "loss": 0.4662,
+      "step": 15720
+    },
+    {
+      "epoch": 2.16,
+      "learning_rate": 0.0018496762639482022,
+      "loss": 0.4787,
+      "step": 15730
+    },
+    {
+      "epoch": 2.16,
+      "learning_rate": 0.0018491252238600358,
+      "loss": 0.4783,
+      "step": 15740
+    },
+    {
+      "epoch": 2.16,
+      "learning_rate": 0.0018485741837718694,
+      "loss": 0.5295,
+      "step": 15750
+    },
+    {
+      "epoch": 2.16,
+      "learning_rate": 0.001848023143683703,
+      "loss": 0.5334,
+      "step": 15760
+    },
+    {
+      "epoch": 2.16,
+      "learning_rate": 0.0018474721035955366,
+      "loss": 0.5291,
+      "step": 15770
+    },
+    {
+      "epoch": 2.16,
+      "learning_rate": 0.00184692106350737,
+      "loss": 0.4833,
+      "step": 15780
+    },
+    {
+      "epoch": 2.16,
+      "learning_rate": 0.0018463700234192038,
+      "loss": 0.4707,
+      "step": 15790
+    },
+    {
+      "epoch": 2.16,
+      "learning_rate": 0.0018458189833310374,
+      "loss": 0.4258,
+      "step": 15800
+    },
+    {
+      "epoch": 2.16,
+      "eval_loss": 0.4388333261013031,
+      "eval_runtime": 1075.4883,
+      "eval_samples_per_second": 10.863,
+      "eval_steps_per_second": 1.358,
+      "eval_wer": 0.7353315746384063,
+      "step": 15800
+    },
+    {
+      "epoch": 2.17,
+      "learning_rate": 0.001845267943242871,
+      "loss": 0.6091,
+      "step": 15810
+    },
+    {
+      "epoch": 2.17,
+      "learning_rate": 0.0018447169031547046,
+      "loss": 0.4915,
+      "step": 15820
+    },
+    {
+      "epoch": 2.17,
+      "learning_rate": 0.0018441658630665382,
+      "loss": 0.5196,
+      "step": 15830
+    },
+    {
+      "epoch": 2.17,
+      "learning_rate": 0.0018436148229783718,
+      "loss": 0.4946,
+      "step": 15840
+    },
+    {
+      "epoch": 2.17,
+      "learning_rate": 0.0018430637828902052,
+      "loss": 0.5427,
+      "step": 15850
+    },
+    {
+      "epoch": 2.17,
+      "learning_rate": 0.001842512742802039,
+      "loss": 0.6653,
+      "step": 15860
+    },
+    {
+      "epoch": 2.17,
+      "learning_rate": 0.0018419617027138724,
+      "loss": 0.5577,
+      "step": 15870
+    },
+    {
+      "epoch": 2.18,
+      "learning_rate": 0.0018414106626257062,
+      "loss": 0.5711,
+      "step": 15880
+    },
+    {
+      "epoch": 2.18,
+      "learning_rate": 0.0018408596225375396,
+      "loss": 0.5572,
+      "step": 15890
+    },
+    {
+      "epoch": 2.18,
+      "learning_rate": 0.0018403085824493734,
+      "loss": 0.6645,
+      "step": 15900
+    },
+    {
+      "epoch": 2.18,
+      "eval_loss": 0.5061728954315186,
+      "eval_runtime": 1073.5954,
+      "eval_samples_per_second": 10.882,
+      "eval_steps_per_second": 1.361,
+      "eval_wer": 0.7838965705448923,
+      "step": 15900
+    },
+    {
+      "epoch": 2.18,
+      "learning_rate": 0.0018397575423612068,
+      "loss": 0.565,
+      "step": 15910
+    },
+    {
+      "epoch": 2.18,
+      "learning_rate": 0.0018392065022730403,
+      "loss": 0.541,
+      "step": 15920
+    },
+    {
+      "epoch": 2.18,
+      "learning_rate": 0.001838655462184874,
+      "loss": 0.4845,
+      "step": 15930
+    },
+    {
+      "epoch": 2.18,
+      "learning_rate": 0.0018381044220967075,
+      "loss": 0.5094,
+      "step": 15940
+    },
+    {
+      "epoch": 2.19,
+      "learning_rate": 0.0018375533820085411,
+      "loss": 0.5269,
+      "step": 15950
+    },
+    {
+      "epoch": 2.19,
+      "learning_rate": 0.0018370023419203747,
+      "loss": 0.539,
+      "step": 15960
+    },
+    {
+      "epoch": 2.19,
+      "learning_rate": 0.0018364513018322083,
+      "loss": 0.5211,
+      "step": 15970
+    },
+    {
+      "epoch": 2.19,
+      "learning_rate": 0.001835900261744042,
+      "loss": 0.5068,
+      "step": 15980
+    },
+    {
+      "epoch": 2.19,
+      "learning_rate": 0.0018353492216558755,
+      "loss": 0.5009,
+      "step": 15990
+    },
+    {
+      "epoch": 2.19,
+      "learning_rate": 0.0018347981815677091,
+      "loss": 0.4875,
+      "step": 16000
+    },
+    {
+      "epoch": 2.19,
+      "eval_loss": 0.49386918544769287,
+      "eval_runtime": 1076.1477,
+      "eval_samples_per_second": 10.856,
+      "eval_steps_per_second": 1.358,
+      "eval_wer": 0.78132675338852,
+      "step": 16000
+    },
+    {
+      "epoch": 2.19,
+      "learning_rate": 0.0018342471414795427,
+      "loss": 0.5676,
+      "step": 16010
+    },
+    {
+      "epoch": 2.19,
+      "learning_rate": 0.0018336961013913763,
+      "loss": 0.4928,
+      "step": 16020
+    },
+    {
+      "epoch": 2.2,
+      "learning_rate": 0.00183314506130321,
+      "loss": 0.4918,
+      "step": 16030
+    },
+    {
+      "epoch": 2.2,
+      "learning_rate": 0.0018325940212150433,
+      "loss": 0.5225,
+      "step": 16040
+    },
+    {
+      "epoch": 2.2,
+      "learning_rate": 0.001832042981126877,
+      "loss": 0.5276,
+      "step": 16050
+    },
+    {
+      "epoch": 2.2,
+      "learning_rate": 0.0018314919410387105,
+      "loss": 0.5817,
+      "step": 16060
+    },
+    {
+      "epoch": 2.2,
+      "learning_rate": 0.0018309409009505443,
+      "loss": 0.5011,
+      "step": 16070
+    },
+    {
+      "epoch": 2.2,
+      "learning_rate": 0.0018303898608623777,
+      "loss": 0.4848,
+      "step": 16080
+    },
+    {
+      "epoch": 2.2,
+      "learning_rate": 0.0018298388207742115,
+      "loss": 0.52,
+      "step": 16090
+    },
+    {
+      "epoch": 2.21,
+      "learning_rate": 0.0018292877806860449,
+      "loss": 0.5195,
+      "step": 16100
+    },
+    {
+      "epoch": 2.21,
+      "eval_loss": 0.4683535397052765,
+      "eval_runtime": 1072.5877,
+      "eval_samples_per_second": 10.892,
+      "eval_steps_per_second": 1.362,
+      "eval_wer": 0.7653393068316201,
+      "step": 16100
+    },
+    {
+      "epoch": 2.21,
+      "learning_rate": 0.0018287367405978785,
+      "loss": 0.5253,
+      "step": 16110
+    },
+    {
+      "epoch": 2.21,
+      "learning_rate": 0.0018281857005097123,
+      "loss": 0.5137,
+      "step": 16120
+    },
+    {
+      "epoch": 2.21,
+      "learning_rate": 0.0018276346604215456,
+      "loss": 0.4561,
+      "step": 16130
+    },
+    {
+      "epoch": 2.21,
+      "learning_rate": 0.0018270836203333795,
+      "loss": 0.4788,
+      "step": 16140
+    },
+    {
+      "epoch": 2.21,
+      "learning_rate": 0.0018265325802452128,
+      "loss": 0.5289,
+      "step": 16150
+    },
+    {
+      "epoch": 2.21,
+      "learning_rate": 0.0018259815401570464,
+      "loss": 0.5685,
+      "step": 16160
+    },
+    {
+      "epoch": 2.22,
+      "learning_rate": 0.00182543050006888,
+      "loss": 0.4932,
+      "step": 16170
+    },
+    {
+      "epoch": 2.22,
+      "learning_rate": 0.0018248794599807136,
+      "loss": 0.4953,
+      "step": 16180
+    },
+    {
+      "epoch": 2.22,
+      "learning_rate": 0.0018243284198925472,
+      "loss": 0.6045,
+      "step": 16190
+    },
+    {
+      "epoch": 2.22,
+      "learning_rate": 0.0018237773798043808,
+      "loss": 0.5162,
+      "step": 16200
+    },
+    {
+      "epoch": 2.22,
+      "eval_loss": 0.4421519339084625,
+      "eval_runtime": 1065.8605,
+      "eval_samples_per_second": 10.961,
+      "eval_steps_per_second": 1.371,
+      "eval_wer": 0.7386973528609115,
+      "step": 16200
+    },
+    {
+      "epoch": 2.22,
+      "learning_rate": 0.0018232263397162144,
+      "loss": 0.5558,
+      "step": 16210
+    },
+    {
+      "epoch": 2.22,
+      "learning_rate": 0.001822675299628048,
+      "loss": 0.5607,
+      "step": 16220
+    },
+    {
+      "epoch": 2.22,
+      "learning_rate": 0.0018221242595398814,
+      "loss": 0.5122,
+      "step": 16230
+    },
+    {
+      "epoch": 2.22,
+      "learning_rate": 0.0018215732194517152,
+      "loss": 0.5324,
+      "step": 16240
+    },
+    {
+      "epoch": 2.23,
+      "learning_rate": 0.0018210221793635486,
+      "loss": 0.5185,
+      "step": 16250
+    },
+    {
+      "epoch": 2.23,
+      "learning_rate": 0.0018204711392753824,
+      "loss": 0.5441,
+      "step": 16260
+    },
+    {
+      "epoch": 2.23,
+      "learning_rate": 0.001819920099187216,
+      "loss": 0.4951,
+      "step": 16270
+    },
+    {
+      "epoch": 2.23,
+      "learning_rate": 0.0018193690590990496,
+      "loss": 0.5159,
+      "step": 16280
+    },
+    {
+      "epoch": 2.23,
+      "learning_rate": 0.0018188180190108832,
+      "loss": 0.5226,
+      "step": 16290
+    },
+    {
+      "epoch": 2.23,
+      "learning_rate": 0.0018182669789227166,
+      "loss": 0.4851,
+      "step": 16300
+    },
+    {
+      "epoch": 2.23,
+      "eval_loss": 0.471422404050827,
+      "eval_runtime": 1053.3991,
+      "eval_samples_per_second": 11.091,
+      "eval_steps_per_second": 1.387,
+      "eval_wer": 0.7618825616301282,
+      "step": 16300
+    },
+    {
+      "epoch": 2.23,
+      "learning_rate": 0.0018177159388345504,
+      "loss": 0.5515,
+      "step": 16310
+    },
+    {
+      "epoch": 2.24,
+      "learning_rate": 0.0018171648987463838,
+      "loss": 0.5406,
+      "step": 16320
+    },
+    {
+      "epoch": 2.24,
+      "learning_rate": 0.0018166138586582176,
+      "loss": 0.5429,
+      "step": 16330
+    },
+    {
+      "epoch": 2.24,
+      "learning_rate": 0.001816062818570051,
+      "loss": 0.5044,
+      "step": 16340
+    },
+    {
+      "epoch": 2.24,
+      "learning_rate": 0.0018155117784818845,
+      "loss": 0.5177,
+      "step": 16350
+    },
+    {
+      "epoch": 2.24,
+      "learning_rate": 0.0018149607383937181,
+      "loss": 0.5407,
+      "step": 16360
+    },
+    {
+      "epoch": 2.24,
+      "learning_rate": 0.0018144096983055517,
+      "loss": 0.4788,
+      "step": 16370
+    },
+    {
+      "epoch": 2.24,
+      "learning_rate": 0.0018138586582173853,
+      "loss": 0.4926,
+      "step": 16380
+    },
+    {
+      "epoch": 2.25,
+      "learning_rate": 0.001813307618129219,
+      "loss": 0.5092,
+      "step": 16390
+    },
+    {
+      "epoch": 2.25,
+      "learning_rate": 0.0018127565780410525,
+      "loss": 0.536,
+      "step": 16400
+    },
+    {
+      "epoch": 2.25,
+      "eval_loss": 0.477826863527298,
+      "eval_runtime": 1053.145,
+      "eval_samples_per_second": 11.093,
+      "eval_steps_per_second": 1.387,
+      "eval_wer": 0.7589602474301829,
+      "step": 16400
+    },
+    {
+      "epoch": 2.25,
+      "learning_rate": 0.0018122055379528861,
+      "loss": 0.5386,
+      "step": 16410
+    },
+    {
+      "epoch": 2.25,
+      "learning_rate": 0.0018116544978647195,
+      "loss": 0.5379,
+      "step": 16420
+    },
+    {
+      "epoch": 2.25,
+      "learning_rate": 0.0018111034577765533,
+      "loss": 0.506,
+      "step": 16430
+    },
+    {
+      "epoch": 2.25,
+      "learning_rate": 0.001810552417688387,
+      "loss": 0.4759,
+      "step": 16440
+    },
+    {
+      "epoch": 2.25,
+      "learning_rate": 0.0018100013776002205,
+      "loss": 0.4609,
+      "step": 16450
+    },
+    {
+      "epoch": 2.26,
+      "learning_rate": 0.001809450337512054,
+      "loss": 0.5147,
+      "step": 16460
+    },
+    {
+      "epoch": 2.26,
+      "learning_rate": 0.0018088992974238877,
+      "loss": 0.4671,
+      "step": 16470
+    },
+    {
+      "epoch": 2.26,
+      "learning_rate": 0.0018083482573357213,
+      "loss": 0.4362,
+      "step": 16480
+    },
+    {
+      "epoch": 2.26,
+      "learning_rate": 0.0018077972172475547,
+      "loss": 0.5137,
+      "step": 16490
+    },
+    {
+      "epoch": 2.26,
+      "learning_rate": 0.0018072461771593885,
+      "loss": 0.403,
+      "step": 16500
+    },
+    {
+      "epoch": 2.26,
+      "eval_loss": 0.4671553075313568,
+      "eval_runtime": 1054.804,
+      "eval_samples_per_second": 11.076,
+      "eval_steps_per_second": 1.385,
+      "eval_wer": 0.7729464204493769,
+      "step": 16500
+    },
+    {
+      "epoch": 2.26,
+      "learning_rate": 0.0018066951370712219,
+      "loss": 0.5662,
+      "step": 16510
+    },
+    {
+      "epoch": 2.26,
+      "learning_rate": 0.0018061440969830557,
+      "loss": 0.4852,
+      "step": 16520
+    },
+    {
+      "epoch": 2.26,
+      "learning_rate": 0.001805593056894889,
+      "loss": 0.4885,
+      "step": 16530
+    },
+    {
+      "epoch": 2.27,
+      "learning_rate": 0.0018050420168067227,
+      "loss": 0.488,
+      "step": 16540
+    },
+    {
+      "epoch": 2.27,
+      "learning_rate": 0.0018044909767185563,
+      "loss": 0.5101,
+      "step": 16550
+    },
+    {
+      "epoch": 2.27,
+      "learning_rate": 0.0018039399366303898,
+      "loss": 0.5494,
+      "step": 16560
+    },
+    {
+      "epoch": 2.27,
+      "learning_rate": 0.0018033888965422234,
+      "loss": 0.5157,
+      "step": 16570
+    },
+    {
+      "epoch": 2.27,
+      "learning_rate": 0.001802837856454057,
+      "loss": 0.4512,
+      "step": 16580
+    },
+    {
+      "epoch": 2.27,
+      "learning_rate": 0.0018022868163658909,
+      "loss": 0.4831,
+      "step": 16590
+    },
+    {
+      "epoch": 2.27,
+      "learning_rate": 0.0018017357762777242,
+      "loss": 0.4217,
+      "step": 16600
+    },
+    {
+      "epoch": 2.27,
+      "eval_loss": 0.4305126965045929,
+      "eval_runtime": 1053.1518,
+      "eval_samples_per_second": 11.093,
+      "eval_steps_per_second": 1.387,
+      "eval_wer": 0.732898207950514,
+      "step": 16600
+    },
+    {
+      "epoch": 2.28,
+      "learning_rate": 0.0018011847361895578,
+      "loss": 0.4984,
+      "step": 16610
+    },
+    {
+      "epoch": 2.28,
+      "learning_rate": 0.0018006336961013914,
+      "loss": 0.4745,
+      "step": 16620
+    },
+    {
+      "epoch": 2.28,
+      "learning_rate": 0.001800082656013225,
+      "loss": 0.4924,
+      "step": 16630
+    },
+    {
+      "epoch": 2.28,
+      "learning_rate": 0.0017995316159250586,
+      "loss": 0.4658,
+      "step": 16640
+    },
+    {
+      "epoch": 2.28,
+      "learning_rate": 0.0017989805758368922,
+      "loss": 0.4441,
+      "step": 16650
+    },
+    {
+      "epoch": 2.28,
+      "learning_rate": 0.0017984295357487256,
+      "loss": 0.4939,
+      "step": 16660
+    },
+    {
+      "epoch": 2.28,
+      "learning_rate": 0.0017978784956605594,
+      "loss": 0.4888,
+      "step": 16670
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 0.0017973274555723928,
+      "loss": 0.5057,
+      "step": 16680
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 0.0017967764154842266,
+      "loss": 0.4993,
+      "step": 16690
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 0.00179622537539606,
+      "loss": 0.4865,
+      "step": 16700
+    },
+    {
+      "epoch": 2.29,
+      "eval_loss": 0.45768412947654724,
+      "eval_runtime": 1052.361,
+      "eval_samples_per_second": 11.102,
+      "eval_steps_per_second": 1.388,
+      "eval_wer": 0.7367301919403256,
+      "step": 16700
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 0.0017956743353078938,
+      "loss": 0.5034,
+      "step": 16710
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 0.0017951232952197272,
+      "loss": 0.4687,
+      "step": 16720
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 0.0017945722551315608,
+      "loss": 0.533,
+      "step": 16730
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 0.0017940212150433946,
+      "loss": 0.485,
+      "step": 16740
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 0.001793470174955228,
+      "loss": 0.5868,
+      "step": 16750
+    },
+    {
+      "epoch": 2.3,
+      "learning_rate": 0.0017929191348670618,
+      "loss": 0.5512,
+      "step": 16760
+    },
+    {
+      "epoch": 2.3,
+      "learning_rate": 0.0017923680947788951,
+      "loss": 0.4544,
+      "step": 16770
+    },
+    {
+      "epoch": 2.3,
+      "learning_rate": 0.001791817054690729,
+      "loss": 0.4859,
+      "step": 16780
+    },
+    {
+      "epoch": 2.3,
+      "learning_rate": 0.0017912660146025623,
+      "loss": 0.4832,
+      "step": 16790
+    },
+    {
+      "epoch": 2.3,
+      "learning_rate": 0.001790714974514396,
+      "loss": 0.4889,
+      "step": 16800
+    },
+    {
+      "epoch": 2.3,
+      "eval_loss": 0.4397953748703003,
+      "eval_runtime": 1054.2308,
+      "eval_samples_per_second": 11.082,
+      "eval_steps_per_second": 1.386,
+      "eval_wer": 0.728156554170836,
+      "step": 16800
+    },
+    {
+      "epoch": 2.3,
+      "learning_rate": 0.0017901639344262295,
+      "loss": 0.4971,
+      "step": 16810
+    },
+    {
+      "epoch": 2.3,
+      "learning_rate": 0.0017896128943380631,
+      "loss": 0.4847,
+      "step": 16820
+    },
+    {
+      "epoch": 2.31,
+      "learning_rate": 0.0017890618542498967,
+      "loss": 0.5,
+      "step": 16830
+    },
+    {
+      "epoch": 2.31,
+      "learning_rate": 0.0017885108141617303,
+      "loss": 0.4368,
+      "step": 16840
+    },
+    {
+      "epoch": 2.31,
+      "learning_rate": 0.0017879597740735637,
+      "loss": 0.4865,
+      "step": 16850
+    },
+    {
+      "epoch": 2.31,
+      "learning_rate": 0.0017874087339853975,
+      "loss": 0.5118,
+      "step": 16860
+    },
+    {
+      "epoch": 2.31,
+      "learning_rate": 0.001786857693897231,
+      "loss": 0.472,
+      "step": 16870
+    },
+    {
+      "epoch": 2.31,
+      "learning_rate": 0.0017863066538090647,
+      "loss": 0.479,
+      "step": 16880
+    },
+    {
+      "epoch": 2.31,
+      "learning_rate": 0.0017857556137208983,
+      "loss": 0.573,
+      "step": 16890
+    },
+    {
+      "epoch": 2.32,
+      "learning_rate": 0.001785204573632732,
+      "loss": 0.4094,
+      "step": 16900
+    },
+    {
+      "epoch": 2.32,
+      "eval_loss": 0.4433184266090393,
+      "eval_runtime": 1053.0718,
+      "eval_samples_per_second": 11.094,
+      "eval_steps_per_second": 1.387,
+      "eval_wer": 0.7244837623942509,
+      "step": 16900
+    },
+    {
+      "epoch": 2.32,
+      "learning_rate": 0.0017846535335445655,
+      "loss": 0.4922,
+      "step": 16910
+    },
+    {
+      "epoch": 2.32,
+      "learning_rate": 0.0017841024934563989,
+      "loss": 0.4766,
+      "step": 16920
+    },
+    {
+      "epoch": 2.32,
+      "learning_rate": 0.0017835514533682327,
+      "loss": 0.5223,
+      "step": 16930
+    },
+    {
+      "epoch": 2.32,
+      "learning_rate": 0.001783000413280066,
+      "loss": 0.5221,
+      "step": 16940
+    },
+    {
+      "epoch": 2.32,
+      "learning_rate": 0.0017824493731918999,
+      "loss": 0.429,
+      "step": 16950
+    },
+    {
+      "epoch": 2.32,
+      "learning_rate": 0.0017818983331037333,
+      "loss": 0.4991,
+      "step": 16960
+    },
+    {
+      "epoch": 2.32,
+      "learning_rate": 0.001781347293015567,
+      "loss": 0.4557,
+      "step": 16970
+    },
+    {
+      "epoch": 2.33,
+      "learning_rate": 0.0017807962529274004,
+      "loss": 0.4615,
+      "step": 16980
+    },
+    {
+      "epoch": 2.33,
+      "learning_rate": 0.001780245212839234,
+      "loss": 0.5181,
+      "step": 16990
+    },
+    {
+      "epoch": 2.33,
+      "learning_rate": 0.0017796941727510676,
+      "loss": 0.4654,
+      "step": 17000
+    },
+    {
+      "epoch": 2.33,
+      "eval_loss": 0.4121365547180176,
+      "eval_runtime": 1055.5565,
+      "eval_samples_per_second": 11.068,
+      "eval_steps_per_second": 1.384,
+      "eval_wer": 0.7142727190030019,
+      "step": 17000
+    },
+    {
+      "epoch": 2.33,
+      "learning_rate": 0.0017791431326629012,
+      "loss": 0.4477,
+      "step": 17010
+    },
+    {
+      "epoch": 2.33,
+      "learning_rate": 0.0017785920925747348,
+      "loss": 0.459,
+      "step": 17020
+    },
+    {
+      "epoch": 2.33,
+      "learning_rate": 0.0017780410524865684,
+      "loss": 0.4788,
+      "step": 17030
+    },
+    {
+      "epoch": 2.33,
+      "learning_rate": 0.0017774900123984018,
+      "loss": 0.4779,
+      "step": 17040
+    },
+    {
+      "epoch": 2.34,
+      "learning_rate": 0.0017769389723102356,
+      "loss": 0.4616,
+      "step": 17050
+    },
+    {
+      "epoch": 2.34,
+      "learning_rate": 0.0017763879322220692,
+      "loss": 0.4624,
+      "step": 17060
+    },
+    {
+      "epoch": 2.34,
+      "learning_rate": 0.0017758368921339028,
+      "loss": 0.4753,
+      "step": 17070
+    },
+    {
+      "epoch": 2.34,
+      "learning_rate": 0.0017752858520457364,
+      "loss": 0.5359,
+      "step": 17080
+    },
+    {
+      "epoch": 2.34,
+      "learning_rate": 0.00177473481195757,
+      "loss": 0.4931,
+      "step": 17090
+    },
+    {
+      "epoch": 2.34,
+      "learning_rate": 0.0017741837718694036,
+      "loss": 0.394,
+      "step": 17100
+    },
+    {
+      "epoch": 2.34,
+      "eval_loss": 0.4642173945903778,
+      "eval_runtime": 1056.2456,
+      "eval_samples_per_second": 11.061,
+      "eval_steps_per_second": 1.383,
+      "eval_wer": 0.7267806786136632,
+      "step": 17100
+    },
+    {
+      "epoch": 2.34,
+      "learning_rate": 0.001773632731781237,
+      "loss": 0.5476,
+      "step": 17110
+    },
+    {
+      "epoch": 2.35,
+      "learning_rate": 0.0017730816916930708,
+      "loss": 0.4746,
+      "step": 17120
+    },
+    {
+      "epoch": 2.35,
+      "learning_rate": 0.0017725306516049042,
+      "loss": 0.4542,
+      "step": 17130
+    },
+    {
+      "epoch": 2.35,
+      "learning_rate": 0.001771979611516738,
+      "loss": 0.4702,
+      "step": 17140
+    },
+    {
+      "epoch": 2.35,
+      "learning_rate": 0.0017714285714285714,
+      "loss": 0.3867,
+      "step": 17150
+    },
+    {
+      "epoch": 2.35,
+      "learning_rate": 0.0017708775313404052,
+      "loss": 0.53,
+      "step": 17160
+    },
+    {
+      "epoch": 2.35,
+      "learning_rate": 0.0017703264912522386,
+      "loss": 0.493,
+      "step": 17170
+    },
+    {
+      "epoch": 2.35,
+      "learning_rate": 0.0017697754511640722,
+      "loss": 0.4669,
+      "step": 17180
+    },
+    {
+      "epoch": 2.36,
+      "learning_rate": 0.0017692244110759058,
+      "loss": 0.463,
+      "step": 17190
+    },
+    {
+      "epoch": 2.36,
+      "learning_rate": 0.0017686733709877393,
+      "loss": 0.4595,
+      "step": 17200
+    },
+    {
+      "epoch": 2.36,
+      "eval_loss": 0.40757834911346436,
+      "eval_runtime": 1054.2362,
+      "eval_samples_per_second": 11.082,
+      "eval_steps_per_second": 1.386,
+      "eval_wer": 0.7034021650140999,
+      "step": 17200
+    },
+    {
+      "epoch": 2.36,
+      "learning_rate": 0.0017681223308995732,
+      "loss": 0.4914,
+      "step": 17210
+    },
+    {
+      "epoch": 2.36,
+      "learning_rate": 0.0017675712908114065,
+      "loss": 0.5259,
+      "step": 17220
+    },
+    {
+      "epoch": 2.36,
+      "learning_rate": 0.0017670202507232401,
+      "loss": 0.4793,
+      "step": 17230
+    },
+    {
+      "epoch": 2.36,
+      "learning_rate": 0.0017664692106350737,
+      "loss": 0.497,
+      "step": 17240
+    },
+    {
+      "epoch": 2.36,
+      "learning_rate": 0.0017659181705469073,
+      "loss": 0.4498,
+      "step": 17250
+    },
+    {
+      "epoch": 2.36,
+      "learning_rate": 0.001765367130458741,
+      "loss": 0.5271,
+      "step": 17260
+    },
+    {
+      "epoch": 2.37,
+      "learning_rate": 0.0017648160903705745,
+      "loss": 0.5563,
+      "step": 17270
+    },
+    {
+      "epoch": 2.37,
+      "learning_rate": 0.0017642650502824081,
+      "loss": 0.453,
+      "step": 17280
+    },
+    {
+      "epoch": 2.37,
+      "learning_rate": 0.0017637140101942417,
+      "loss": 0.4934,
+      "step": 17290
+    },
+    {
+      "epoch": 2.37,
+      "learning_rate": 0.001763162970106075,
+      "loss": 0.4463,
+      "step": 17300
+    },
+    {
+      "epoch": 2.37,
+      "eval_loss": 0.4476642608642578,
+      "eval_runtime": 1055.6534,
+      "eval_samples_per_second": 11.067,
+      "eval_steps_per_second": 1.384,
+      "eval_wer": 0.7619280451196215,
+      "step": 17300
+    },
+    {
+      "epoch": 2.37,
+      "learning_rate": 0.001762611930017909,
+      "loss": 0.5428,
+      "step": 17310
+    },
+    {
+      "epoch": 2.37,
+      "learning_rate": 0.0017620608899297423,
+      "loss": 0.4991,
+      "step": 17320
+    },
+    {
+      "epoch": 2.37,
+      "learning_rate": 0.001761509849841576,
+      "loss": 0.4893,
+      "step": 17330
+    },
+    {
+      "epoch": 2.38,
+      "learning_rate": 0.0017609588097534095,
+      "loss": 0.4443,
+      "step": 17340
+    },
+    {
+      "epoch": 2.38,
+      "learning_rate": 0.0017604077696652433,
+      "loss": 0.41,
+      "step": 17350
+    },
+    {
+      "epoch": 2.38,
+      "learning_rate": 0.0017598567295770769,
+      "loss": 0.5842,
+      "step": 17360
+    },
+    {
+      "epoch": 2.38,
+      "learning_rate": 0.0017593056894889103,
+      "loss": 0.5285,
+      "step": 17370
+    },
+    {
+      "epoch": 2.38,
+      "learning_rate": 0.001758754649400744,
+      "loss": 0.4419,
+      "step": 17380
+    },
+    {
+      "epoch": 2.38,
+      "learning_rate": 0.0017582036093125775,
+      "loss": 0.4966,
+      "step": 17390
+    },
+    {
+      "epoch": 2.38,
+      "learning_rate": 0.0017576525692244113,
+      "loss": 0.5096,
+      "step": 17400
+    },
+    {
+      "epoch": 2.38,
+      "eval_loss": 0.4394005835056305,
+      "eval_runtime": 1055.0929,
+      "eval_samples_per_second": 11.073,
+      "eval_steps_per_second": 1.385,
+      "eval_wer": 0.7323637769489675,
+      "step": 17400
+    },
+    {
+      "epoch": 2.39,
+      "learning_rate": 0.0017571015291362446,
+      "loss": 0.4738,
+      "step": 17410
+    },
+    {
+      "epoch": 2.39,
+      "learning_rate": 0.0017565504890480782,
+      "loss": 0.4552,
+      "step": 17420
+    },
+    {
+      "epoch": 2.39,
+      "learning_rate": 0.0017559994489599118,
+      "loss": 0.4751,
+      "step": 17430
+    },
+    {
+      "epoch": 2.39,
+      "learning_rate": 0.0017554484088717454,
+      "loss": 0.4169,
+      "step": 17440
+    },
+    {
+      "epoch": 2.39,
+      "learning_rate": 0.001754897368783579,
+      "loss": 0.4897,
+      "step": 17450
+    },
+    {
+      "epoch": 2.39,
+      "learning_rate": 0.0017543463286954126,
+      "loss": 0.4707,
+      "step": 17460
+    },
+    {
+      "epoch": 2.39,
+      "learning_rate": 0.0017537952886072462,
+      "loss": 0.4352,
+      "step": 17470
+    },
+    {
+      "epoch": 2.39,
+      "learning_rate": 0.0017532442485190798,
+      "loss": 0.4455,
+      "step": 17480
+    },
+    {
+      "epoch": 2.4,
+      "learning_rate": 0.0017526932084309132,
+      "loss": 0.4235,
+      "step": 17490
+    },
+    {
+      "epoch": 2.4,
+      "learning_rate": 0.001752142168342747,
+      "loss": 0.3934,
+      "step": 17500
+    },
+    {
+      "epoch": 2.4,
+      "eval_loss": 0.3927477300167084,
+      "eval_runtime": 1054.3105,
+      "eval_samples_per_second": 11.081,
+      "eval_steps_per_second": 1.386,
+      "eval_wer": 0.7054375511689257,
+      "step": 17500
+    },
+    {
+      "epoch": 2.4,
+      "learning_rate": 0.0017515911282545806,
+      "loss": 0.458,
+      "step": 17510
+    },
+    {
+      "epoch": 2.4,
+      "learning_rate": 0.0017510400881664142,
+      "loss": 0.4481,
+      "step": 17520
+    },
+    {
+      "epoch": 2.4,
+      "learning_rate": 0.0017504890480782478,
+      "loss": 0.4682,
+      "step": 17530
+    },
+    {
+      "epoch": 2.4,
+      "learning_rate": 0.0017499380079900814,
+      "loss": 0.4444,
+      "step": 17540
+    },
+    {
+      "epoch": 2.4,
+      "learning_rate": 0.001749386967901915,
+      "loss": 0.3908,
+      "step": 17550
+    },
+    {
+      "epoch": 2.41,
+      "learning_rate": 0.0017488359278137484,
+      "loss": 0.4802,
+      "step": 17560
+    },
+    {
+      "epoch": 2.41,
+      "learning_rate": 0.0017482848877255822,
+      "loss": 0.4737,
+      "step": 17570
+    },
+    {
+      "epoch": 2.41,
+      "learning_rate": 0.0017477338476374156,
+      "loss": 0.4303,
+      "step": 17580
+    },
+    {
+      "epoch": 2.41,
+      "learning_rate": 0.0017471828075492494,
+      "loss": 0.4784,
+      "step": 17590
+    },
+    {
+      "epoch": 2.41,
+      "learning_rate": 0.0017466317674610828,
+      "loss": 0.3766,
+      "step": 17600
+    },
+    {
+      "epoch": 2.41,
+      "eval_loss": 0.4441586136817932,
+      "eval_runtime": 1055.7713,
+      "eval_samples_per_second": 11.066,
+      "eval_steps_per_second": 1.384,
+      "eval_wer": 0.7404712089511507,
+      "step": 17600
+    },
+    {
+      "epoch": 2.41,
+      "learning_rate": 0.0017460807273729164,
+      "loss": 0.5494,
+      "step": 17610
+    },
+    {
+      "epoch": 2.41,
+      "learning_rate": 0.00174552968728475,
+      "loss": 0.4816,
+      "step": 17620
+    },
+    {
+      "epoch": 2.42,
+      "learning_rate": 0.0017449786471965835,
+      "loss": 0.477,
+      "step": 17630
+    },
+    {
+      "epoch": 2.42,
+      "learning_rate": 0.0017444276071084171,
+      "loss": 0.4655,
+      "step": 17640
+    },
+    {
+      "epoch": 2.42,
+      "learning_rate": 0.0017438765670202507,
+      "loss": 0.4493,
+      "step": 17650
+    },
+    {
+      "epoch": 2.42,
+      "learning_rate": 0.0017433255269320843,
+      "loss": 0.5322,
+      "step": 17660
+    },
+    {
+      "epoch": 2.42,
+      "learning_rate": 0.001742774486843918,
+      "loss": 0.469,
+      "step": 17670
+    },
+    {
+      "epoch": 2.42,
+      "learning_rate": 0.0017422234467557515,
+      "loss": 0.4617,
+      "step": 17680
+    },
+    {
+      "epoch": 2.42,
+      "learning_rate": 0.0017416724066675851,
+      "loss": 0.4681,
+      "step": 17690
+    },
+    {
+      "epoch": 2.42,
+      "learning_rate": 0.0017411213665794187,
+      "loss": 0.4216,
+      "step": 17700
+    },
+    {
+      "epoch": 2.42,
+      "eval_loss": 0.4131408631801605,
+      "eval_runtime": 1055.5874,
+      "eval_samples_per_second": 11.068,
+      "eval_steps_per_second": 1.384,
+      "eval_wer": 0.6990926043846084,
+      "step": 17700
+    },
+    {
+      "epoch": 2.43,
+      "learning_rate": 0.0017405703264912523,
+      "loss": 0.4797,
+      "step": 17710
+    },
+    {
+      "epoch": 2.43,
+      "learning_rate": 0.001740019286403086,
+      "loss": 0.5957,
+      "step": 17720
+    },
+    {
+      "epoch": 2.43,
+      "learning_rate": 0.0017394682463149195,
+      "loss": 0.4808,
+      "step": 17730
+    },
+    {
+      "epoch": 2.43,
+      "learning_rate": 0.001738917206226753,
+      "loss": 0.4488,
+      "step": 17740
+    },
+    {
+      "epoch": 2.43,
+      "learning_rate": 0.0017383661661385865,
+      "loss": 0.4166,
+      "step": 17750
+    },
+    {
+      "epoch": 2.43,
+      "learning_rate": 0.0017378151260504203,
+      "loss": 0.5195,
+      "step": 17760
+    },
+    {
+      "epoch": 2.43,
+      "learning_rate": 0.0017372640859622537,
+      "loss": 0.4294,
+      "step": 17770
+    },
+    {
+      "epoch": 2.44,
+      "learning_rate": 0.0017367130458740875,
+      "loss": 0.4121,
+      "step": 17780
+    },
+    {
+      "epoch": 2.44,
+      "learning_rate": 0.0017361620057859209,
+      "loss": 0.4333,
+      "step": 17790
+    },
+    {
+      "epoch": 2.44,
+      "learning_rate": 0.0017356109656977545,
+      "loss": 0.3822,
+      "step": 17800
+    },
+    {
+      "epoch": 2.44,
+      "eval_loss": 0.3996317386627197,
+      "eval_runtime": 1055.4182,
+      "eval_samples_per_second": 11.07,
+      "eval_steps_per_second": 1.384,
+      "eval_wer": 0.7095196943509506,
+      "step": 17800
+    },
+    {
+      "epoch": 2.44,
+      "learning_rate": 0.001735059925609588,
+      "loss": 0.4921,
+      "step": 17810
+    },
+    {
+      "epoch": 2.44,
+      "learning_rate": 0.0017345088855214217,
+      "loss": 0.4498,
+      "step": 17820
+    },
+    {
+      "epoch": 2.44,
+      "learning_rate": 0.0017339578454332555,
+      "loss": 0.5018,
+      "step": 17830
+    },
+    {
+      "epoch": 2.44,
+      "learning_rate": 0.0017334068053450888,
+      "loss": 0.4637,
+      "step": 17840
+    },
+    {
+      "epoch": 2.45,
+      "learning_rate": 0.0017328557652569227,
+      "loss": 0.4192,
+      "step": 17850
+    },
+    {
+      "epoch": 2.45,
+      "learning_rate": 0.001732304725168756,
+      "loss": 0.4713,
+      "step": 17860
+    },
+    {
+      "epoch": 2.45,
+      "learning_rate": 0.0017317536850805896,
+      "loss": 0.4238,
+      "step": 17870
+    },
+    {
+      "epoch": 2.45,
+      "learning_rate": 0.0017312026449924232,
+      "loss": 0.4235,
+      "step": 17880
+    },
+    {
+      "epoch": 2.45,
+      "learning_rate": 0.0017306516049042568,
+      "loss": 0.4431,
+      "step": 17890
+    },
+    {
+      "epoch": 2.45,
+      "learning_rate": 0.0017301005648160904,
+      "loss": 0.5685,
+      "step": 17900
+    },
+    {
+      "epoch": 2.45,
+      "eval_loss": 0.378450870513916,
+      "eval_runtime": 1057.2816,
+      "eval_samples_per_second": 11.05,
+      "eval_steps_per_second": 1.382,
+      "eval_wer": 0.6970344764850359,
+      "step": 17900
+    },
+    {
+      "epoch": 2.45,
+      "learning_rate": 0.001729549524727924,
+      "loss": 0.4426,
+      "step": 17910
+    },
+    {
+      "epoch": 2.46,
+      "learning_rate": 0.0017289984846397576,
+      "loss": 0.4376,
+      "step": 17920
+    },
+    {
+      "epoch": 2.46,
+      "learning_rate": 0.0017284474445515912,
+      "loss": 0.4101,
+      "step": 17930
+    },
+    {
+      "epoch": 2.46,
+      "learning_rate": 0.0017278964044634246,
+      "loss": 0.4139,
+      "step": 17940
+    },
+    {
+      "epoch": 2.46,
+      "learning_rate": 0.0017273453643752584,
+      "loss": 0.3741,
+      "step": 17950
+    },
+    {
+      "epoch": 2.46,
+      "learning_rate": 0.0017267943242870918,
+      "loss": 0.5031,
+      "step": 17960
+    },
+    {
+      "epoch": 2.46,
+      "learning_rate": 0.0017262432841989256,
+      "loss": 0.4883,
+      "step": 17970
+    },
+    {
+      "epoch": 2.46,
+      "learning_rate": 0.0017256922441107592,
+      "loss": 0.4426,
+      "step": 17980
+    },
+    {
+      "epoch": 2.46,
+      "learning_rate": 0.0017251412040225926,
+      "loss": 0.4383,
+      "step": 17990
+    },
+    {
+      "epoch": 2.47,
+      "learning_rate": 0.0017245901639344264,
+      "loss": 0.5032,
+      "step": 18000
+    },
+    {
+      "epoch": 2.47,
+      "eval_loss": 0.40151506662368774,
+      "eval_runtime": 1056.4263,
+      "eval_samples_per_second": 11.059,
+      "eval_steps_per_second": 1.383,
+      "eval_wer": 0.7012985536250341,
+      "step": 18000
+    },
+    {
+      "epoch": 2.47,
+      "learning_rate": 0.0017240391238462598,
+      "loss": 0.4593,
+      "step": 18010
+    },
+    {
+      "epoch": 2.47,
+      "learning_rate": 0.0017234880837580936,
+      "loss": 0.4189,
+      "step": 18020
+    },
+    {
+      "epoch": 2.47,
+      "learning_rate": 0.001722937043669927,
+      "loss": 0.4719,
+      "step": 18030
+    },
+    {
+      "epoch": 2.47,
+      "learning_rate": 0.0017223860035817608,
+      "loss": 0.4496,
+      "step": 18040
+    },
+    {
+      "epoch": 2.47,
+      "learning_rate": 0.0017218349634935941,
+      "loss": 0.3813,
+      "step": 18050
+    },
+    {
+      "epoch": 2.47,
+      "learning_rate": 0.0017212839234054277,
+      "loss": 0.4301,
+      "step": 18060
+    },
+    {
+      "epoch": 2.48,
+      "learning_rate": 0.0017207328833172613,
+      "loss": 0.4474,
+      "step": 18070
+    },
+    {
+      "epoch": 2.48,
+      "learning_rate": 0.001720181843229095,
+      "loss": 0.4507,
+      "step": 18080
+    },
+    {
+      "epoch": 2.48,
+      "learning_rate": 0.0017196308031409285,
+      "loss": 0.4455,
+      "step": 18090
+    },
+    {
+      "epoch": 2.48,
+      "learning_rate": 0.0017190797630527621,
+      "loss": 0.3936,
+      "step": 18100
+    },
+    {
+      "epoch": 2.48,
+      "eval_loss": 0.3789331316947937,
+      "eval_runtime": 1056.8996,
+      "eval_samples_per_second": 11.054,
+      "eval_steps_per_second": 1.382,
+      "eval_wer": 0.6921450013645046,
+      "step": 18100
+    },
+    {
+      "epoch": 2.48,
+      "learning_rate": 0.0017185287229645955,
+      "loss": 0.4716,
+      "step": 18110
+    },
+    {
+      "epoch": 2.48,
+      "learning_rate": 0.0017179776828764293,
+      "loss": 0.3993,
+      "step": 18120
+    },
+    {
+      "epoch": 2.48,
+      "learning_rate": 0.0017174266427882627,
+      "loss": 0.4358,
+      "step": 18130
+    },
+    {
+      "epoch": 2.49,
+      "learning_rate": 0.0017168756027000965,
+      "loss": 0.4541,
+      "step": 18140
+    },
+    {
+      "epoch": 2.49,
+      "learning_rate": 0.0017163245626119301,
+      "loss": 0.382,
+      "step": 18150
+    },
+    {
+      "epoch": 2.49,
+      "learning_rate": 0.0017157735225237637,
+      "loss": 0.506,
+      "step": 18160
+    },
+    {
+      "epoch": 2.49,
+      "learning_rate": 0.0017152224824355973,
+      "loss": 0.4647,
+      "step": 18170
+    },
+    {
+      "epoch": 2.49,
+      "learning_rate": 0.0017146714423474307,
+      "loss": 0.4587,
+      "step": 18180
+    },
+    {
+      "epoch": 2.49,
+      "learning_rate": 0.0017141204022592645,
+      "loss": 0.4351,
+      "step": 18190
+    },
+    {
+      "epoch": 2.49,
+      "learning_rate": 0.0017135693621710979,
+      "loss": 0.4129,
+      "step": 18200
+    },
+    {
+      "epoch": 2.49,
+      "eval_loss": 0.4373599886894226,
+      "eval_runtime": 1054.9432,
+      "eval_samples_per_second": 11.075,
+      "eval_steps_per_second": 1.385,
+      "eval_wer": 0.7274856727008097,
+      "step": 18200
+    },
+    {
+      "epoch": 2.49,
+      "learning_rate": 0.0017130183220829317,
+      "loss": 0.5001,
+      "step": 18210
+    },
+    {
+      "epoch": 2.5,
+      "learning_rate": 0.001712467281994765,
+      "loss": 0.4584,
+      "step": 18220
+    },
+    {
+      "epoch": 2.5,
+      "learning_rate": 0.0017119162419065989,
+      "loss": 0.4184,
+      "step": 18230
+    },
+    {
+      "epoch": 2.5,
+      "learning_rate": 0.0017113652018184323,
+      "loss": 0.4813,
+      "step": 18240
+    },
+    {
+      "epoch": 2.5,
+      "learning_rate": 0.0017108141617302659,
+      "loss": 0.514,
+      "step": 18250
+    },
+    {
+      "epoch": 2.5,
+      "learning_rate": 0.0017102631216420994,
+      "loss": 0.4884,
+      "step": 18260
+    },
+    {
+      "epoch": 2.5,
+      "learning_rate": 0.001709712081553933,
+      "loss": 0.4582,
+      "step": 18270
+    },
+    {
+      "epoch": 2.5,
+      "learning_rate": 0.0017091610414657666,
+      "loss": 0.5004,
+      "step": 18280
+    },
+    {
+      "epoch": 2.51,
+      "learning_rate": 0.0017086100013776002,
+      "loss": 0.544,
+      "step": 18290
+    },
+    {
+      "epoch": 2.51,
+      "learning_rate": 0.001708058961289434,
+      "loss": 0.4559,
+      "step": 18300
+    },
+    {
+      "epoch": 2.51,
+      "eval_loss": 0.46133074164390564,
+      "eval_runtime": 1056.1136,
+      "eval_samples_per_second": 11.062,
+      "eval_steps_per_second": 1.383,
+      "eval_wer": 0.7437119075775493,
+      "step": 18300
+    },
+    {
+      "epoch": 2.51,
+      "learning_rate": 0.0017075079212012674,
+      "loss": 0.5014,
+      "step": 18310
+    },
+    {
+      "epoch": 2.51,
+      "learning_rate": 0.001706956881113101,
+      "loss": 0.4689,
+      "step": 18320
+    },
+    {
+      "epoch": 2.51,
+      "learning_rate": 0.0017064058410249346,
+      "loss": 0.456,
+      "step": 18330
+    },
+    {
+      "epoch": 2.51,
+      "learning_rate": 0.0017058548009367682,
+      "loss": 0.4801,
+      "step": 18340
+    },
+    {
+      "epoch": 2.51,
+      "learning_rate": 0.0017053037608486018,
+      "loss": 0.4157,
+      "step": 18350
+    },
+    {
+      "epoch": 2.52,
+      "learning_rate": 0.0017047527207604354,
+      "loss": 0.4815,
+      "step": 18360
+    },
+    {
+      "epoch": 2.52,
+      "learning_rate": 0.0017042016806722688,
+      "loss": 0.48,
+      "step": 18370
+    },
+    {
+      "epoch": 2.52,
+      "learning_rate": 0.0017036506405841026,
+      "loss": 0.5935,
+      "step": 18380
+    },
+    {
+      "epoch": 2.52,
+      "learning_rate": 0.001703099600495936,
+      "loss": 0.4989,
+      "step": 18390
+    },
+    {
+      "epoch": 2.52,
+      "learning_rate": 0.0017025485604077698,
+      "loss": 0.4168,
+      "step": 18400
+    },
+    {
+      "epoch": 2.52,
+      "eval_loss": 0.4182513654232025,
+      "eval_runtime": 1076.2361,
+      "eval_samples_per_second": 10.855,
+      "eval_steps_per_second": 1.358,
+      "eval_wer": 0.7092354225416174,
+      "step": 18400
+    },
+    {
+      "epoch": 2.52,
+      "learning_rate": 0.0017019975203196032,
+      "loss": 0.4833,
+      "step": 18410
+    },
+    {
+      "epoch": 2.52,
+      "learning_rate": 0.001701446480231437,
+      "loss": 0.416,
+      "step": 18420
+    },
+    {
+      "epoch": 2.53,
+      "learning_rate": 0.0017008954401432704,
+      "loss": 0.4896,
+      "step": 18430
+    },
+    {
+      "epoch": 2.53,
+      "learning_rate": 0.001700344400055104,
+      "loss": 0.4726,
+      "step": 18440
+    },
+    {
+      "epoch": 2.53,
+      "learning_rate": 0.0016997933599669378,
+      "loss": 0.4152,
+      "step": 18450
+    },
+    {
+      "epoch": 2.53,
+      "learning_rate": 0.0016992423198787712,
+      "loss": 0.5045,
+      "step": 18460
+    },
+    {
+      "epoch": 2.53,
+      "learning_rate": 0.001698691279790605,
+      "loss": 0.4485,
+      "step": 18470
+    },
+    {
+      "epoch": 2.53,
+      "learning_rate": 0.0016981402397024383,
+      "loss": 0.4399,
+      "step": 18480
+    },
+    {
+      "epoch": 2.53,
+      "learning_rate": 0.001697589199614272,
+      "loss": 0.4487,
+      "step": 18490
+    },
+    {
+      "epoch": 2.53,
+      "learning_rate": 0.0016970381595261055,
+      "loss": 0.3521,
+      "step": 18500
+    },
+    {
+      "epoch": 2.53,
+      "eval_loss": 0.4058149755001068,
+      "eval_runtime": 1077.6187,
+      "eval_samples_per_second": 10.841,
+      "eval_steps_per_second": 1.356,
+      "eval_wer": 0.7055740016374056,
+      "step": 18500
+    },
+    {
+      "epoch": 2.54,
+      "learning_rate": 0.0016964871194379391,
+      "loss": 0.4788,
+      "step": 18510
+    },
+    {
+      "epoch": 2.54,
+      "learning_rate": 0.0016959360793497727,
+      "loss": 0.4587,
+      "step": 18520
+    },
+    {
+      "epoch": 2.54,
+      "learning_rate": 0.0016953850392616063,
+      "loss": 0.477,
+      "step": 18530
+    },
+    {
+      "epoch": 2.54,
+      "learning_rate": 0.00169483399917344,
+      "loss": 0.4734,
+      "step": 18540
+    },
+    {
+      "epoch": 2.54,
+      "learning_rate": 0.0016942829590852735,
+      "loss": 0.4395,
+      "step": 18550
+    },
+    {
+      "epoch": 2.54,
+      "learning_rate": 0.001693731918997107,
+      "loss": 0.5108,
+      "step": 18560
+    },
+    {
+      "epoch": 2.54,
+      "learning_rate": 0.0016931808789089407,
+      "loss": 0.4812,
+      "step": 18570
+    },
+    {
+      "epoch": 2.55,
+      "learning_rate": 0.001692629838820774,
+      "loss": 0.4657,
+      "step": 18580
+    },
+    {
+      "epoch": 2.55,
+      "learning_rate": 0.001692078798732608,
+      "loss": 0.4463,
+      "step": 18590
+    },
+    {
+      "epoch": 2.55,
+      "learning_rate": 0.0016915277586444415,
+      "loss": 0.4208,
+      "step": 18600
+    },
+    {
+      "epoch": 2.55,
+      "eval_loss": 0.4372923970222473,
+      "eval_runtime": 1121.3451,
+      "eval_samples_per_second": 10.419,
+      "eval_steps_per_second": 1.303,
+      "eval_wer": 0.7353770581278996,
+      "step": 18600
+    },
+    {
+      "epoch": 2.55,
+      "learning_rate": 0.001690976718556275,
+      "loss": 0.5681,
+      "step": 18610
+    },
+    {
+      "epoch": 2.55,
+      "learning_rate": 0.0016904256784681087,
+      "loss": 0.4643,
+      "step": 18620
+    },
+    {
+      "epoch": 2.55,
+      "learning_rate": 0.001689874638379942,
+      "loss": 0.4764,
+      "step": 18630
+    },
+    {
+      "epoch": 2.55,
+      "learning_rate": 0.0016893235982917759,
+      "loss": 0.531,
+      "step": 18640
+    },
+    {
+      "epoch": 2.56,
+      "learning_rate": 0.0016887725582036093,
+      "loss": 0.4855,
+      "step": 18650
+    },
+    {
+      "epoch": 2.56,
+      "learning_rate": 0.001688221518115443,
+      "loss": 0.5388,
+      "step": 18660
+    },
+    {
+      "epoch": 2.56,
+      "learning_rate": 0.0016876704780272765,
+      "loss": 0.4379,
+      "step": 18670
+    },
+    {
+      "epoch": 2.56,
+      "learning_rate": 0.00168711943793911,
+      "loss": 0.4793,
+      "step": 18680
+    },
+    {
+      "epoch": 2.56,
+      "learning_rate": 0.0016865683978509436,
+      "loss": 0.4609,
+      "step": 18690
+    },
+    {
+      "epoch": 2.56,
+      "learning_rate": 0.0016860173577627772,
+      "loss": 0.3918,
+      "step": 18700
+    },
+    {
+      "epoch": 2.56,
+      "eval_loss": 0.4256938099861145,
+      "eval_runtime": 1111.9535,
+      "eval_samples_per_second": 10.507,
+      "eval_steps_per_second": 1.314,
+      "eval_wer": 0.7229828072409715,
+      "step": 18700
+    },
+    {
+      "epoch": 2.56,
+      "learning_rate": 0.0016854663176746108,
+      "loss": 0.533,
+      "step": 18710
+    },
+    {
+      "epoch": 2.56,
+      "learning_rate": 0.0016849152775864444,
+      "loss": 0.487,
+      "step": 18720
+    },
+    {
+      "epoch": 2.57,
+      "learning_rate": 0.001684364237498278,
+      "loss": 0.4289,
+      "step": 18730
+    },
+    {
+      "epoch": 2.57,
+      "learning_rate": 0.0016838131974101116,
+      "loss": 0.4594,
+      "step": 18740
+    },
+    {
+      "epoch": 2.57,
+      "learning_rate": 0.001683262157321945,
+      "loss": 0.3775,
+      "step": 18750
+    },
+    {
+      "epoch": 2.57,
+      "learning_rate": 0.0016827111172337788,
+      "loss": 0.4828,
+      "step": 18760
+    },
+    {
+      "epoch": 2.57,
+      "learning_rate": 0.0016821600771456124,
+      "loss": 0.4511,
+      "step": 18770
+    },
+    {
+      "epoch": 2.57,
+      "learning_rate": 0.001681609037057446,
+      "loss": 0.4459,
+      "step": 18780
+    },
+    {
+      "epoch": 2.57,
+      "learning_rate": 0.0016810579969692796,
+      "loss": 0.4982,
+      "step": 18790
+    },
+    {
+      "epoch": 2.58,
+      "learning_rate": 0.0016805069568811132,
+      "loss": 0.3886,
+      "step": 18800
+    },
+    {
+      "epoch": 2.58,
+      "eval_loss": 0.41439610719680786,
+      "eval_runtime": 1077.6735,
+      "eval_samples_per_second": 10.841,
+      "eval_steps_per_second": 1.356,
+      "eval_wer": 0.7191621941235331,
+      "step": 18800
+    },
+    {
+      "epoch": 2.58,
+      "learning_rate": 0.0016799559167929468,
+      "loss": 0.57,
+      "step": 18810
+    },
+    {
+      "epoch": 2.58,
+      "learning_rate": 0.0016794048767047802,
+      "loss": 0.4562,
+      "step": 18820
+    },
+    {
+      "epoch": 2.58,
+      "learning_rate": 0.001678853836616614,
+      "loss": 0.4418,
+      "step": 18830
+    },
+    {
+      "epoch": 2.58,
+      "learning_rate": 0.0016783027965284474,
+      "loss": 0.4872,
+      "step": 18840
+    },
+    {
+      "epoch": 2.58,
+      "learning_rate": 0.0016777517564402812,
+      "loss": 0.4984,
+      "step": 18850
+    },
+    {
+      "epoch": 2.58,
+      "learning_rate": 0.0016772007163521146,
+      "loss": 0.4835,
+      "step": 18860
+    },
+    {
+      "epoch": 2.59,
+      "learning_rate": 0.0016766496762639482,
+      "loss": 0.4779,
+      "step": 18870
+    },
+    {
+      "epoch": 2.59,
+      "learning_rate": 0.0016760986361757818,
+      "loss": 0.4548,
+      "step": 18880
+    },
+    {
+      "epoch": 2.59,
+      "learning_rate": 0.0016755475960876154,
+      "loss": 0.4794,
+      "step": 18890
+    },
+    {
+      "epoch": 2.59,
+      "learning_rate": 0.001674996555999449,
+      "loss": 0.5022,
+      "step": 18900
+    },
+    {
+      "epoch": 2.59,
+      "eval_loss": 0.3865416646003723,
+      "eval_runtime": 1077.7105,
+      "eval_samples_per_second": 10.841,
+      "eval_steps_per_second": 1.356,
+      "eval_wer": 0.6997293732375148,
+      "step": 18900
+    },
+    {
+      "epoch": 2.59,
+      "learning_rate": 0.0016744455159112825,
+      "loss": 0.485,
+      "step": 18910
+    },
+    {
+      "epoch": 2.59,
+      "learning_rate": 0.0016738944758231164,
+      "loss": 0.4502,
+      "step": 18920
+    },
+    {
+      "epoch": 2.59,
+      "learning_rate": 0.0016733434357349497,
+      "loss": 0.4419,
+      "step": 18930
+    },
+    {
+      "epoch": 2.59,
+      "learning_rate": 0.0016727923956467833,
+      "loss": 0.4835,
+      "step": 18940
+    },
+    {
+      "epoch": 2.6,
+      "learning_rate": 0.001672241355558617,
+      "loss": 0.468,
+      "step": 18950
+    },
+    {
+      "epoch": 2.6,
+      "learning_rate": 0.0016716903154704505,
+      "loss": 0.4594,
+      "step": 18960
+    },
+    {
+      "epoch": 2.6,
+      "learning_rate": 0.0016711392753822841,
+      "loss": 0.4318,
+      "step": 18970
+    },
+    {
+      "epoch": 2.6,
+      "learning_rate": 0.0016705882352941177,
+      "loss": 0.45,
+      "step": 18980
+    },
+    {
+      "epoch": 2.6,
+      "learning_rate": 0.0016700371952059513,
+      "loss": 0.4783,
+      "step": 18990
+    },
+    {
+      "epoch": 2.6,
+      "learning_rate": 0.001669486155117785,
+      "loss": 0.4236,
+      "step": 19000
+    },
+    {
+      "epoch": 2.6,
+      "eval_loss": 0.40951940417289734,
+      "eval_runtime": 1077.4216,
+      "eval_samples_per_second": 10.843,
+      "eval_steps_per_second": 1.356,
+      "eval_wer": 0.7155348858364414,
+      "step": 19000
+    },
+    {
+      "epoch": 2.6,
+      "learning_rate": 0.0016689351150296183,
+      "loss": 0.452,
+      "step": 19010
+    },
+    {
+      "epoch": 2.61,
+      "learning_rate": 0.001668384074941452,
+      "loss": 0.7663,
+      "step": 19020
+    },
+    {
+      "epoch": 2.61,
+      "learning_rate": 0.0016678330348532855,
+      "loss": 0.4473,
+      "step": 19030
+    },
+    {
+      "epoch": 2.61,
+      "learning_rate": 0.0016672819947651193,
+      "loss": 0.4585,
+      "step": 19040
+    },
+    {
+      "epoch": 2.61,
+      "learning_rate": 0.0016667309546769527,
+      "loss": 0.4386,
+      "step": 19050
+    },
+    {
+      "epoch": 2.61,
+      "learning_rate": 0.0016661799145887863,
+      "loss": 0.5152,
+      "step": 19060
+    },
+    {
+      "epoch": 2.61,
+      "learning_rate": 0.00166562887450062,
+      "loss": 0.4124,
+      "step": 19070
+    },
+    {
+      "epoch": 2.61,
+      "learning_rate": 0.0016650778344124535,
+      "loss": 0.4615,
+      "step": 19080
+    },
+    {
+      "epoch": 2.62,
+      "learning_rate": 0.0016645267943242873,
+      "loss": 0.4855,
+      "step": 19090
+    },
+    {
+      "epoch": 2.62,
+      "learning_rate": 0.0016639757542361207,
+      "loss": 0.3934,
+      "step": 19100
+    },
+    {
+      "epoch": 2.62,
+      "eval_loss": 0.41373133659362793,
+      "eval_runtime": 1079.2652,
+      "eval_samples_per_second": 10.825,
+      "eval_steps_per_second": 1.354,
+      "eval_wer": 0.7208678249795324,
+      "step": 19100
+    },
+    {
+      "epoch": 2.62,
+      "learning_rate": 0.0016634247141479545,
+      "loss": 0.5345,
+      "step": 19110
+    },
+    {
+      "epoch": 2.62,
+      "learning_rate": 0.0016628736740597878,
+      "loss": 0.4635,
+      "step": 19120
+    },
+    {
+      "epoch": 2.62,
+      "learning_rate": 0.0016623226339716214,
+      "loss": 0.393,
+      "step": 19130
+    },
+    {
+      "epoch": 2.62,
+      "learning_rate": 0.001661771593883455,
+      "loss": 0.4539,
+      "step": 19140
+    },
+    {
+      "epoch": 2.62,
+      "learning_rate": 0.0016612205537952886,
+      "loss": 0.4016,
+      "step": 19150
+    },
+    {
+      "epoch": 2.63,
+      "learning_rate": 0.0016606695137071222,
+      "loss": 0.4623,
+      "step": 19160
+    },
+    {
+      "epoch": 2.63,
+      "learning_rate": 0.0016601184736189558,
+      "loss": 0.447,
+      "step": 19170
+    },
+    {
+      "epoch": 2.63,
+      "learning_rate": 0.0016595674335307894,
+      "loss": 0.4317,
+      "step": 19180
+    },
+    {
+      "epoch": 2.63,
+      "learning_rate": 0.001659016393442623,
+      "loss": 0.3862,
+      "step": 19190
+    },
+    {
+      "epoch": 2.63,
+      "learning_rate": 0.0016584653533544564,
+      "loss": 0.4274,
+      "step": 19200
+    },
+    {
+      "epoch": 2.63,
+      "eval_loss": 0.3846190571784973,
+      "eval_runtime": 1079.2259,
+      "eval_samples_per_second": 10.825,
+      "eval_steps_per_second": 1.354,
+      "eval_wer": 0.6821841171654689,
+      "step": 19200
+    },
+    {
+      "epoch": 2.63,
+      "learning_rate": 0.0016579143132662902,
+      "loss": 0.4518,
+      "step": 19210
+    },
+    {
+      "epoch": 2.63,
+      "learning_rate": 0.0016573632731781238,
+      "loss": 0.4569,
+      "step": 19220
+    },
+    {
+      "epoch": 2.63,
+      "learning_rate": 0.0016568122330899574,
+      "loss": 0.3971,
+      "step": 19230
+    },
+    {
+      "epoch": 2.64,
+      "learning_rate": 0.001656261193001791,
+      "loss": 0.4855,
+      "step": 19240
+    },
+    {
+      "epoch": 2.64,
+      "learning_rate": 0.0016557101529136244,
+      "loss": 0.35,
+      "step": 19250
+    },
+    {
+      "epoch": 2.64,
+      "learning_rate": 0.0016551591128254582,
+      "loss": 0.4922,
+      "step": 19260
+    },
+    {
+      "epoch": 2.64,
+      "learning_rate": 0.0016546080727372916,
+      "loss": 0.4174,
+      "step": 19270
+    },
+    {
+      "epoch": 2.64,
+      "learning_rate": 0.0016540570326491254,
+      "loss": 0.4181,
+      "step": 19280
+    },
+    {
+      "epoch": 2.64,
+      "learning_rate": 0.0016535059925609588,
+      "loss": 0.4532,
+      "step": 19290
+    },
+    {
+      "epoch": 2.64,
+      "learning_rate": 0.0016529549524727926,
+      "loss": 0.3991,
+      "step": 19300
+    },
+    {
+      "epoch": 2.64,
+      "eval_loss": 0.37318894267082214,
+      "eval_runtime": 1080.6107,
+      "eval_samples_per_second": 10.811,
+      "eval_steps_per_second": 1.352,
+      "eval_wer": 0.6723028290730465,
+      "step": 19300
+    },
+    {
+      "epoch": 2.65,
+      "learning_rate": 0.001652403912384626,
+      "loss": 0.441,
+      "step": 19310
+    },
+    {
+      "epoch": 2.65,
+      "learning_rate": 0.0016518528722964596,
+      "loss": 0.4311,
+      "step": 19320
+    },
+    {
+      "epoch": 2.65,
+      "learning_rate": 0.0016513018322082931,
+      "loss": 0.5161,
+      "step": 19330
+    },
+    {
+      "epoch": 2.65,
+      "learning_rate": 0.0016507507921201267,
+      "loss": 0.4995,
+      "step": 19340
+    },
+    {
+      "epoch": 2.65,
+      "learning_rate": 0.0016501997520319603,
+      "loss": 0.4335,
+      "step": 19350
+    },
+    {
+      "epoch": 2.65,
+      "learning_rate": 0.001649648711943794,
+      "loss": 0.4368,
+      "step": 19360
+    },
+    {
+      "epoch": 2.65,
+      "learning_rate": 0.0016490976718556275,
+      "loss": 0.4128,
+      "step": 19370
+    },
+    {
+      "epoch": 2.66,
+      "learning_rate": 0.0016485466317674611,
+      "loss": 0.4289,
+      "step": 19380
+    },
+    {
+      "epoch": 2.66,
+      "learning_rate": 0.0016479955916792947,
+      "loss": 0.4784,
+      "step": 19390
+    },
+    {
+      "epoch": 2.66,
+      "learning_rate": 0.0016474445515911283,
+      "loss": 0.4125,
+      "step": 19400
+    },
+    {
+      "epoch": 2.66,
+      "eval_loss": 0.4049933850765228,
+      "eval_runtime": 1061.7449,
+      "eval_samples_per_second": 11.004,
+      "eval_steps_per_second": 1.376,
+      "eval_wer": 0.6991267170017283,
+      "step": 19400
+    },
+    {
+      "epoch": 2.66,
+      "learning_rate": 0.001646893511502962,
+      "loss": 0.5169,
+      "step": 19410
+    },
+    {
+      "epoch": 2.66,
+      "learning_rate": 0.0016463424714147955,
+      "loss": 0.4409,
+      "step": 19420
+    },
+    {
+      "epoch": 2.66,
+      "learning_rate": 0.001645791431326629,
+      "loss": 0.4307,
+      "step": 19430
+    },
+    {
+      "epoch": 2.66,
+      "learning_rate": 0.0016452403912384625,
+      "loss": 0.4878,
+      "step": 19440
+    },
+    {
+      "epoch": 2.66,
+      "learning_rate": 0.0016446893511502963,
+      "loss": 0.3275,
+      "step": 19450
+    },
+    {
+      "epoch": 2.67,
+      "learning_rate": 0.0016441383110621297,
+      "loss": 0.5385,
+      "step": 19460
+    },
+    {
+      "epoch": 2.67,
+      "learning_rate": 0.0016435872709739635,
+      "loss": 0.4699,
+      "step": 19470
+    },
+    {
+      "epoch": 2.67,
+      "learning_rate": 0.0016430362308857969,
+      "loss": 0.4564,
+      "step": 19480
+    },
+    {
+      "epoch": 2.67,
+      "learning_rate": 0.0016424851907976307,
+      "loss": 0.4872,
+      "step": 19490
+    },
+    {
+      "epoch": 2.67,
+      "learning_rate": 0.001641934150709464,
+      "loss": 0.3494,
+      "step": 19500
+    },
+    {
+      "epoch": 2.67,
+      "eval_loss": 0.3713689148426056,
+      "eval_runtime": 1055.3116,
+      "eval_samples_per_second": 11.071,
+      "eval_steps_per_second": 1.384,
+      "eval_wer": 0.6873692349677067,
+      "step": 19500
+    },
+    {
+      "epoch": 2.67,
+      "learning_rate": 0.0016413831106212977,
+      "loss": 0.4652,
+      "step": 19510
+    },
+    {
+      "epoch": 2.67,
+      "learning_rate": 0.0016408320705331313,
+      "loss": 0.3952,
+      "step": 19520
+    },
+    {
+      "epoch": 2.68,
+      "learning_rate": 0.0016402810304449649,
+      "loss": 0.428,
+      "step": 19530
+    },
+    {
+      "epoch": 2.68,
+      "learning_rate": 0.0016397299903567987,
+      "loss": 0.4589,
+      "step": 19540
+    },
+    {
+      "epoch": 2.68,
+      "learning_rate": 0.001639178950268632,
+      "loss": 0.3793,
+      "step": 19550
+    },
+    {
+      "epoch": 2.68,
+      "learning_rate": 0.0016386279101804659,
+      "loss": 0.4458,
+      "step": 19560
+    },
+    {
+      "epoch": 2.68,
+      "learning_rate": 0.0016380768700922992,
+      "loss": 0.4573,
+      "step": 19570
+    },
+    {
+      "epoch": 2.68,
+      "learning_rate": 0.0016375258300041328,
+      "loss": 0.4676,
+      "step": 19580
+    },
+    {
+      "epoch": 2.68,
+      "learning_rate": 0.0016369747899159664,
+      "loss": 0.4153,
+      "step": 19590
+    },
+    {
+      "epoch": 2.69,
+      "learning_rate": 0.0016364237498278,
+      "loss": 0.4416,
+      "step": 19600
+    },
+    {
+      "epoch": 2.69,
+      "eval_loss": 0.37622639536857605,
+      "eval_runtime": 1065.7745,
+      "eval_samples_per_second": 10.962,
+      "eval_steps_per_second": 1.371,
+      "eval_wer": 0.684162648958428,
+      "step": 19600
+    },
+    {
+      "epoch": 2.69,
+      "learning_rate": 0.0016358727097396336,
+      "loss": 0.4389,
+      "step": 19610
+    },
+    {
+      "epoch": 2.69,
+      "learning_rate": 0.0016353216696514672,
+      "loss": 0.4624,
+      "step": 19620
+    },
+    {
+      "epoch": 2.69,
+      "learning_rate": 0.0016347706295633006,
+      "loss": 0.4273,
+      "step": 19630
+    },
+    {
+      "epoch": 2.69,
+      "learning_rate": 0.0016342195894751344,
+      "loss": 0.4338,
+      "step": 19640
+    },
+    {
+      "epoch": 2.69,
+      "learning_rate": 0.0016336685493869678,
+      "loss": 0.4413,
+      "step": 19650
+    },
+    {
+      "epoch": 2.69,
+      "learning_rate": 0.0016331175092988016,
+      "loss": 0.4636,
+      "step": 19660
+    },
+    {
+      "epoch": 2.69,
+      "learning_rate": 0.001632566469210635,
+      "loss": 0.4253,
+      "step": 19670
+    },
+    {
+      "epoch": 2.7,
+      "learning_rate": 0.0016320154291224688,
+      "loss": 0.4218,
+      "step": 19680
+    },
+    {
+      "epoch": 2.7,
+      "learning_rate": 0.0016314643890343024,
+      "loss": 0.4508,
+      "step": 19690
+    },
+    {
+      "epoch": 2.7,
+      "learning_rate": 0.0016309133489461358,
+      "loss": 0.397,
+      "step": 19700
+    },
+    {
+      "epoch": 2.7,
+      "eval_loss": 0.37222862243652344,
+      "eval_runtime": 1055.777,
+      "eval_samples_per_second": 11.066,
+      "eval_steps_per_second": 1.384,
+      "eval_wer": 0.682115891931229,
+      "step": 19700
+    },
+    {
+      "epoch": 2.7,
+      "learning_rate": 0.0016303623088579696,
+      "loss": 0.4811,
+      "step": 19710
+    },
+    {
+      "epoch": 2.7,
+      "learning_rate": 0.001629811268769803,
+      "loss": 0.4314,
+      "step": 19720
+    },
+    {
+      "epoch": 2.7,
+      "learning_rate": 0.0016292602286816368,
+      "loss": 0.4363,
+      "step": 19730
+    },
+    {
+      "epoch": 2.7,
+      "learning_rate": 0.0016287091885934702,
+      "loss": 0.4375,
+      "step": 19740
+    },
+    {
+      "epoch": 2.71,
+      "learning_rate": 0.001628158148505304,
+      "loss": 0.397,
+      "step": 19750
+    },
+    {
+      "epoch": 2.71,
+      "learning_rate": 0.0016276071084171373,
+      "loss": 0.4463,
+      "step": 19760
+    },
+    {
+      "epoch": 2.71,
+      "learning_rate": 0.001627056068328971,
+      "loss": 0.4658,
+      "step": 19770
+    },
+    {
+      "epoch": 2.71,
+      "learning_rate": 0.0016265050282408045,
+      "loss": 0.4143,
+      "step": 19780
+    },
+    {
+      "epoch": 2.71,
+      "learning_rate": 0.0016259539881526381,
+      "loss": 0.4589,
+      "step": 19790
+    },
+    {
+      "epoch": 2.71,
+      "learning_rate": 0.0016254029480644717,
+      "loss": 0.4796,
+      "step": 19800
+    },
+    {
+      "epoch": 2.71,
+      "eval_loss": 0.3822312355041504,
+      "eval_runtime": 1055.0013,
+      "eval_samples_per_second": 11.074,
+      "eval_steps_per_second": 1.385,
+      "eval_wer": 0.6959315018648231,
+      "step": 19800
+    },
+    {
+      "epoch": 2.71,
+      "learning_rate": 0.0016248519079763053,
+      "loss": 0.4656,
+      "step": 19810
+    },
+    {
+      "epoch": 2.72,
+      "learning_rate": 0.0016243008678881387,
+      "loss": 0.4431,
+      "step": 19820
+    },
+    {
+      "epoch": 2.72,
+      "learning_rate": 0.0016237498277999725,
+      "loss": 0.428,
+      "step": 19830
+    },
+    {
+      "epoch": 2.72,
+      "learning_rate": 0.001623198787711806,
+      "loss": 0.4491,
+      "step": 19840
+    },
+    {
+      "epoch": 2.72,
+      "learning_rate": 0.0016226477476236397,
+      "loss": 0.3989,
+      "step": 19850
+    },
+    {
+      "epoch": 2.72,
+      "learning_rate": 0.0016220967075354733,
+      "loss": 0.4572,
+      "step": 19860
+    },
+    {
+      "epoch": 2.72,
+      "learning_rate": 0.001621545667447307,
+      "loss": 0.4338,
+      "step": 19870
+    },
+    {
+      "epoch": 2.72,
+      "learning_rate": 0.0016209946273591405,
+      "loss": 0.4743,
+      "step": 19880
+    },
+    {
+      "epoch": 2.73,
+      "learning_rate": 0.0016204435872709739,
+      "loss": 0.4287,
+      "step": 19890
+    },
+    {
+      "epoch": 2.73,
+      "learning_rate": 0.0016198925471828077,
+      "loss": 0.3637,
+      "step": 19900
+    },
+    {
+      "epoch": 2.73,
+      "eval_loss": 0.3864554464817047,
+      "eval_runtime": 1055.7923,
+      "eval_samples_per_second": 11.066,
+      "eval_steps_per_second": 1.384,
+      "eval_wer": 0.6979214045301555,
+      "step": 19900
+    },
+    {
+      "epoch": 2.73,
+      "learning_rate": 0.001619341507094641,
+      "loss": 0.5414,
+      "step": 19910
+    },
+    {
+      "epoch": 2.73,
+      "learning_rate": 0.0016187904670064749,
+      "loss": 0.468,
+      "step": 19920
+    },
+    {
+      "epoch": 2.73,
+      "learning_rate": 0.0016182394269183083,
+      "loss": 0.4326,
+      "step": 19930
+    },
+    {
+      "epoch": 2.73,
+      "learning_rate": 0.0016176883868301419,
+      "loss": 0.4912,
+      "step": 19940
+    },
+    {
+      "epoch": 2.73,
+      "learning_rate": 0.0016171373467419755,
+      "loss": 0.4261,
+      "step": 19950
+    },
+    {
+      "epoch": 2.73,
+      "learning_rate": 0.001616586306653809,
+      "loss": 0.4505,
+      "step": 19960
+    },
+    {
+      "epoch": 2.74,
+      "learning_rate": 0.0016160352665656426,
+      "loss": 0.4708,
+      "step": 19970
+    },
+    {
+      "epoch": 2.74,
+      "learning_rate": 0.0016154842264774762,
+      "loss": 0.4064,
+      "step": 19980
+    },
+    {
+      "epoch": 2.74,
+      "learning_rate": 0.0016149331863893098,
+      "loss": 0.4524,
+      "step": 19990
+    },
+    {
+      "epoch": 2.74,
+      "learning_rate": 0.0016143821463011434,
+      "loss": 0.4456,
+      "step": 20000
+    },
+    {
+      "epoch": 2.74,
+      "eval_loss": 0.39026540517807007,
+      "eval_runtime": 1055.8745,
+      "eval_samples_per_second": 11.065,
+      "eval_steps_per_second": 1.384,
+      "eval_wer": 0.7037774038024197,
+      "step": 20000
+    },
+    {
+      "epoch": 2.74,
+      "learning_rate": 0.001613831106212977,
+      "loss": 0.484,
+      "step": 20010
+    },
+    {
+      "epoch": 2.74,
+      "learning_rate": 0.0016132800661248106,
+      "loss": 0.4055,
+      "step": 20020
+    },
+    {
+      "epoch": 2.74,
+      "learning_rate": 0.0016127290260366442,
+      "loss": 0.4346,
+      "step": 20030
+    },
+    {
+      "epoch": 2.75,
+      "learning_rate": 0.0016121779859484778,
+      "loss": 0.4343,
+      "step": 20040
+    },
+    {
+      "epoch": 2.75,
+      "learning_rate": 0.0016116269458603114,
+      "loss": 0.3801,
+      "step": 20050
+    },
+    {
+      "epoch": 2.75,
+      "learning_rate": 0.001611075905772145,
+      "loss": 0.4218,
+      "step": 20060
+    },
+    {
+      "epoch": 2.75,
+      "learning_rate": 0.0016105248656839786,
+      "loss": 0.4118,
+      "step": 20070
+    },
+    {
+      "epoch": 2.75,
+      "learning_rate": 0.001609973825595812,
+      "loss": 0.4138,
+      "step": 20080
+    },
+    {
+      "epoch": 2.75,
+      "learning_rate": 0.0016094227855076458,
+      "loss": 0.4119,
+      "step": 20090
+    },
+    {
+      "epoch": 2.75,
+      "learning_rate": 0.0016088717454194792,
+      "loss": 0.3338,
+      "step": 20100
+    },
+    {
+      "epoch": 2.75,
+      "eval_loss": 0.3513972759246826,
+      "eval_runtime": 1054.5365,
+      "eval_samples_per_second": 11.079,
+      "eval_steps_per_second": 1.385,
+      "eval_wer": 0.6548599108523606,
+      "step": 20100
+    },
+    {
+      "epoch": 2.76,
+      "learning_rate": 0.001608320705331313,
+      "loss": 0.426,
+      "step": 20110
+    },
+    {
+      "epoch": 2.76,
+      "learning_rate": 0.0016077696652431464,
+      "loss": 0.5807,
+      "step": 20120
+    },
+    {
+      "epoch": 2.76,
+      "learning_rate": 0.00160721862515498,
+      "loss": 0.4345,
+      "step": 20130
+    },
+    {
+      "epoch": 2.76,
+      "learning_rate": 0.0016066675850668136,
+      "loss": 0.4679,
+      "step": 20140
+    },
+    {
+      "epoch": 2.76,
+      "learning_rate": 0.0016061165449786472,
+      "loss": 0.4381,
+      "step": 20150
+    },
+    {
+      "epoch": 2.76,
+      "learning_rate": 0.001605565504890481,
+      "loss": 0.4752,
+      "step": 20160
+    },
+    {
+      "epoch": 2.76,
+      "learning_rate": 0.0016050144648023144,
+      "loss": 0.4445,
+      "step": 20170
+    },
+    {
+      "epoch": 2.76,
+      "learning_rate": 0.0016044634247141482,
+      "loss": 0.4492,
+      "step": 20180
+    },
+    {
+      "epoch": 2.77,
+      "learning_rate": 0.0016039123846259815,
+      "loss": 0.4185,
+      "step": 20190
+    },
+    {
+      "epoch": 2.77,
+      "learning_rate": 0.0016033613445378151,
+      "loss": 0.3541,
+      "step": 20200
+    },
+    {
+      "epoch": 2.77,
+      "eval_loss": 0.3696558177471161,
+      "eval_runtime": 1056.1721,
+      "eval_samples_per_second": 11.062,
+      "eval_steps_per_second": 1.383,
+      "eval_wer": 0.678170199217684,
+      "step": 20200
+    },
+    {
+      "epoch": 2.77,
+      "learning_rate": 0.0016028103044496487,
+      "loss": 0.4462,
+      "step": 20210
+    },
+    {
+      "epoch": 2.77,
+      "learning_rate": 0.0016022592643614823,
+      "loss": 0.4229,
+      "step": 20220
+    },
+    {
+      "epoch": 2.77,
+      "learning_rate": 0.001601708224273316,
+      "loss": 0.4437,
+      "step": 20230
+    },
+    {
+      "epoch": 2.77,
+      "learning_rate": 0.0016011571841851495,
+      "loss": 0.3853,
+      "step": 20240
+    },
+    {
+      "epoch": 2.77,
+      "learning_rate": 0.0016006061440969831,
+      "loss": 0.4451,
+      "step": 20250
+    },
+    {
+      "epoch": 2.78,
+      "learning_rate": 0.0016000551040088167,
+      "loss": 0.4434,
+      "step": 20260
+    },
+    {
+      "epoch": 2.78,
+      "learning_rate": 0.00159950406392065,
+      "loss": 0.4604,
+      "step": 20270
+    },
+    {
+      "epoch": 2.78,
+      "learning_rate": 0.001598953023832484,
+      "loss": 0.4426,
+      "step": 20280
+    },
+    {
+      "epoch": 2.78,
+      "learning_rate": 0.0015984019837443173,
+      "loss": 0.4068,
+      "step": 20290
+    },
+    {
+      "epoch": 2.78,
+      "learning_rate": 0.001597850943656151,
+      "loss": 0.4417,
+      "step": 20300
+    },
+    {
+      "epoch": 2.78,
+      "eval_loss": 0.36540818214416504,
+      "eval_runtime": 1053.9323,
+      "eval_samples_per_second": 11.085,
+      "eval_steps_per_second": 1.386,
+      "eval_wer": 0.6810356590557628,
+      "step": 20300
+    },
+    {
+      "epoch": 2.78,
+      "learning_rate": 0.0015972999035679847,
+      "loss": 0.4574,
+      "step": 20310
+    },
+    {
+      "epoch": 2.78,
+      "learning_rate": 0.001596748863479818,
+      "loss": 0.4448,
+      "step": 20320
+    },
+    {
+      "epoch": 2.79,
+      "learning_rate": 0.0015961978233916519,
+      "loss": 0.4009,
+      "step": 20330
+    },
+    {
+      "epoch": 2.79,
+      "learning_rate": 0.0015956467833034853,
+      "loss": 0.4218,
+      "step": 20340
+    },
+    {
+      "epoch": 2.79,
+      "learning_rate": 0.001595095743215319,
+      "loss": 0.4281,
+      "step": 20350
+    },
+    {
+      "epoch": 2.79,
+      "learning_rate": 0.0015945447031271525,
+      "loss": 0.4509,
+      "step": 20360
+    },
+    {
+      "epoch": 2.79,
+      "learning_rate": 0.0015939936630389863,
+      "loss": 0.3913,
+      "step": 20370
+    },
+    {
+      "epoch": 2.79,
+      "learning_rate": 0.0015934426229508197,
+      "loss": 0.4214,
+      "step": 20380
+    },
+    {
+      "epoch": 2.79,
+      "learning_rate": 0.0015928915828626532,
+      "loss": 0.4138,
+      "step": 20390
+    },
+    {
+      "epoch": 2.79,
+      "learning_rate": 0.0015923405427744868,
+      "loss": 0.3682,
+      "step": 20400
+    },
+    {
+      "epoch": 2.79,
+      "eval_loss": 0.3626611828804016,
+      "eval_runtime": 1053.9016,
+      "eval_samples_per_second": 11.085,
+      "eval_steps_per_second": 1.386,
+      "eval_wer": 0.6612844537432911,
+      "step": 20400
+    },
+    {
+      "epoch": 2.8,
+      "learning_rate": 0.0015917895026863204,
+      "loss": 0.4199,
+      "step": 20410
+    },
+    {
+      "epoch": 2.8,
+      "learning_rate": 0.001591238462598154,
+      "loss": 0.4418,
+      "step": 20420
+    },
+    {
+      "epoch": 2.8,
+      "learning_rate": 0.0015906874225099876,
+      "loss": 0.3738,
+      "step": 20430
+    },
+    {
+      "epoch": 2.8,
+      "learning_rate": 0.0015901363824218212,
+      "loss": 0.4465,
+      "step": 20440
+    },
+    {
+      "epoch": 2.8,
+      "learning_rate": 0.0015895853423336548,
+      "loss": 0.334,
+      "step": 20450
+    },
+    {
+      "epoch": 2.8,
+      "learning_rate": 0.0015890343022454882,
+      "loss": 0.4826,
+      "step": 20460
+    },
+    {
+      "epoch": 2.8,
+      "learning_rate": 0.001588483262157322,
+      "loss": 0.47,
+      "step": 20470
+    },
+    {
+      "epoch": 2.81,
+      "learning_rate": 0.0015879322220691556,
+      "loss": 0.3618,
+      "step": 20480
+    },
+    {
+      "epoch": 2.81,
+      "learning_rate": 0.0015873811819809892,
+      "loss": 0.4534,
+      "step": 20490
+    },
+    {
+      "epoch": 2.81,
+      "learning_rate": 0.0015868852459016394,
+      "loss": 0.3196,
+      "step": 20500
+    },
+    {
+      "epoch": 2.81,
+      "eval_loss": 0.3464365005493164,
+      "eval_runtime": 1058.5615,
+      "eval_samples_per_second": 11.037,
+      "eval_steps_per_second": 1.38,
+      "eval_wer": 0.6649572455198763,
+      "step": 20500
+    },
+    {
+      "epoch": 2.81,
+      "learning_rate": 0.0015863342058134728,
+      "loss": 0.4225,
+      "step": 20510
+    },
+    {
+      "epoch": 2.81,
+      "learning_rate": 0.0015857831657253066,
+      "loss": 0.5448,
+      "step": 20520
+    },
+    {
+      "epoch": 2.81,
+      "learning_rate": 0.00158523212563714,
+      "loss": 0.4192,
+      "step": 20530
+    },
+    {
+      "epoch": 2.81,
+      "learning_rate": 0.0015846810855489736,
+      "loss": 0.405,
+      "step": 20540
+    },
+    {
+      "epoch": 2.82,
+      "learning_rate": 0.0015841300454608074,
+      "loss": 0.3801,
+      "step": 20550
+    },
+    {
+      "epoch": 2.82,
+      "learning_rate": 0.0015835790053726408,
+      "loss": 0.4334,
+      "step": 20560
+    },
+    {
+      "epoch": 2.82,
+      "learning_rate": 0.0015830279652844746,
+      "loss": 0.3815,
+      "step": 20570
+    },
+    {
+      "epoch": 2.82,
+      "learning_rate": 0.001582476925196308,
+      "loss": 0.3965,
+      "step": 20580
+    },
+    {
+      "epoch": 2.82,
+      "learning_rate": 0.0015819258851081418,
+      "loss": 0.3885,
+      "step": 20590
+    },
+    {
+      "epoch": 2.82,
+      "learning_rate": 0.0015813748450199752,
+      "loss": 0.3729,
+      "step": 20600
+    },
+    {
+      "epoch": 2.82,
+      "eval_loss": 0.3278181552886963,
+      "eval_runtime": 1055.9112,
+      "eval_samples_per_second": 11.064,
+      "eval_steps_per_second": 1.384,
+      "eval_wer": 0.6508459929045757,
+      "step": 20600
+    },
+    {
+      "epoch": 2.82,
+      "learning_rate": 0.0015808238049318088,
+      "loss": 0.4336,
+      "step": 20610
+    },
+    {
+      "epoch": 2.83,
+      "learning_rate": 0.0015802727648436424,
+      "loss": 0.4096,
+      "step": 20620
+    },
+    {
+      "epoch": 2.83,
+      "learning_rate": 0.001579721724755476,
+      "loss": 0.3993,
+      "step": 20630
+    },
+    {
+      "epoch": 2.83,
+      "learning_rate": 0.0015791706846673096,
+      "loss": 0.5915,
+      "step": 20640
+    },
+    {
+      "epoch": 2.83,
+      "learning_rate": 0.0015786196445791432,
+      "loss": 0.3481,
+      "step": 20650
+    },
+    {
+      "epoch": 2.83,
+      "learning_rate": 0.0015780686044909768,
+      "loss": 0.4306,
+      "step": 20660
+    },
+    {
+      "epoch": 2.83,
+      "learning_rate": 0.0015775175644028104,
+      "loss": 0.4459,
+      "step": 20670
+    },
+    {
+      "epoch": 2.83,
+      "learning_rate": 0.0015769665243146437,
+      "loss": 0.392,
+      "step": 20680
+    },
+    {
+      "epoch": 2.83,
+      "learning_rate": 0.0015764154842264776,
+      "loss": 0.4154,
+      "step": 20690
+    },
+    {
+      "epoch": 2.84,
+      "learning_rate": 0.0015758644441383112,
+      "loss": 0.3838,
+      "step": 20700
+    },
+    {
+      "epoch": 2.84,
+      "eval_loss": 0.3434087932109833,
+      "eval_runtime": 1061.8801,
+      "eval_samples_per_second": 11.002,
+      "eval_steps_per_second": 1.376,
+      "eval_wer": 0.6575548076048394,
+      "step": 20700
+    },
+    {
+      "epoch": 2.84,
+      "learning_rate": 0.0015753134040501447,
+      "loss": 0.4283,
+      "step": 20710
+    },
+    {
+      "epoch": 2.84,
+      "learning_rate": 0.0015747623639619783,
+      "loss": 1.0957,
+      "step": 20720
+    },
+    {
+      "epoch": 2.84,
+      "learning_rate": 0.0015742113238738117,
+      "loss": 0.4283,
+      "step": 20730
+    },
+    {
+      "epoch": 2.84,
+      "learning_rate": 0.0015736602837856455,
+      "loss": 0.4244,
+      "step": 20740
+    },
+    {
+      "epoch": 2.84,
+      "learning_rate": 0.001573109243697479,
+      "loss": 0.4558,
+      "step": 20750
+    },
+    {
+      "epoch": 2.84,
+      "learning_rate": 0.0015725582036093127,
+      "loss": 0.4625,
+      "step": 20760
+    },
+    {
+      "epoch": 2.85,
+      "learning_rate": 0.001572007163521146,
+      "loss": 0.4495,
+      "step": 20770
+    },
+    {
+      "epoch": 2.85,
+      "learning_rate": 0.00157145612343298,
+      "loss": 0.4193,
+      "step": 20780
+    },
+    {
+      "epoch": 2.85,
+      "learning_rate": 0.0015709050833448133,
+      "loss": 0.3851,
+      "step": 20790
+    },
+    {
+      "epoch": 2.85,
+      "learning_rate": 0.001570354043256647,
+      "loss": 0.3506,
+      "step": 20800
+    },
+    {
+      "epoch": 2.85,
+      "eval_loss": 0.3494613766670227,
+      "eval_runtime": 1055.7531,
+      "eval_samples_per_second": 11.066,
+      "eval_steps_per_second": 1.384,
+      "eval_wer": 0.6607386518693714,
+      "step": 20800
+    },
+    {
+      "epoch": 2.85,
+      "learning_rate": 0.0015698030031684805,
+      "loss": 0.4661,
+      "step": 20810
+    },
+    {
+      "epoch": 2.85,
+      "learning_rate": 0.001569251963080314,
+      "loss": 0.4263,
+      "step": 20820
+    },
+    {
+      "epoch": 2.85,
+      "learning_rate": 0.0015687009229921477,
+      "loss": 0.4223,
+      "step": 20830
+    },
+    {
+      "epoch": 2.86,
+      "learning_rate": 0.0015681498829039813,
+      "loss": 0.4374,
+      "step": 20840
+    },
+    {
+      "epoch": 2.86,
+      "learning_rate": 0.0015675988428158149,
+      "loss": 0.3702,
+      "step": 20850
+    },
+    {
+      "epoch": 2.86,
+      "learning_rate": 0.0015670478027276485,
+      "loss": 0.4727,
+      "step": 20860
+    },
+    {
+      "epoch": 2.86,
+      "learning_rate": 0.001566496762639482,
+      "loss": 0.413,
+      "step": 20870
+    },
+    {
+      "epoch": 2.86,
+      "learning_rate": 0.0015659457225513157,
+      "loss": 0.3899,
+      "step": 20880
+    },
+    {
+      "epoch": 2.86,
+      "learning_rate": 0.0015653946824631493,
+      "loss": 0.3742,
+      "step": 20890
+    },
+    {
+      "epoch": 2.86,
+      "learning_rate": 0.0015648436423749829,
+      "loss": 0.3215,
+      "step": 20900
+    },
+    {
+      "epoch": 2.86,
+      "eval_loss": 0.33483704924583435,
+      "eval_runtime": 1056.7073,
+      "eval_samples_per_second": 11.056,
+      "eval_steps_per_second": 1.383,
+      "eval_wer": 0.6416469571545529,
+      "step": 20900
+    },
+    {
+      "epoch": 2.86,
+      "learning_rate": 0.0015642926022868165,
+      "loss": 0.4138,
+      "step": 20910
+    },
+    {
+      "epoch": 2.87,
+      "learning_rate": 0.0015637415621986498,
+      "loss": 0.4063,
+      "step": 20920
+    },
+    {
+      "epoch": 2.87,
+      "learning_rate": 0.0015631905221104836,
+      "loss": 0.387,
+      "step": 20930
+    },
+    {
+      "epoch": 2.87,
+      "learning_rate": 0.001562639482022317,
+      "loss": 0.389,
+      "step": 20940
+    },
+    {
+      "epoch": 2.87,
+      "learning_rate": 0.0015620884419341508,
+      "loss": 0.3564,
+      "step": 20950
+    },
+    {
+      "epoch": 2.87,
+      "learning_rate": 0.0015615374018459842,
+      "loss": 0.4296,
+      "step": 20960
+    },
+    {
+      "epoch": 2.87,
+      "learning_rate": 0.001560986361757818,
+      "loss": 0.413,
+      "step": 20970
+    },
+    {
+      "epoch": 2.87,
+      "learning_rate": 0.0015604353216696514,
+      "loss": 0.4051,
+      "step": 20980
+    },
+    {
+      "epoch": 2.88,
+      "learning_rate": 0.001559884281581485,
+      "loss": 0.4033,
+      "step": 20990
+    },
+    {
+      "epoch": 2.88,
+      "learning_rate": 0.0015593332414933186,
+      "loss": 0.3806,
+      "step": 21000
+    },
+    {
+      "epoch": 2.88,
+      "eval_loss": 0.3501095771789551,
+      "eval_runtime": 1056.0562,
+      "eval_samples_per_second": 11.063,
+      "eval_steps_per_second": 1.383,
+      "eval_wer": 0.6777494769398709,
+      "step": 21000
+    },
+    {
+      "epoch": 2.88,
+      "learning_rate": 0.0015587822014051522,
+      "loss": 0.4279,
+      "step": 21010
+    },
+    {
+      "epoch": 2.88,
+      "learning_rate": 0.001558231161316986,
+      "loss": 0.4158,
+      "step": 21020
+    },
+    {
+      "epoch": 2.88,
+      "learning_rate": 0.0015576801212288194,
+      "loss": 0.401,
+      "step": 21030
+    },
+    {
+      "epoch": 2.88,
+      "learning_rate": 0.0015571290811406532,
+      "loss": 0.3997,
+      "step": 21040
+    },
+    {
+      "epoch": 2.88,
+      "learning_rate": 0.0015565780410524866,
+      "loss": 0.425,
+      "step": 21050
+    },
+    {
+      "epoch": 2.89,
+      "learning_rate": 0.0015560270009643202,
+      "loss": 0.3948,
+      "step": 21060
+    },
+    {
+      "epoch": 2.89,
+      "learning_rate": 0.0015554759608761538,
+      "loss": 0.3637,
+      "step": 21070
+    },
+    {
+      "epoch": 2.89,
+      "learning_rate": 0.0015549249207879874,
+      "loss": 0.3674,
+      "step": 21080
+    },
+    {
+      "epoch": 2.89,
+      "learning_rate": 0.001554373880699821,
+      "loss": 0.4302,
+      "step": 21090
+    },
+    {
+      "epoch": 2.89,
+      "learning_rate": 0.0015538228406116546,
+      "loss": 0.311,
+      "step": 21100
+    },
+    {
+      "epoch": 2.89,
+      "eval_loss": 0.33735784888267517,
+      "eval_runtime": 1056.336,
+      "eval_samples_per_second": 11.06,
+      "eval_steps_per_second": 1.383,
+      "eval_wer": 0.655667242790867,
+      "step": 21100
+    },
+    {
+      "epoch": 2.89,
+      "learning_rate": 0.001553271800523488,
+      "loss": 0.4267,
+      "step": 21110
+    },
+    {
+      "epoch": 2.89,
+      "learning_rate": 0.0015527207604353218,
+      "loss": 0.4322,
+      "step": 21120
+    },
+    {
+      "epoch": 2.89,
+      "learning_rate": 0.0015521697203471551,
+      "loss": 0.3795,
+      "step": 21130
+    },
+    {
+      "epoch": 2.9,
+      "learning_rate": 0.001551618680258989,
+      "loss": 0.4125,
+      "step": 21140
+    },
+    {
+      "epoch": 2.9,
+      "learning_rate": 0.0015510676401708223,
+      "loss": 0.3832,
+      "step": 21150
+    },
+    {
+      "epoch": 2.9,
+      "learning_rate": 0.0015505166000826561,
+      "loss": 0.5865,
+      "step": 21160
+    },
+    {
+      "epoch": 2.9,
+      "learning_rate": 0.0015499655599944897,
+      "loss": 0.4587,
+      "step": 21170
+    },
+    {
+      "epoch": 2.9,
+      "learning_rate": 0.0015494145199063231,
+      "loss": 0.4576,
+      "step": 21180
+    },
+    {
+      "epoch": 2.9,
+      "learning_rate": 0.001548863479818157,
+      "loss": 0.4487,
+      "step": 21190
+    },
+    {
+      "epoch": 2.9,
+      "learning_rate": 0.0015483124397299903,
+      "loss": 0.33,
+      "step": 21200
+    },
+    {
+      "epoch": 2.9,
+      "eval_loss": 0.3296695351600647,
+      "eval_runtime": 1057.3212,
+      "eval_samples_per_second": 11.05,
+      "eval_steps_per_second": 1.382,
+      "eval_wer": 0.6424884017101792,
+      "step": 21200
+    },
+    {
+      "epoch": 2.91,
+      "learning_rate": 0.0015477613996418241,
+      "loss": 0.4127,
+      "step": 21210
+    },
+    {
+      "epoch": 2.91,
+      "learning_rate": 0.0015472103595536575,
+      "loss": 0.3976,
+      "step": 21220
+    },
+    {
+      "epoch": 2.91,
+      "learning_rate": 0.0015466593194654913,
+      "loss": 0.3987,
+      "step": 21230
+    },
+    {
+      "epoch": 2.91,
+      "learning_rate": 0.0015461082793773247,
+      "loss": 0.4227,
+      "step": 21240
+    },
+    {
+      "epoch": 2.91,
+      "learning_rate": 0.0015455572392891583,
+      "loss": 0.3223,
+      "step": 21250
+    },
+    {
+      "epoch": 2.91,
+      "learning_rate": 0.0015450061992009919,
+      "loss": 0.4304,
+      "step": 21260
+    },
+    {
+      "epoch": 2.91,
+      "learning_rate": 0.0015444551591128255,
+      "loss": 0.4278,
+      "step": 21270
+    },
+    {
+      "epoch": 2.92,
+      "learning_rate": 0.001543904119024659,
+      "loss": 0.3953,
+      "step": 21280
+    },
+    {
+      "epoch": 2.92,
+      "learning_rate": 0.0015433530789364927,
+      "loss": 0.4034,
+      "step": 21290
+    },
+    {
+      "epoch": 2.92,
+      "learning_rate": 0.001542802038848326,
+      "loss": 0.3916,
+      "step": 21300
+    },
+    {
+      "epoch": 2.92,
+      "eval_loss": 0.3227943778038025,
+      "eval_runtime": 1056.0733,
+      "eval_samples_per_second": 11.063,
+      "eval_steps_per_second": 1.383,
+      "eval_wer": 0.6373373965250614,
+      "step": 21300
+    },
+    {
+      "epoch": 2.92,
+      "learning_rate": 0.0015422509987601599,
+      "loss": 0.4118,
+      "step": 21310
+    },
+    {
+      "epoch": 2.92,
+      "learning_rate": 0.0015416999586719935,
+      "loss": 0.4211,
+      "step": 21320
+    },
+    {
+      "epoch": 2.92,
+      "learning_rate": 0.001541148918583827,
+      "loss": 0.3988,
+      "step": 21330
+    },
+    {
+      "epoch": 2.92,
+      "learning_rate": 0.0015405978784956607,
+      "loss": 0.4166,
+      "step": 21340
+    },
+    {
+      "epoch": 2.93,
+      "learning_rate": 0.0015400468384074942,
+      "loss": 0.3534,
+      "step": 21350
+    },
+    {
+      "epoch": 2.93,
+      "learning_rate": 0.0015394957983193278,
+      "loss": 0.4279,
+      "step": 21360
+    },
+    {
+      "epoch": 2.93,
+      "learning_rate": 0.0015389447582311612,
+      "loss": 0.3945,
+      "step": 21370
+    },
+    {
+      "epoch": 2.93,
+      "learning_rate": 0.001538393718142995,
+      "loss": 0.4255,
+      "step": 21380
+    },
+    {
+      "epoch": 2.93,
+      "learning_rate": 0.0015378426780548284,
+      "loss": 0.4504,
+      "step": 21390
+    },
+    {
+      "epoch": 2.93,
+      "learning_rate": 0.0015372916379666622,
+      "loss": 0.3767,
+      "step": 21400
+    },
+    {
+      "epoch": 2.93,
+      "eval_loss": 0.3391410708427429,
+      "eval_runtime": 1057.3579,
+      "eval_samples_per_second": 11.049,
+      "eval_steps_per_second": 1.382,
+      "eval_wer": 0.6496634221777495,
+      "step": 21400
+    },
+    {
+      "epoch": 2.93,
+      "learning_rate": 0.0015367405978784956,
+      "loss": 0.4185,
+      "step": 21410
+    },
+    {
+      "epoch": 2.93,
+      "learning_rate": 0.0015361895577903294,
+      "loss": 0.396,
+      "step": 21420
+    },
+    {
+      "epoch": 2.94,
+      "learning_rate": 0.0015356385177021628,
+      "loss": 0.3527,
+      "step": 21430
+    },
+    {
+      "epoch": 2.94,
+      "learning_rate": 0.0015350874776139964,
+      "loss": 0.4202,
+      "step": 21440
+    },
+    {
+      "epoch": 2.94,
+      "learning_rate": 0.00153453643752583,
+      "loss": 0.4095,
+      "step": 21450
+    },
+    {
+      "epoch": 2.94,
+      "learning_rate": 0.0015339853974376636,
+      "loss": 0.4219,
+      "step": 21460
+    },
+    {
+      "epoch": 2.94,
+      "learning_rate": 0.0015334343573494972,
+      "loss": 0.3977,
+      "step": 21470
+    },
+    {
+      "epoch": 2.94,
+      "learning_rate": 0.0015328833172613308,
+      "loss": 0.534,
+      "step": 21480
+    },
+    {
+      "epoch": 2.94,
+      "learning_rate": 0.0015323322771731644,
+      "loss": 0.4425,
+      "step": 21490
+    },
+    {
+      "epoch": 2.95,
+      "learning_rate": 0.001531781237084998,
+      "loss": 0.3951,
+      "step": 21500
+    },
+    {
+      "epoch": 2.95,
+      "eval_loss": 0.41201913356781006,
+      "eval_runtime": 1059.787,
+      "eval_samples_per_second": 11.024,
+      "eval_steps_per_second": 1.379,
+      "eval_wer": 0.7108387155462567,
+      "step": 21500
+    },
+    {
+      "epoch": 2.95,
+      "learning_rate": 0.0015312301969968316,
+      "loss": 0.4798,
+      "step": 21510
+    },
+    {
+      "epoch": 2.95,
+      "learning_rate": 0.0015306791569086652,
+      "loss": 0.4608,
+      "step": 21520
+    },
+    {
+      "epoch": 2.95,
+      "learning_rate": 0.0015301281168204988,
+      "loss": 0.4221,
+      "step": 21530
+    },
+    {
+      "epoch": 2.95,
+      "learning_rate": 0.0015295770767323324,
+      "loss": 0.3802,
+      "step": 21540
+    },
+    {
+      "epoch": 2.95,
+      "learning_rate": 0.001529026036644166,
+      "loss": 0.3376,
+      "step": 21550
+    },
+    {
+      "epoch": 2.95,
+      "learning_rate": 0.0015284749965559993,
+      "loss": 0.4184,
+      "step": 21560
+    },
+    {
+      "epoch": 2.96,
+      "learning_rate": 0.0015279239564678331,
+      "loss": 0.4184,
+      "step": 21570
+    },
+    {
+      "epoch": 2.96,
+      "learning_rate": 0.0015273729163796665,
+      "loss": 0.4041,
+      "step": 21580
+    },
+    {
+      "epoch": 2.96,
+      "learning_rate": 0.0015268218762915003,
+      "loss": 0.4006,
+      "step": 21590
+    },
+    {
+      "epoch": 2.96,
+      "learning_rate": 0.0015262708362033337,
+      "loss": 0.3302,
+      "step": 21600
+    },
+    {
+      "epoch": 2.96,
+      "eval_loss": 0.32705143094062805,
+      "eval_runtime": 1059.5723,
+      "eval_samples_per_second": 11.026,
+      "eval_steps_per_second": 1.379,
+      "eval_wer": 0.6308787410170108,
+      "step": 21600
+    },
+    {
+      "epoch": 2.96,
+      "learning_rate": 0.0015257197961151675,
+      "loss": 0.3692,
+      "step": 21610
+    },
+    {
+      "epoch": 2.96,
+      "learning_rate": 0.001525168756027001,
+      "loss": 0.4081,
+      "step": 21620
+    },
+    {
+      "epoch": 2.96,
+      "learning_rate": 0.0015246177159388345,
+      "loss": 0.3444,
+      "step": 21630
+    },
+    {
+      "epoch": 2.96,
+      "learning_rate": 0.0015240666758506683,
+      "loss": 0.8576,
+      "step": 21640
+    },
+    {
+      "epoch": 2.97,
+      "learning_rate": 0.0015235156357625017,
+      "loss": 0.3874,
+      "step": 21650
+    },
+    {
+      "epoch": 2.97,
+      "learning_rate": 0.0015229645956743355,
+      "loss": 0.3926,
+      "step": 21660
+    },
+    {
+      "epoch": 2.97,
+      "learning_rate": 0.0015224135555861689,
+      "loss": 0.3731,
+      "step": 21670
+    },
+    {
+      "epoch": 2.97,
+      "learning_rate": 0.0015218625154980025,
+      "loss": 0.3838,
+      "step": 21680
+    },
+    {
+      "epoch": 2.97,
+      "learning_rate": 0.001521311475409836,
+      "loss": 0.392,
+      "step": 21690
+    },
+    {
+      "epoch": 2.97,
+      "learning_rate": 0.0015207604353216697,
+      "loss": 0.3759,
+      "step": 21700
+    },
+    {
+      "epoch": 2.97,
+      "eval_loss": 0.3210052251815796,
+      "eval_runtime": 1055.7957,
+      "eval_samples_per_second": 11.066,
+      "eval_steps_per_second": 1.384,
+      "eval_wer": 0.6292981897571182,
+      "step": 21700
+    },
+    {
+      "epoch": 2.97,
+      "learning_rate": 0.0015202093952335033,
+      "loss": 0.5184,
+      "step": 21710
+    },
+    {
+      "epoch": 2.98,
+      "learning_rate": 0.0015196583551453369,
+      "loss": 0.3832,
+      "step": 21720
+    },
+    {
+      "epoch": 2.98,
+      "learning_rate": 0.0015191073150571705,
+      "loss": 0.4144,
+      "step": 21730
+    },
+    {
+      "epoch": 2.98,
+      "learning_rate": 0.001518556274969004,
+      "loss": 0.4993,
+      "step": 21740
+    },
+    {
+      "epoch": 2.98,
+      "learning_rate": 0.0015180052348808374,
+      "loss": 0.4006,
+      "step": 21750
+    },
+    {
+      "epoch": 2.98,
+      "learning_rate": 0.0015174541947926713,
+      "loss": 0.4301,
+      "step": 21760
+    },
+    {
+      "epoch": 2.98,
+      "learning_rate": 0.0015169031547045046,
+      "loss": 0.3952,
+      "step": 21770
+    },
+    {
+      "epoch": 2.98,
+      "learning_rate": 0.0015163521146163384,
+      "loss": 0.4074,
+      "step": 21780
+    },
+    {
+      "epoch": 2.99,
+      "learning_rate": 0.001515801074528172,
+      "loss": 0.4626,
+      "step": 21790
+    },
+    {
+      "epoch": 2.99,
+      "learning_rate": 0.0015152500344400056,
+      "loss": 0.3984,
+      "step": 21800
+    },
+    {
+      "epoch": 2.99,
+      "eval_loss": 0.35977602005004883,
+      "eval_runtime": 1055.7157,
+      "eval_samples_per_second": 11.066,
+      "eval_steps_per_second": 1.384,
+      "eval_wer": 0.6565314290912398,
+      "step": 21800
+    },
+    {
+      "epoch": 2.99,
+      "learning_rate": 0.0015146989943518392,
+      "loss": 0.4148,
+      "step": 21810
+    },
+    {
+      "epoch": 2.99,
+      "learning_rate": 0.0015141479542636726,
+      "loss": 0.37,
+      "step": 21820
+    },
+    {
+      "epoch": 2.99,
+      "learning_rate": 0.0015135969141755064,
+      "loss": 0.3451,
+      "step": 21830
+    },
+    {
+      "epoch": 2.99,
+      "learning_rate": 0.0015130458740873398,
+      "loss": 0.42,
+      "step": 21840
+    },
+    {
+      "epoch": 2.99,
+      "learning_rate": 0.0015124948339991736,
+      "loss": 0.3692,
+      "step": 21850
+    },
+    {
+      "epoch": 2.99,
+      "learning_rate": 0.001511943793911007,
+      "loss": 0.4293,
+      "step": 21860
+    },
+    {
+      "epoch": 3.0,
+      "learning_rate": 0.0015113927538228406,
+      "loss": 0.4089,
+      "step": 21870
+    },
+    {
+      "epoch": 3.0,
+      "learning_rate": 0.0015108417137346742,
+      "loss": 0.4345,
+      "step": 21880
+    },
+    {
+      "epoch": 3.0,
+      "learning_rate": 0.0015102906736465078,
+      "loss": 0.3802,
+      "step": 21890
+    },
+    {
+      "epoch": 3.0,
+      "learning_rate": 0.0015097396335583414,
+      "loss": 0.3391,
+      "step": 21900
+    },
+    {
+      "epoch": 3.0,
+      "eval_loss": 0.3367140591144562,
+      "eval_runtime": 1057.8742,
+      "eval_samples_per_second": 11.044,
+      "eval_steps_per_second": 1.381,
+      "eval_wer": 0.6470367506595106,
+      "step": 21900
+    },
+    {
+      "epoch": 3.0,
+      "learning_rate": 0.001509188593470175,
+      "loss": 0.4421,
+      "step": 21910
+    },
+    {
+      "epoch": 3.0,
+      "learning_rate": 0.0015086375533820086,
+      "loss": 0.3905,
+      "step": 21920
+    },
+    {
+      "epoch": 3.0,
+      "learning_rate": 0.0015080865132938422,
+      "loss": 0.3833,
+      "step": 21930
+    },
+    {
+      "epoch": 3.01,
+      "learning_rate": 0.0015075354732056756,
+      "loss": 0.4201,
+      "step": 21940
+    },
+    {
+      "epoch": 3.01,
+      "learning_rate": 0.0015069844331175094,
+      "loss": 0.3726,
+      "step": 21950
+    },
+    {
+      "epoch": 3.01,
+      "learning_rate": 0.001506433393029343,
+      "loss": 0.4154,
+      "step": 21960
+    },
+    {
+      "epoch": 3.01,
+      "learning_rate": 0.0015058823529411766,
+      "loss": 0.4069,
+      "step": 21970
+    },
+    {
+      "epoch": 3.01,
+      "learning_rate": 0.0015053313128530102,
+      "loss": 0.3588,
+      "step": 21980
+    },
+    {
+      "epoch": 3.01,
+      "learning_rate": 0.0015047802727648435,
+      "loss": 0.3781,
+      "step": 21990
+    },
+    {
+      "epoch": 3.01,
+      "learning_rate": 0.0015042292326766773,
+      "loss": 0.3271,
+      "step": 22000
+    },
+    {
+      "epoch": 3.01,
+      "eval_loss": 0.32491961121559143,
+      "eval_runtime": 1066.1419,
+      "eval_samples_per_second": 10.958,
+      "eval_steps_per_second": 1.37,
+      "eval_wer": 0.635245156008369,
+      "step": 22000
+    },
+    {
+      "epoch": 3.02,
+      "learning_rate": 0.0015036781925885107,
+      "loss": 0.396,
+      "step": 22010
+    },
+    {
+      "epoch": 3.02,
+      "learning_rate": 0.0015031271525003445,
+      "loss": 0.3676,
+      "step": 22020
+    },
+    {
+      "epoch": 3.02,
+      "learning_rate": 0.001502576112412178,
+      "loss": 0.4184,
+      "step": 22030
+    },
+    {
+      "epoch": 3.02,
+      "learning_rate": 0.0015020250723240117,
+      "loss": 0.4136,
+      "step": 22040
+    },
+    {
+      "epoch": 3.02,
+      "learning_rate": 0.001501474032235845,
+      "loss": 0.3392,
+      "step": 22050
+    },
+    {
+      "epoch": 3.02,
+      "learning_rate": 0.0015009229921476787,
+      "loss": 0.3918,
+      "step": 22060
+    },
+    {
+      "epoch": 3.02,
+      "learning_rate": 0.0015003719520595123,
+      "loss": 0.3956,
+      "step": 22070
+    },
+    {
+      "epoch": 3.03,
+      "learning_rate": 0.001499820911971346,
+      "loss": 0.3424,
+      "step": 22080
+    },
+    {
+      "epoch": 3.03,
+      "learning_rate": 0.0014992698718831795,
+      "loss": 0.399,
+      "step": 22090
+    },
+    {
+      "epoch": 3.03,
+      "learning_rate": 0.001498718831795013,
+      "loss": 0.3205,
+      "step": 22100
+    },
+    {
+      "epoch": 3.03,
+      "eval_loss": 0.34936413168907166,
+      "eval_runtime": 1060.7821,
+      "eval_samples_per_second": 11.014,
+      "eval_steps_per_second": 1.377,
+      "eval_wer": 0.6460588556354043,
+      "step": 22100
+    },
+    {
+      "epoch": 3.03,
+      "learning_rate": 0.001498167791706847,
+      "loss": 0.4046,
+      "step": 22110
+    },
+    {
+      "epoch": 3.03,
+      "learning_rate": 0.0014976167516186803,
+      "loss": 0.3984,
+      "step": 22120
+    },
+    {
+      "epoch": 3.03,
+      "learning_rate": 0.0014970657115305139,
+      "loss": 0.3439,
+      "step": 22130
+    },
+    {
+      "epoch": 3.03,
+      "learning_rate": 0.0014965146714423475,
+      "loss": 0.4122,
+      "step": 22140
+    },
+    {
+      "epoch": 3.03,
+      "learning_rate": 0.001495963631354181,
+      "loss": 0.3346,
+      "step": 22150
+    },
+    {
+      "epoch": 3.04,
+      "learning_rate": 0.0014954125912660147,
+      "loss": 0.3684,
+      "step": 22160
+    },
+    {
+      "epoch": 3.04,
+      "learning_rate": 0.0014948615511778483,
+      "loss": 0.3935,
+      "step": 22170
+    },
+    {
+      "epoch": 3.04,
+      "learning_rate": 0.0014943105110896816,
+      "loss": 0.3674,
+      "step": 22180
+    },
+    {
+      "epoch": 3.04,
+      "learning_rate": 0.0014937594710015155,
+      "loss": 0.3813,
+      "step": 22190
+    },
+    {
+      "epoch": 3.04,
+      "learning_rate": 0.0014932084309133488,
+      "loss": 0.3037,
+      "step": 22200
+    },
+    {
+      "epoch": 3.04,
+      "eval_loss": 0.3112682104110718,
+      "eval_runtime": 1070.8405,
+      "eval_samples_per_second": 10.91,
+      "eval_steps_per_second": 1.364,
+      "eval_wer": 0.6190985172382425,
+      "step": 22200
+    },
+    {
+      "epoch": 3.04,
+      "learning_rate": 0.0014926573908251826,
+      "loss": 0.4043,
+      "step": 22210
+    },
+    {
+      "epoch": 3.04,
+      "learning_rate": 0.001492106350737016,
+      "loss": 0.3272,
+      "step": 22220
+    },
+    {
+      "epoch": 3.05,
+      "learning_rate": 0.0014915553106488498,
+      "loss": 0.3387,
+      "step": 22230
+    },
+    {
+      "epoch": 3.05,
+      "learning_rate": 0.0014910042705606832,
+      "loss": 0.3813,
+      "step": 22240
+    },
+    {
+      "epoch": 3.05,
+      "learning_rate": 0.0014904532304725168,
+      "loss": 0.3449,
+      "step": 22250
+    },
+    {
+      "epoch": 3.05,
+      "learning_rate": 0.0014899021903843506,
+      "loss": 0.4813,
+      "step": 22260
+    },
+    {
+      "epoch": 3.05,
+      "learning_rate": 0.001489351150296184,
+      "loss": 0.3656,
+      "step": 22270
+    },
+    {
+      "epoch": 3.05,
+      "learning_rate": 0.0014888001102080178,
+      "loss": 0.3738,
+      "step": 22280
+    },
+    {
+      "epoch": 3.05,
+      "learning_rate": 0.0014882490701198512,
+      "loss": 0.377,
+      "step": 22290
+    },
+    {
+      "epoch": 3.06,
+      "learning_rate": 0.001487698030031685,
+      "loss": 0.3389,
+      "step": 22300
+    },
+    {
+      "epoch": 3.06,
+      "eval_loss": 0.3393755853176117,
+      "eval_runtime": 1057.7853,
+      "eval_samples_per_second": 11.045,
+      "eval_steps_per_second": 1.381,
+      "eval_wer": 0.6502092240516693,
+      "step": 22300
+    },
+    {
+      "epoch": 3.06,
+      "learning_rate": 0.0014871469899435184,
+      "loss": 0.433,
+      "step": 22310
+    },
+    {
+      "epoch": 3.06,
+      "learning_rate": 0.001486595949855352,
+      "loss": 0.3889,
+      "step": 22320
+    },
+    {
+      "epoch": 3.06,
+      "learning_rate": 0.0014860449097671856,
+      "loss": 0.4139,
+      "step": 22330
+    },
+    {
+      "epoch": 3.06,
+      "learning_rate": 0.0014854938696790192,
+      "loss": 0.373,
+      "step": 22340
+    },
+    {
+      "epoch": 3.06,
+      "learning_rate": 0.0014849428295908528,
+      "loss": 0.3772,
+      "step": 22350
+    },
+    {
+      "epoch": 3.06,
+      "learning_rate": 0.0014843917895026864,
+      "loss": 0.4006,
+      "step": 22360
+    },
+    {
+      "epoch": 3.06,
+      "learning_rate": 0.0014838407494145197,
+      "loss": 0.3773,
+      "step": 22370
+    },
+    {
+      "epoch": 3.07,
+      "learning_rate": 0.0014832897093263536,
+      "loss": 0.4174,
+      "step": 22380
+    },
+    {
+      "epoch": 3.07,
+      "learning_rate": 0.001482738669238187,
+      "loss": 0.4052,
+      "step": 22390
+    },
+    {
+      "epoch": 3.07,
+      "learning_rate": 0.0014821876291500208,
+      "loss": 0.3817,
+      "step": 22400
+    },
+    {
+      "epoch": 3.07,
+      "eval_loss": 0.32760316133499146,
+      "eval_runtime": 1055.4202,
+      "eval_samples_per_second": 11.07,
+      "eval_steps_per_second": 1.384,
+      "eval_wer": 0.6281952151369053,
+      "step": 22400
+    },
+    {
+      "epoch": 3.07,
+      "learning_rate": 0.0014816365890618544,
+      "loss": 0.4287,
+      "step": 22410
+    },
+    {
+      "epoch": 3.07,
+      "learning_rate": 0.001481085548973688,
+      "loss": 0.3791,
+      "step": 22420
+    },
+    {
+      "epoch": 3.07,
+      "learning_rate": 0.0014805345088855215,
+      "loss": 0.3855,
+      "step": 22430
+    },
+    {
+      "epoch": 3.07,
+      "learning_rate": 0.001479983468797355,
+      "loss": 0.3903,
+      "step": 22440
+    },
+    {
+      "epoch": 3.08,
+      "learning_rate": 0.0014794324287091887,
+      "loss": 0.3331,
+      "step": 22450
+    },
+    {
+      "epoch": 3.08,
+      "learning_rate": 0.0014788813886210221,
+      "loss": 0.4316,
+      "step": 22460
+    },
+    {
+      "epoch": 3.08,
+      "learning_rate": 0.001478330348532856,
+      "loss": 0.3967,
+      "step": 22470
+    },
+    {
+      "epoch": 3.08,
+      "learning_rate": 0.0014777793084446893,
+      "loss": 0.3996,
+      "step": 22480
+    },
+    {
+      "epoch": 3.08,
+      "learning_rate": 0.0014772282683565231,
+      "loss": 0.3698,
+      "step": 22490
+    },
+    {
+      "epoch": 3.08,
+      "learning_rate": 0.0014766772282683565,
+      "loss": 0.3801,
+      "step": 22500
+    },
+    {
+      "epoch": 3.08,
+      "eval_loss": 0.3176908791065216,
+      "eval_runtime": 1062.6676,
+      "eval_samples_per_second": 10.994,
+      "eval_steps_per_second": 1.375,
+      "eval_wer": 0.6243177476576003,
+      "step": 22500
+    },
+    {
+      "epoch": 3.08,
+      "learning_rate": 0.00147612618818019,
+      "loss": 0.4173,
+      "step": 22510
+    },
+    {
+      "epoch": 3.09,
+      "learning_rate": 0.0014755751480920237,
+      "loss": 0.3822,
+      "step": 22520
+    },
+    {
+      "epoch": 3.09,
+      "learning_rate": 0.0014750241080038573,
+      "loss": 0.3721,
+      "step": 22530
+    },
+    {
+      "epoch": 3.09,
+      "learning_rate": 0.0014744730679156909,
+      "loss": 0.3528,
+      "step": 22540
+    },
+    {
+      "epoch": 3.09,
+      "learning_rate": 0.0014739220278275245,
+      "loss": 0.3836,
+      "step": 22550
+    },
+    {
+      "epoch": 3.09,
+      "learning_rate": 0.0014733709877393579,
+      "loss": 0.4047,
+      "step": 22560
+    },
+    {
+      "epoch": 3.09,
+      "learning_rate": 0.0014728199476511917,
+      "loss": 0.3896,
+      "step": 22570
+    },
+    {
+      "epoch": 3.09,
+      "learning_rate": 0.0014722689075630253,
+      "loss": 0.3879,
+      "step": 22580
+    },
+    {
+      "epoch": 3.09,
+      "learning_rate": 0.0014717178674748589,
+      "loss": 0.368,
+      "step": 22590
+    },
+    {
+      "epoch": 3.1,
+      "learning_rate": 0.0014711668273866925,
+      "loss": 0.4181,
+      "step": 22600
+    },
+    {
+      "epoch": 3.1,
+      "eval_loss": 0.29948726296424866,
+      "eval_runtime": 1059.4942,
+      "eval_samples_per_second": 11.027,
+      "eval_steps_per_second": 1.379,
+      "eval_wer": 0.6090580369325934,
+      "step": 22600
+    },
+    {
+      "epoch": 3.1,
+      "learning_rate": 0.001470615787298526,
+      "loss": 0.3732,
+      "step": 22610
+    },
+    {
+      "epoch": 3.1,
+      "learning_rate": 0.0014700647472103597,
+      "loss": 0.3595,
+      "step": 22620
+    },
+    {
+      "epoch": 3.1,
+      "learning_rate": 0.001469513707122193,
+      "loss": 0.3467,
+      "step": 22630
+    },
+    {
+      "epoch": 3.1,
+      "learning_rate": 0.0014689626670340268,
+      "loss": 0.3763,
+      "step": 22640
+    },
+    {
+      "epoch": 3.1,
+      "learning_rate": 0.0014684116269458602,
+      "loss": 0.3571,
+      "step": 22650
+    },
+    {
+      "epoch": 3.1,
+      "learning_rate": 0.001467860586857694,
+      "loss": 0.3786,
+      "step": 22660
+    },
+    {
+      "epoch": 3.11,
+      "learning_rate": 0.0014673095467695274,
+      "loss": 0.3746,
+      "step": 22670
+    },
+    {
+      "epoch": 3.11,
+      "learning_rate": 0.0014667585066813612,
+      "loss": 0.395,
+      "step": 22680
+    },
+    {
+      "epoch": 3.11,
+      "learning_rate": 0.0014662074665931946,
+      "loss": 0.341,
+      "step": 22690
+    },
+    {
+      "epoch": 3.11,
+      "learning_rate": 0.0014656564265050282,
+      "loss": 0.2837,
+      "step": 22700
+    },
+    {
+      "epoch": 3.11,
+      "eval_loss": 0.3071697950363159,
+      "eval_runtime": 1058.3364,
+      "eval_samples_per_second": 11.039,
+      "eval_steps_per_second": 1.38,
+      "eval_wer": 0.6149822614390976,
+      "step": 22700
+    },
+    {
+      "epoch": 3.11,
+      "learning_rate": 0.0014651053864168618,
+      "loss": 0.4122,
+      "step": 22710
+    },
+    {
+      "epoch": 3.11,
+      "learning_rate": 0.0014645543463286954,
+      "loss": 0.3733,
+      "step": 22720
+    },
+    {
+      "epoch": 3.11,
+      "learning_rate": 0.0014640033062405292,
+      "loss": 0.3306,
+      "step": 22730
+    },
+    {
+      "epoch": 3.12,
+      "learning_rate": 0.0014634522661523626,
+      "loss": 0.4102,
+      "step": 22740
+    },
+    {
+      "epoch": 3.12,
+      "learning_rate": 0.0014629012260641962,
+      "loss": 0.3455,
+      "step": 22750
+    },
+    {
+      "epoch": 3.12,
+      "learning_rate": 0.0014623501859760298,
+      "loss": 0.3551,
+      "step": 22760
+    },
+    {
+      "epoch": 3.12,
+      "learning_rate": 0.0014617991458878634,
+      "loss": 0.4049,
+      "step": 22770
+    },
+    {
+      "epoch": 3.12,
+      "learning_rate": 0.001461248105799697,
+      "loss": 0.3728,
+      "step": 22780
+    },
+    {
+      "epoch": 3.12,
+      "learning_rate": 0.0014606970657115306,
+      "loss": 0.3598,
+      "step": 22790
+    },
+    {
+      "epoch": 3.12,
+      "learning_rate": 0.0014601460256233642,
+      "loss": 0.3423,
+      "step": 22800
+    },
+    {
+      "epoch": 3.12,
+      "eval_loss": 0.3116106688976288,
+      "eval_runtime": 1062.7181,
+      "eval_samples_per_second": 10.994,
+      "eval_steps_per_second": 1.375,
+      "eval_wer": 0.610843263895206,
+      "step": 22800
+    },
+    {
+      "epoch": 3.13,
+      "learning_rate": 0.0014595949855351978,
+      "loss": 0.3975,
+      "step": 22810
+    },
+    {
+      "epoch": 3.13,
+      "learning_rate": 0.0014590439454470311,
+      "loss": 0.3703,
+      "step": 22820
+    },
+    {
+      "epoch": 3.13,
+      "learning_rate": 0.001458492905358865,
+      "loss": 0.3535,
+      "step": 22830
+    },
+    {
+      "epoch": 3.13,
+      "learning_rate": 0.0014579418652706983,
+      "loss": 0.3809,
+      "step": 22840
+    },
+    {
+      "epoch": 3.13,
+      "learning_rate": 0.0014573908251825321,
+      "loss": 0.3399,
+      "step": 22850
+    },
+    {
+      "epoch": 3.13,
+      "learning_rate": 0.0014568397850943655,
+      "loss": 0.4107,
+      "step": 22860
+    },
+    {
+      "epoch": 3.13,
+      "learning_rate": 0.0014562887450061993,
+      "loss": 0.3787,
+      "step": 22870
+    },
+    {
+      "epoch": 3.13,
+      "learning_rate": 0.001455737704918033,
+      "loss": 0.3596,
+      "step": 22880
+    },
+    {
+      "epoch": 3.14,
+      "learning_rate": 0.0014551866648298663,
+      "loss": 0.3887,
+      "step": 22890
+    },
+    {
+      "epoch": 3.14,
+      "learning_rate": 0.0014546356247417001,
+      "loss": 0.32,
+      "step": 22900
+    },
+    {
+      "epoch": 3.14,
+      "eval_loss": 0.3129037022590637,
+      "eval_runtime": 1063.1859,
+      "eval_samples_per_second": 10.989,
+      "eval_steps_per_second": 1.374,
+      "eval_wer": 0.6183252979168562,
+      "step": 22900
+    },
+    {
+      "epoch": 3.14,
+      "learning_rate": 0.0014540845846535335,
+      "loss": 0.4169,
+      "step": 22910
+    },
+    {
+      "epoch": 3.14,
+      "learning_rate": 0.0014535335445653673,
+      "loss": 0.3686,
+      "step": 22920
+    },
+    {
+      "epoch": 3.14,
+      "learning_rate": 0.0014529825044772007,
+      "loss": 0.3671,
+      "step": 22930
+    },
+    {
+      "epoch": 3.14,
+      "learning_rate": 0.0014524314643890343,
+      "loss": 0.376,
+      "step": 22940
+    },
+    {
+      "epoch": 3.14,
+      "learning_rate": 0.0014518804243008679,
+      "loss": 0.3904,
+      "step": 22950
+    },
+    {
+      "epoch": 3.15,
+      "learning_rate": 0.0014513293842127015,
+      "loss": 0.3969,
+      "step": 22960
+    },
+    {
+      "epoch": 3.15,
+      "learning_rate": 0.001450778344124535,
+      "loss": 0.4143,
+      "step": 22970
+    },
+    {
+      "epoch": 3.15,
+      "learning_rate": 0.0014502273040363687,
+      "loss": 0.3858,
+      "step": 22980
+    },
+    {
+      "epoch": 3.15,
+      "learning_rate": 0.0014496762639482023,
+      "loss": 0.361,
+      "step": 22990
+    },
+    {
+      "epoch": 3.15,
+      "learning_rate": 0.0014491252238600359,
+      "loss": 0.3138,
+      "step": 23000
+    },
+    {
+      "epoch": 3.15,
+      "eval_loss": 0.30257460474967957,
+      "eval_runtime": 1058.5621,
+      "eval_samples_per_second": 11.037,
+      "eval_steps_per_second": 1.38,
+      "eval_wer": 0.6174042572546166,
+      "step": 23000
+    },
+    {
+      "epoch": 3.15,
+      "learning_rate": 0.0014485741837718692,
+      "loss": 0.3663,
+      "step": 23010
+    },
+    {
+      "epoch": 3.15,
+      "learning_rate": 0.001448023143683703,
+      "loss": 0.3418,
+      "step": 23020
+    },
+    {
+      "epoch": 3.16,
+      "learning_rate": 0.0014474721035955367,
+      "loss": 0.388,
+      "step": 23030
+    },
+    {
+      "epoch": 3.16,
+      "learning_rate": 0.0014469210635073703,
+      "loss": 0.4072,
+      "step": 23040
+    },
+    {
+      "epoch": 3.16,
+      "learning_rate": 0.0014463700234192038,
+      "loss": 0.2476,
+      "step": 23050
+    },
+    {
+      "epoch": 3.16,
+      "learning_rate": 0.0014458189833310374,
+      "loss": 0.3547,
+      "step": 23060
+    },
+    {
+      "epoch": 3.16,
+      "learning_rate": 0.001445267943242871,
+      "loss": 0.3887,
+      "step": 23070
+    },
+    {
+      "epoch": 3.16,
+      "learning_rate": 0.0014447169031547044,
+      "loss": 0.4251,
+      "step": 23080
+    },
+    {
+      "epoch": 3.16,
+      "learning_rate": 0.0014441658630665382,
+      "loss": 0.3735,
+      "step": 23090
+    },
+    {
+      "epoch": 3.16,
+      "learning_rate": 0.0014436148229783716,
+      "loss": 0.3693,
+      "step": 23100
+    },
+    {
+      "epoch": 3.16,
+      "eval_loss": 0.31039053201675415,
+      "eval_runtime": 1057.2125,
+      "eval_samples_per_second": 11.051,
+      "eval_steps_per_second": 1.382,
+      "eval_wer": 0.6233284817611208,
+      "step": 23100
+    },
+    {
+      "epoch": 3.17,
+      "learning_rate": 0.0014430637828902054,
+      "loss": 0.4078,
+      "step": 23110
+    },
+    {
+      "epoch": 3.17,
+      "learning_rate": 0.0014425127428020388,
+      "loss": 0.3863,
+      "step": 23120
+    },
+    {
+      "epoch": 3.17,
+      "learning_rate": 0.0014419617027138724,
+      "loss": 0.3487,
+      "step": 23130
+    },
+    {
+      "epoch": 3.17,
+      "learning_rate": 0.001441410662625706,
+      "loss": 0.4019,
+      "step": 23140
+    },
+    {
+      "epoch": 3.17,
+      "learning_rate": 0.0014408596225375396,
+      "loss": 0.3452,
+      "step": 23150
+    },
+    {
+      "epoch": 3.17,
+      "learning_rate": 0.0014403085824493732,
+      "loss": 0.3983,
+      "step": 23160
+    },
+    {
+      "epoch": 3.17,
+      "learning_rate": 0.0014397575423612068,
+      "loss": 0.3907,
+      "step": 23170
+    },
+    {
+      "epoch": 3.18,
+      "learning_rate": 0.0014392065022730404,
+      "loss": 0.3874,
+      "step": 23180
+    },
+    {
+      "epoch": 3.18,
+      "learning_rate": 0.001438655462184874,
+      "loss": 0.424,
+      "step": 23190
+    },
+    {
+      "epoch": 3.18,
+      "learning_rate": 0.0014381044220967076,
+      "loss": 0.3765,
+      "step": 23200
+    },
+    {
+      "epoch": 3.18,
+      "eval_loss": 0.319587379693985,
+      "eval_runtime": 1058.2649,
+      "eval_samples_per_second": 11.04,
+      "eval_steps_per_second": 1.381,
+      "eval_wer": 0.6313222050395706,
+      "step": 23200
+    },
+    {
+      "epoch": 3.18,
+      "learning_rate": 0.0014375533820085412,
+      "loss": 0.4507,
+      "step": 23210
+    },
+    {
+      "epoch": 3.18,
+      "learning_rate": 0.0014370023419203748,
+      "loss": 0.411,
+      "step": 23220
+    },
+    {
+      "epoch": 3.18,
+      "learning_rate": 0.0014364513018322084,
+      "loss": 0.3246,
+      "step": 23230
+    },
+    {
+      "epoch": 3.18,
+      "learning_rate": 0.001435900261744042,
+      "loss": 0.4098,
+      "step": 23240
+    },
+    {
+      "epoch": 3.19,
+      "learning_rate": 0.0014353492216558756,
+      "loss": 0.3301,
+      "step": 23250
+    },
+    {
+      "epoch": 3.19,
+      "learning_rate": 0.0014347981815677092,
+      "loss": 0.3735,
+      "step": 23260
+    },
+    {
+      "epoch": 3.19,
+      "learning_rate": 0.0014342471414795425,
+      "loss": 0.3889,
+      "step": 23270
+    },
+    {
+      "epoch": 3.19,
+      "learning_rate": 0.0014336961013913763,
+      "loss": 0.3536,
+      "step": 23280
+    },
+    {
+      "epoch": 3.19,
+      "learning_rate": 0.0014331450613032097,
+      "loss": 0.3577,
+      "step": 23290
+    },
+    {
+      "epoch": 3.19,
+      "learning_rate": 0.0014325940212150435,
+      "loss": 0.4339,
+      "step": 23300
+    },
+    {
+      "epoch": 3.19,
+      "eval_loss": 0.31256937980651855,
+      "eval_runtime": 1065.7333,
+      "eval_samples_per_second": 10.962,
+      "eval_steps_per_second": 1.371,
+      "eval_wer": 0.6231920312926408,
+      "step": 23300
+    },
+    {
+      "epoch": 3.19,
+      "learning_rate": 0.001432042981126877,
+      "loss": 0.4106,
+      "step": 23310
+    },
+    {
+      "epoch": 3.19,
+      "learning_rate": 0.0014314919410387105,
+      "loss": 0.3162,
+      "step": 23320
+    },
+    {
+      "epoch": 3.2,
+      "learning_rate": 0.001430940900950544,
+      "loss": 0.3702,
+      "step": 23330
+    },
+    {
+      "epoch": 3.2,
+      "learning_rate": 0.0014303898608623777,
+      "loss": 0.3331,
+      "step": 23340
+    },
+    {
+      "epoch": 3.2,
+      "learning_rate": 0.0014298388207742115,
+      "loss": 0.3117,
+      "step": 23350
+    },
+    {
+      "epoch": 3.2,
+      "learning_rate": 0.001429287780686045,
+      "loss": 0.4024,
+      "step": 23360
+    },
+    {
+      "epoch": 3.2,
+      "learning_rate": 0.0014287367405978787,
+      "loss": 0.3959,
+      "step": 23370
+    },
+    {
+      "epoch": 3.2,
+      "learning_rate": 0.001428185700509712,
+      "loss": 0.3577,
+      "step": 23380
+    },
+    {
+      "epoch": 3.2,
+      "learning_rate": 0.0014276346604215457,
+      "loss": 0.3585,
+      "step": 23390
+    },
+    {
+      "epoch": 3.21,
+      "learning_rate": 0.0014270836203333793,
+      "loss": 0.3197,
+      "step": 23400
+    },
+    {
+      "epoch": 3.21,
+      "eval_loss": 0.2984280288219452,
+      "eval_runtime": 1059.8144,
+      "eval_samples_per_second": 11.024,
+      "eval_steps_per_second": 1.379,
+      "eval_wer": 0.5982784499226781,
+      "step": 23400
+    },
+    {
+      "epoch": 3.21,
+      "learning_rate": 0.0014265325802452129,
+      "loss": 0.4092,
+      "step": 23410
+    },
+    {
+      "epoch": 3.21,
+      "learning_rate": 0.0014259815401570465,
+      "loss": 0.3743,
+      "step": 23420
+    },
+    {
+      "epoch": 3.21,
+      "learning_rate": 0.00142543050006888,
+      "loss": 0.3445,
+      "step": 23430
+    },
+    {
+      "epoch": 3.21,
+      "learning_rate": 0.0014248794599807134,
+      "loss": 0.3876,
+      "step": 23440
+    },
+    {
+      "epoch": 3.21,
+      "learning_rate": 0.0014243284198925473,
+      "loss": 0.2887,
+      "step": 23450
+    },
+    {
+      "epoch": 3.21,
+      "learning_rate": 0.0014237773798043806,
+      "loss": 0.3827,
+      "step": 23460
+    },
+    {
+      "epoch": 3.22,
+      "learning_rate": 0.0014232263397162145,
+      "loss": 0.37,
+      "step": 23470
+    },
+    {
+      "epoch": 3.22,
+      "learning_rate": 0.0014226752996280478,
+      "loss": 0.3342,
+      "step": 23480
+    },
+    {
+      "epoch": 3.22,
+      "learning_rate": 0.0014221242595398816,
+      "loss": 0.3695,
+      "step": 23490
+    },
+    {
+      "epoch": 3.22,
+      "learning_rate": 0.0014215732194517152,
+      "loss": 0.3411,
+      "step": 23500
+    },
+    {
+      "epoch": 3.22,
+      "eval_loss": 0.3055705726146698,
+      "eval_runtime": 1058.0032,
+      "eval_samples_per_second": 11.042,
+      "eval_steps_per_second": 1.381,
+      "eval_wer": 0.6092968252524333,
+      "step": 23500
+    },
+    {
+      "epoch": 3.22,
+      "learning_rate": 0.0014210221793635486,
+      "loss": 0.359,
+      "step": 23510
+    },
+    {
+      "epoch": 3.22,
+      "learning_rate": 0.0014204711392753824,
+      "loss": 0.3597,
+      "step": 23520
+    },
+    {
+      "epoch": 3.22,
+      "learning_rate": 0.0014199200991872158,
+      "loss": 0.3939,
+      "step": 23530
+    },
+    {
+      "epoch": 3.23,
+      "learning_rate": 0.0014193690590990496,
+      "loss": 0.3629,
+      "step": 23540
+    },
+    {
+      "epoch": 3.23,
+      "learning_rate": 0.001418818019010883,
+      "loss": 0.3421,
+      "step": 23550
+    },
+    {
+      "epoch": 3.23,
+      "learning_rate": 0.0014182669789227168,
+      "loss": 0.3617,
+      "step": 23560
+    },
+    {
+      "epoch": 3.23,
+      "learning_rate": 0.0014177159388345502,
+      "loss": 0.3585,
+      "step": 23570
+    },
+    {
+      "epoch": 3.23,
+      "learning_rate": 0.0014171648987463838,
+      "loss": 0.3634,
+      "step": 23580
+    },
+    {
+      "epoch": 3.23,
+      "learning_rate": 0.0014166138586582174,
+      "loss": 0.3442,
+      "step": 23590
+    },
+    {
+      "epoch": 3.23,
+      "learning_rate": 0.001416062818570051,
+      "loss": 0.3468,
+      "step": 23600
+    },
+    {
+      "epoch": 3.23,
+      "eval_loss": 0.30944326519966125,
+      "eval_runtime": 1060.5881,
+      "eval_samples_per_second": 11.016,
+      "eval_steps_per_second": 1.378,
+      "eval_wer": 0.6145956517784045,
+      "step": 23600
+    },
+    {
+      "epoch": 3.23,
+      "learning_rate": 0.0014155117784818846,
+      "loss": 0.3673,
+      "step": 23610
+    },
+    {
+      "epoch": 3.24,
+      "learning_rate": 0.0014149607383937182,
+      "loss": 0.4099,
+      "step": 23620
+    },
+    {
+      "epoch": 3.24,
+      "learning_rate": 0.0014144096983055516,
+      "loss": 0.3567,
+      "step": 23630
+    },
+    {
+      "epoch": 3.24,
+      "learning_rate": 0.0014138586582173854,
+      "loss": 0.3604,
+      "step": 23640
+    },
+    {
+      "epoch": 3.24,
+      "learning_rate": 0.0014133076181292187,
+      "loss": 0.4102,
+      "step": 23650
+    },
+    {
+      "epoch": 3.24,
+      "learning_rate": 0.0014127565780410526,
+      "loss": 0.3829,
+      "step": 23660
+    },
+    {
+      "epoch": 3.24,
+      "learning_rate": 0.0014122055379528862,
+      "loss": 0.3594,
+      "step": 23670
+    },
+    {
+      "epoch": 3.24,
+      "learning_rate": 0.0014116544978647198,
+      "loss": 0.3356,
+      "step": 23680
+    },
+    {
+      "epoch": 3.25,
+      "learning_rate": 0.0014111034577765533,
+      "loss": 0.3871,
+      "step": 23690
+    },
+    {
+      "epoch": 3.25,
+      "learning_rate": 0.0014105524176883867,
+      "loss": 0.3297,
+      "step": 23700
+    },
+    {
+      "epoch": 3.25,
+      "eval_loss": 0.30329951643943787,
+      "eval_runtime": 1071.3817,
+      "eval_samples_per_second": 10.905,
+      "eval_steps_per_second": 1.364,
+      "eval_wer": 0.6029859910852361,
+      "step": 23700
+    },
+    {
+      "epoch": 3.25,
+      "learning_rate": 0.0014100013776002205,
+      "loss": 0.3445,
+      "step": 23710
+    },
+    {
+      "epoch": 3.25,
+      "learning_rate": 0.001409450337512054,
+      "loss": 0.3391,
+      "step": 23720
+    },
+    {
+      "epoch": 3.25,
+      "learning_rate": 0.0014088992974238877,
+      "loss": 0.3081,
+      "step": 23730
+    },
+    {
+      "epoch": 3.25,
+      "learning_rate": 0.0014083482573357211,
+      "loss": 0.3575,
+      "step": 23740
+    },
+    {
+      "epoch": 3.25,
+      "learning_rate": 0.001407797217247555,
+      "loss": 0.3166,
+      "step": 23750
+    },
+    {
+      "epoch": 3.26,
+      "learning_rate": 0.0014072461771593883,
+      "loss": 0.3675,
+      "step": 23760
+    },
+    {
+      "epoch": 3.26,
+      "learning_rate": 0.001406695137071222,
+      "loss": 0.3658,
+      "step": 23770
+    },
+    {
+      "epoch": 3.26,
+      "learning_rate": 0.0014061440969830555,
+      "loss": 0.3794,
+      "step": 23780
+    },
+    {
+      "epoch": 3.26,
+      "learning_rate": 0.001405593056894889,
+      "loss": 0.3719,
+      "step": 23790
+    },
+    {
+      "epoch": 3.26,
+      "learning_rate": 0.0014050420168067227,
+      "loss": 0.3018,
+      "step": 23800
+    },
+    {
+      "epoch": 3.26,
+      "eval_loss": 0.28488004207611084,
+      "eval_runtime": 1068.2725,
+      "eval_samples_per_second": 10.936,
+      "eval_steps_per_second": 1.368,
+      "eval_wer": 0.5892954607477485,
+      "step": 23800
+    },
+    {
+      "epoch": 3.26,
+      "learning_rate": 0.0014044909767185563,
+      "loss": 0.4599,
+      "step": 23810
+    },
+    {
+      "epoch": 3.26,
+      "learning_rate": 0.00140393993663039,
+      "loss": 0.3719,
+      "step": 23820
+    },
+    {
+      "epoch": 3.26,
+      "learning_rate": 0.0014033888965422235,
+      "loss": 0.3774,
+      "step": 23830
+    },
+    {
+      "epoch": 3.27,
+      "learning_rate": 0.001402837856454057,
+      "loss": 0.3245,
+      "step": 23840
+    },
+    {
+      "epoch": 3.27,
+      "learning_rate": 0.0014022868163658907,
+      "loss": 0.317,
+      "step": 23850
+    },
+    {
+      "epoch": 3.27,
+      "learning_rate": 0.0014017357762777243,
+      "loss": 0.3819,
+      "step": 23860
+    },
+    {
+      "epoch": 3.27,
+      "learning_rate": 0.0014011847361895579,
+      "loss": 0.435,
+      "step": 23870
+    },
+    {
+      "epoch": 3.27,
+      "learning_rate": 0.0014006336961013915,
+      "loss": 0.3643,
+      "step": 23880
+    },
+    {
+      "epoch": 3.27,
+      "learning_rate": 0.0014000826560132248,
+      "loss": 0.3665,
+      "step": 23890
+    },
+    {
+      "epoch": 3.27,
+      "learning_rate": 0.0013995316159250587,
+      "loss": 0.3119,
+      "step": 23900
+    },
+    {
+      "epoch": 3.27,
+      "eval_loss": 0.2943757176399231,
+      "eval_runtime": 1059.2793,
+      "eval_samples_per_second": 11.029,
+      "eval_steps_per_second": 1.379,
+      "eval_wer": 0.604680251068862,
+      "step": 23900
+    },
+    {
+      "epoch": 3.28,
+      "learning_rate": 0.001398980575836892,
+      "loss": 0.3733,
+      "step": 23910
+    },
+    {
+      "epoch": 3.28,
+      "learning_rate": 0.0013984295357487258,
+      "loss": 0.6583,
+      "step": 23920
+    },
+    {
+      "epoch": 3.28,
+      "learning_rate": 0.0013978784956605592,
+      "loss": 0.3481,
+      "step": 23930
+    },
+    {
+      "epoch": 3.28,
+      "learning_rate": 0.001397327455572393,
+      "loss": 0.4019,
+      "step": 23940
+    },
+    {
+      "epoch": 3.28,
+      "learning_rate": 0.0013967764154842264,
+      "loss": 0.3297,
+      "step": 23950
+    },
+    {
+      "epoch": 3.28,
+      "learning_rate": 0.00139622537539606,
+      "loss": 0.3791,
+      "step": 23960
+    },
+    {
+      "epoch": 3.28,
+      "learning_rate": 0.0013956743353078938,
+      "loss": 0.3408,
+      "step": 23970
+    },
+    {
+      "epoch": 3.29,
+      "learning_rate": 0.0013951232952197272,
+      "loss": 0.338,
+      "step": 23980
+    },
+    {
+      "epoch": 3.29,
+      "learning_rate": 0.001394572255131561,
+      "loss": 0.3664,
+      "step": 23990
+    },
+    {
+      "epoch": 3.29,
+      "learning_rate": 0.0013940212150433944,
+      "loss": 0.2679,
+      "step": 24000
+    },
+    {
+      "epoch": 3.29,
+      "eval_loss": 0.28660818934440613,
+      "eval_runtime": 1060.6854,
+      "eval_samples_per_second": 11.015,
+      "eval_steps_per_second": 1.377,
+      "eval_wer": 0.5912739925407077,
+      "step": 24000
+    },
+    {
+      "epoch": 3.29,
+      "learning_rate": 0.001393470174955228,
+      "loss": 0.3627,
+      "step": 24010
+    },
+    {
+      "epoch": 3.29,
+      "learning_rate": 0.0013929191348670616,
+      "loss": 0.3534,
+      "step": 24020
+    },
+    {
+      "epoch": 3.29,
+      "learning_rate": 0.0013923680947788952,
+      "loss": 0.3551,
+      "step": 24030
+    },
+    {
+      "epoch": 3.29,
+      "learning_rate": 0.0013918170546907288,
+      "loss": 0.3707,
+      "step": 24040
+    },
+    {
+      "epoch": 3.29,
+      "learning_rate": 0.0013912660146025624,
+      "loss": 0.3101,
+      "step": 24050
+    },
+    {
+      "epoch": 3.3,
+      "learning_rate": 0.001390714974514396,
+      "loss": 0.3497,
+      "step": 24060
+    },
+    {
+      "epoch": 3.3,
+      "learning_rate": 0.0013901639344262296,
+      "loss": 0.3411,
+      "step": 24070
+    },
+    {
+      "epoch": 3.3,
+      "learning_rate": 0.001389612894338063,
+      "loss": 0.3422,
+      "step": 24080
+    },
+    {
+      "epoch": 3.3,
+      "learning_rate": 0.0013890618542498968,
+      "loss": 0.3686,
+      "step": 24090
+    },
+    {
+      "epoch": 3.3,
+      "learning_rate": 0.0013885108141617301,
+      "loss": 0.4425,
+      "step": 24100
+    },
+    {
+      "epoch": 3.3,
+      "eval_loss": 0.29774948954582214,
+      "eval_runtime": 1062.9705,
+      "eval_samples_per_second": 10.991,
+      "eval_steps_per_second": 1.374,
+      "eval_wer": 0.608625943782407,
+      "step": 24100
+    },
+    {
+      "epoch": 3.3,
+      "learning_rate": 0.001387959774073564,
+      "loss": 0.3618,
+      "step": 24110
+    },
+    {
+      "epoch": 3.3,
+      "learning_rate": 0.0013874087339853975,
+      "loss": 0.4554,
+      "step": 24120
+    },
+    {
+      "epoch": 3.31,
+      "learning_rate": 0.0013868576938972311,
+      "loss": 0.3775,
+      "step": 24130
+    },
+    {
+      "epoch": 3.31,
+      "learning_rate": 0.0013863066538090647,
+      "loss": 0.3617,
+      "step": 24140
+    },
+    {
+      "epoch": 3.31,
+      "learning_rate": 0.0013857556137208981,
+      "loss": 0.3159,
+      "step": 24150
+    },
+    {
+      "epoch": 3.31,
+      "learning_rate": 0.001385204573632732,
+      "loss": 0.388,
+      "step": 24160
+    },
+    {
+      "epoch": 3.31,
+      "learning_rate": 0.0013846535335445653,
+      "loss": 0.3417,
+      "step": 24170
+    },
+    {
+      "epoch": 3.31,
+      "learning_rate": 0.0013841024934563991,
+      "loss": 0.3551,
+      "step": 24180
+    },
+    {
+      "epoch": 3.31,
+      "learning_rate": 0.0013835514533682325,
+      "loss": 0.3587,
+      "step": 24190
+    },
+    {
+      "epoch": 3.32,
+      "learning_rate": 0.001383000413280066,
+      "loss": 0.2961,
+      "step": 24200
+    },
+    {
+      "epoch": 3.32,
+      "eval_loss": 0.2929608225822449,
+      "eval_runtime": 1061.4709,
+      "eval_samples_per_second": 11.006,
+      "eval_steps_per_second": 1.376,
+      "eval_wer": 0.6002228690985172,
+      "step": 24200
+    },
+    {
+      "epoch": 3.32,
+      "learning_rate": 0.0013824493731918997,
+      "loss": 0.382,
+      "step": 24210
+    },
+    {
+      "epoch": 3.32,
+      "learning_rate": 0.0013818983331037333,
+      "loss": 0.3386,
+      "step": 24220
+    },
+    {
+      "epoch": 3.32,
+      "learning_rate": 0.0013813472930155669,
+      "loss": 0.3695,
+      "step": 24230
+    },
+    {
+      "epoch": 3.32,
+      "learning_rate": 0.0013807962529274005,
+      "loss": 0.3387,
+      "step": 24240
+    },
+    {
+      "epoch": 3.32,
+      "learning_rate": 0.001380245212839234,
+      "loss": 0.3554,
+      "step": 24250
+    },
+    {
+      "epoch": 3.32,
+      "learning_rate": 0.0013796941727510677,
+      "loss": 0.3619,
+      "step": 24260
+    },
+    {
+      "epoch": 3.33,
+      "learning_rate": 0.001379143132662901,
+      "loss": 0.327,
+      "step": 24270
+    },
+    {
+      "epoch": 3.33,
+      "learning_rate": 0.0013785920925747349,
+      "loss": 0.3668,
+      "step": 24280
+    },
+    {
+      "epoch": 3.33,
+      "learning_rate": 0.0013780410524865685,
+      "loss": 0.3575,
+      "step": 24290
+    },
+    {
+      "epoch": 3.33,
+      "learning_rate": 0.001377490012398402,
+      "loss": 0.2888,
+      "step": 24300
+    },
+    {
+      "epoch": 3.33,
+      "eval_loss": 0.2751685380935669,
+      "eval_runtime": 1060.6245,
+      "eval_samples_per_second": 11.015,
+      "eval_steps_per_second": 1.377,
+      "eval_wer": 0.577367415628127,
+      "step": 24300
+    },
+    {
+      "epoch": 3.33,
+      "learning_rate": 0.0013769389723102357,
+      "loss": 0.4011,
+      "step": 24310
+    },
+    {
+      "epoch": 3.33,
+      "learning_rate": 0.0013763879322220693,
+      "loss": 0.3574,
+      "step": 24320
+    },
+    {
+      "epoch": 3.33,
+      "learning_rate": 0.0013758368921339028,
+      "loss": 0.352,
+      "step": 24330
+    },
+    {
+      "epoch": 3.33,
+      "learning_rate": 0.0013752858520457362,
+      "loss": 0.4028,
+      "step": 24340
+    },
+    {
+      "epoch": 3.34,
+      "learning_rate": 0.00137473481195757,
+      "loss": 0.2912,
+      "step": 24350
+    },
+    {
+      "epoch": 3.34,
+      "learning_rate": 0.0013741837718694034,
+      "loss": 0.3668,
+      "step": 24360
+    },
+    {
+      "epoch": 3.34,
+      "learning_rate": 0.0013736327317812372,
+      "loss": 0.3542,
+      "step": 24370
+    },
+    {
+      "epoch": 3.34,
+      "learning_rate": 0.0013730816916930706,
+      "loss": 0.3635,
+      "step": 24380
+    },
+    {
+      "epoch": 3.34,
+      "learning_rate": 0.0013725306516049042,
+      "loss": 0.333,
+      "step": 24390
+    },
+    {
+      "epoch": 3.34,
+      "learning_rate": 0.0013719796115167378,
+      "loss": 0.3084,
+      "step": 24400
+    },
+    {
+      "epoch": 3.34,
+      "eval_loss": 0.28903189301490784,
+      "eval_runtime": 1062.658,
+      "eval_samples_per_second": 10.994,
+      "eval_steps_per_second": 1.375,
+      "eval_wer": 0.5944919494223597,
+      "step": 24400
+    },
+    {
+      "epoch": 3.34,
+      "learning_rate": 0.0013714285714285714,
+      "loss": 0.4138,
+      "step": 24410
+    },
+    {
+      "epoch": 3.35,
+      "learning_rate": 0.001370877531340405,
+      "loss": 0.3165,
+      "step": 24420
+    },
+    {
+      "epoch": 3.35,
+      "learning_rate": 0.0013703264912522386,
+      "loss": 0.3825,
+      "step": 24430
+    },
+    {
+      "epoch": 3.35,
+      "learning_rate": 0.0013697754511640724,
+      "loss": 0.3512,
+      "step": 24440
+    },
+    {
+      "epoch": 3.35,
+      "learning_rate": 0.0013692244110759058,
+      "loss": 0.3323,
+      "step": 24450
+    },
+    {
+      "epoch": 3.35,
+      "learning_rate": 0.0013686733709877394,
+      "loss": 0.368,
+      "step": 24460
+    },
+    {
+      "epoch": 3.35,
+      "learning_rate": 0.001368122330899573,
+      "loss": 0.3674,
+      "step": 24470
+    },
+    {
+      "epoch": 3.35,
+      "learning_rate": 0.0013675712908114066,
+      "loss": 0.343,
+      "step": 24480
+    },
+    {
+      "epoch": 3.36,
+      "learning_rate": 0.0013670202507232402,
+      "loss": 0.3613,
+      "step": 24490
+    },
+    {
+      "epoch": 3.36,
+      "learning_rate": 0.0013664692106350738,
+      "loss": 0.3406,
+      "step": 24500
+    },
+    {
+      "epoch": 3.36,
+      "eval_loss": 0.2922148108482361,
+      "eval_runtime": 1061.106,
+      "eval_samples_per_second": 11.01,
+      "eval_steps_per_second": 1.377,
+      "eval_wer": 0.610047302829073,
+      "step": 24500
+    },
+    {
+      "epoch": 3.36,
+      "learning_rate": 0.0013659181705469074,
+      "loss": 0.3899,
+      "step": 24510
+    },
+    {
+      "epoch": 3.36,
+      "learning_rate": 0.001365367130458741,
+      "loss": 0.3469,
+      "step": 24520
+    },
+    {
+      "epoch": 3.36,
+      "learning_rate": 0.0013648160903705743,
+      "loss": 0.3626,
+      "step": 24530
+    },
+    {
+      "epoch": 3.36,
+      "learning_rate": 0.0013642650502824082,
+      "loss": 0.3259,
+      "step": 24540
+    },
+    {
+      "epoch": 3.36,
+      "learning_rate": 0.0013637140101942415,
+      "loss": 0.2798,
+      "step": 24550
+    },
+    {
+      "epoch": 3.36,
+      "learning_rate": 0.0013631629701060753,
+      "loss": 0.3315,
+      "step": 24560
+    },
+    {
+      "epoch": 3.37,
+      "learning_rate": 0.0013626119300179087,
+      "loss": 0.3539,
+      "step": 24570
+    },
+    {
+      "epoch": 3.37,
+      "learning_rate": 0.0013620608899297423,
+      "loss": 0.3379,
+      "step": 24580
+    },
+    {
+      "epoch": 3.37,
+      "learning_rate": 0.0013615098498415761,
+      "loss": 0.3573,
+      "step": 24590
+    },
+    {
+      "epoch": 3.37,
+      "learning_rate": 0.0013609588097534095,
+      "loss": 0.2947,
+      "step": 24600
+    },
+    {
+      "epoch": 3.37,
+      "eval_loss": 0.2909373641014099,
+      "eval_runtime": 1138.9666,
+      "eval_samples_per_second": 10.258,
+      "eval_steps_per_second": 1.283,
+      "eval_wer": 0.6024970435731829,
+      "step": 24600
+    },
+    {
+      "epoch": 3.37,
+      "learning_rate": 0.0013604077696652433,
+      "loss": 0.4116,
+      "step": 24610
+    },
+    {
+      "epoch": 3.37,
+      "learning_rate": 0.0013598567295770767,
+      "loss": 0.3524,
+      "step": 24620
+    },
+    {
+      "epoch": 3.37,
+      "learning_rate": 0.0013593056894889105,
+      "loss": 0.3336,
+      "step": 24630
+    },
+    {
+      "epoch": 3.38,
+      "learning_rate": 0.001358754649400744,
+      "loss": 0.3646,
+      "step": 24640
+    },
+    {
+      "epoch": 3.38,
+      "learning_rate": 0.0013582036093125775,
+      "loss": 0.2947,
+      "step": 24650
+    },
+    {
+      "epoch": 3.38,
+      "learning_rate": 0.001357652569224411,
+      "loss": 0.3618,
+      "step": 24660
+    },
+    {
+      "epoch": 3.38,
+      "learning_rate": 0.0013571015291362447,
+      "loss": 0.3248,
+      "step": 24670
+    },
+    {
+      "epoch": 3.38,
+      "learning_rate": 0.0013565504890480783,
+      "loss": 0.3481,
+      "step": 24680
+    },
+    {
+      "epoch": 3.38,
+      "learning_rate": 0.0013559994489599119,
+      "loss": 0.3934,
+      "step": 24690
+    },
+    {
+      "epoch": 3.38,
+      "learning_rate": 0.0013554484088717455,
+      "loss": 0.284,
+      "step": 24700
+    },
+    {
+      "epoch": 3.38,
+      "eval_loss": 0.2940269112586975,
+      "eval_runtime": 1078.5788,
+      "eval_samples_per_second": 10.832,
+      "eval_steps_per_second": 1.355,
+      "eval_wer": 0.592013099244974,
+      "step": 24700
+    },
+    {
+      "epoch": 3.39,
+      "learning_rate": 0.001354897368783579,
+      "loss": 0.3577,
+      "step": 24710
+    },
+    {
+      "epoch": 3.39,
+      "learning_rate": 0.0013543463286954124,
+      "loss": 0.365,
+      "step": 24720
+    },
+    {
+      "epoch": 3.39,
+      "learning_rate": 0.0013537952886072463,
+      "loss": 0.2942,
+      "step": 24730
+    },
+    {
+      "epoch": 3.39,
+      "learning_rate": 0.0013532442485190799,
+      "loss": 0.3699,
+      "step": 24740
+    },
+    {
+      "epoch": 3.39,
+      "learning_rate": 0.0013526932084309135,
+      "loss": 0.3059,
+      "step": 24750
+    },
+    {
+      "epoch": 3.39,
+      "learning_rate": 0.001352142168342747,
+      "loss": 0.3597,
+      "step": 24760
+    },
+    {
+      "epoch": 3.39,
+      "learning_rate": 0.0013515911282545804,
+      "loss": 0.3257,
+      "step": 24770
+    },
+    {
+      "epoch": 3.39,
+      "learning_rate": 0.0013510400881664142,
+      "loss": 0.3464,
+      "step": 24780
+    },
+    {
+      "epoch": 3.4,
+      "learning_rate": 0.0013504890480782476,
+      "loss": 0.3492,
+      "step": 24790
+    },
+    {
+      "epoch": 3.4,
+      "learning_rate": 0.0013499380079900814,
+      "loss": 0.2991,
+      "step": 24800
+    },
+    {
+      "epoch": 3.4,
+      "eval_loss": 0.2803167700767517,
+      "eval_runtime": 1069.809,
+      "eval_samples_per_second": 10.921,
+      "eval_steps_per_second": 1.366,
+      "eval_wer": 0.5775948330755936,
+      "step": 24800
+    },
+    {
+      "epoch": 3.4,
+      "learning_rate": 0.0013493869679019148,
+      "loss": 0.3691,
+      "step": 24810
+    },
+    {
+      "epoch": 3.4,
+      "learning_rate": 0.0013488359278137486,
+      "loss": 0.354,
+      "step": 24820
+    },
+    {
+      "epoch": 3.4,
+      "learning_rate": 0.001348284887725582,
+      "loss": 0.3624,
+      "step": 24830
+    },
+    {
+      "epoch": 3.4,
+      "learning_rate": 0.0013477338476374156,
+      "loss": 0.3352,
+      "step": 24840
+    },
+    {
+      "epoch": 3.4,
+      "learning_rate": 0.0013471828075492492,
+      "loss": 0.3011,
+      "step": 24850
+    },
+    {
+      "epoch": 3.41,
+      "learning_rate": 0.0013466317674610828,
+      "loss": 0.3592,
+      "step": 24860
+    },
+    {
+      "epoch": 3.41,
+      "learning_rate": 0.0013460807273729164,
+      "loss": 0.3621,
+      "step": 24870
+    },
+    {
+      "epoch": 3.41,
+      "learning_rate": 0.00134552968728475,
+      "loss": 0.3359,
+      "step": 24880
+    },
+    {
+      "epoch": 3.41,
+      "learning_rate": 0.0013449786471965834,
+      "loss": 0.3516,
+      "step": 24890
+    },
+    {
+      "epoch": 3.41,
+      "learning_rate": 0.0013444276071084172,
+      "loss": 0.3419,
+      "step": 24900
+    },
+    {
+      "epoch": 3.41,
+      "eval_loss": 0.273798406124115,
+      "eval_runtime": 1066.2109,
+      "eval_samples_per_second": 10.957,
+      "eval_steps_per_second": 1.37,
+      "eval_wer": 0.5698285272446102,
+      "step": 24900
+    },
+    {
+      "epoch": 3.41,
+      "learning_rate": 0.0013438765670202508,
+      "loss": 0.3429,
+      "step": 24910
+    },
+    {
+      "epoch": 3.41,
+      "learning_rate": 0.0013433255269320844,
+      "loss": 0.3922,
+      "step": 24920
+    },
+    {
+      "epoch": 3.42,
+      "learning_rate": 0.001342774486843918,
+      "loss": 0.3287,
+      "step": 24930
+    },
+    {
+      "epoch": 3.42,
+      "learning_rate": 0.0013422234467557516,
+      "loss": 0.3662,
+      "step": 24940
+    },
+    {
+      "epoch": 3.42,
+      "learning_rate": 0.0013416724066675852,
+      "loss": 0.3308,
+      "step": 24950
+    },
+    {
+      "epoch": 3.42,
+      "learning_rate": 0.0013411213665794185,
+      "loss": 0.3322,
+      "step": 24960
+    },
+    {
+      "epoch": 3.42,
+      "learning_rate": 0.0013405703264912523,
+      "loss": 0.3569,
+      "step": 24970
+    },
+    {
+      "epoch": 3.42,
+      "learning_rate": 0.0013400192864030857,
+      "loss": 0.3487,
+      "step": 24980
+    },
+    {
+      "epoch": 3.42,
+      "learning_rate": 0.0013394682463149195,
+      "loss": 0.2873,
+      "step": 24990
+    },
+    {
+      "epoch": 3.43,
+      "learning_rate": 0.001338917206226753,
+      "loss": 0.3274,
+      "step": 25000
+    },
+    {
+      "epoch": 3.43,
+      "eval_loss": 0.2817874848842621,
+      "eval_runtime": 1064.2829,
+      "eval_samples_per_second": 10.977,
+      "eval_steps_per_second": 1.373,
+      "eval_wer": 0.5924906758846539,
+      "step": 25000
+    },
+    {
+      "epoch": 3.43,
+      "learning_rate": 0.0013383661661385867,
+      "loss": 0.3584,
+      "step": 25010
+    },
+    {
+      "epoch": 3.43,
+      "learning_rate": 0.0013378151260504201,
+      "loss": 0.322,
+      "step": 25020
+    },
+    {
+      "epoch": 3.43,
+      "learning_rate": 0.0013372640859622537,
+      "loss": 0.3526,
+      "step": 25030
+    },
+    {
+      "epoch": 3.43,
+      "learning_rate": 0.0013367130458740873,
+      "loss": 0.3369,
+      "step": 25040
+    },
+    {
+      "epoch": 3.43,
+      "learning_rate": 0.001336162005785921,
+      "loss": 0.3127,
+      "step": 25050
+    },
+    {
+      "epoch": 3.43,
+      "learning_rate": 0.0013356109656977547,
+      "loss": 0.3429,
+      "step": 25060
+    },
+    {
+      "epoch": 3.43,
+      "learning_rate": 0.001335059925609588,
+      "loss": 0.2917,
+      "step": 25070
+    },
+    {
+      "epoch": 3.44,
+      "learning_rate": 0.001334508885521422,
+      "loss": 0.334,
+      "step": 25080
+    },
+    {
+      "epoch": 3.44,
+      "learning_rate": 0.0013339578454332553,
+      "loss": 0.3375,
+      "step": 25090
+    },
+    {
+      "epoch": 3.44,
+      "learning_rate": 0.0013334068053450889,
+      "loss": 0.2594,
+      "step": 25100
+    },
+    {
+      "epoch": 3.44,
+      "eval_loss": 0.2670289874076843,
+      "eval_runtime": 1064.7523,
+      "eval_samples_per_second": 10.973,
+      "eval_steps_per_second": 1.372,
+      "eval_wer": 0.5709542436095697,
+      "step": 25100
+    },
+    {
+      "epoch": 3.44,
+      "learning_rate": 0.0013328557652569225,
+      "loss": 0.3653,
+      "step": 25110
+    },
+    {
+      "epoch": 3.44,
+      "learning_rate": 0.001332304725168756,
+      "loss": 0.3423,
+      "step": 25120
+    },
+    {
+      "epoch": 3.44,
+      "learning_rate": 0.0013317536850805897,
+      "loss": 0.3087,
+      "step": 25130
+    },
+    {
+      "epoch": 3.44,
+      "learning_rate": 0.0013312026449924233,
+      "loss": 0.3469,
+      "step": 25140
+    },
+    {
+      "epoch": 3.45,
+      "learning_rate": 0.0013306516049042566,
+      "loss": 0.3374,
+      "step": 25150
+    },
+    {
+      "epoch": 3.45,
+      "learning_rate": 0.0013301005648160905,
+      "loss": 0.3376,
+      "step": 25160
+    },
+    {
+      "epoch": 3.45,
+      "learning_rate": 0.0013295495247279238,
+      "loss": 0.34,
+      "step": 25170
+    },
+    {
+      "epoch": 3.45,
+      "learning_rate": 0.0013289984846397577,
+      "loss": 0.3612,
+      "step": 25180
+    },
+    {
+      "epoch": 3.45,
+      "learning_rate": 0.001328447444551591,
+      "loss": 0.3192,
+      "step": 25190
+    },
+    {
+      "epoch": 3.45,
+      "learning_rate": 0.0013278964044634248,
+      "loss": 0.3437,
+      "step": 25200
+    },
+    {
+      "epoch": 3.45,
+      "eval_loss": 0.2782919406890869,
+      "eval_runtime": 1066.3915,
+      "eval_samples_per_second": 10.956,
+      "eval_steps_per_second": 1.37,
+      "eval_wer": 0.573421722914582,
+      "step": 25200
+    },
+    {
+      "epoch": 3.45,
+      "learning_rate": 0.0013273453643752584,
+      "loss": 0.3415,
+      "step": 25210
+    },
+    {
+      "epoch": 3.46,
+      "learning_rate": 0.0013267943242870918,
+      "loss": 0.3306,
+      "step": 25220
+    },
+    {
+      "epoch": 3.46,
+      "learning_rate": 0.0013262432841989256,
+      "loss": 0.3411,
+      "step": 25230
+    },
+    {
+      "epoch": 3.46,
+      "learning_rate": 0.001325692244110759,
+      "loss": 0.345,
+      "step": 25240
+    },
+    {
+      "epoch": 3.46,
+      "learning_rate": 0.0013251412040225928,
+      "loss": 0.3189,
+      "step": 25250
+    },
+    {
+      "epoch": 3.46,
+      "learning_rate": 0.0013245901639344262,
+      "loss": 0.3251,
+      "step": 25260
+    },
+    {
+      "epoch": 3.46,
+      "learning_rate": 0.00132403912384626,
+      "loss": 0.3675,
+      "step": 25270
+    },
+    {
+      "epoch": 3.46,
+      "learning_rate": 0.0013234880837580934,
+      "loss": 0.3647,
+      "step": 25280
+    },
+    {
+      "epoch": 3.46,
+      "learning_rate": 0.001322937043669927,
+      "loss": 0.3582,
+      "step": 25290
+    },
+    {
+      "epoch": 3.47,
+      "learning_rate": 0.0013223860035817606,
+      "loss": 0.2723,
+      "step": 25300
+    },
+    {
+      "epoch": 3.47,
+      "eval_loss": 0.2774510383605957,
+      "eval_runtime": 1066.6539,
+      "eval_samples_per_second": 10.953,
+      "eval_steps_per_second": 1.37,
+      "eval_wer": 0.5868279814427363,
+      "step": 25300
+    },
+    {
+      "epoch": 3.47,
+      "learning_rate": 0.0013218349634935942,
+      "loss": 0.3642,
+      "step": 25310
+    },
+    {
+      "epoch": 3.47,
+      "learning_rate": 0.0013212839234054278,
+      "loss": 0.3377,
+      "step": 25320
+    },
+    {
+      "epoch": 3.47,
+      "learning_rate": 0.0013207328833172614,
+      "loss": 0.33,
+      "step": 25330
+    },
+    {
+      "epoch": 3.47,
+      "learning_rate": 0.0013201818432290948,
+      "loss": 0.3416,
+      "step": 25340
+    },
+    {
+      "epoch": 3.47,
+      "learning_rate": 0.0013196308031409286,
+      "loss": 0.2675,
+      "step": 25350
+    },
+    {
+      "epoch": 3.47,
+      "learning_rate": 0.001319079763052762,
+      "loss": 0.312,
+      "step": 25360
+    },
+    {
+      "epoch": 3.48,
+      "learning_rate": 0.0013185287229645958,
+      "loss": 0.3344,
+      "step": 25370
+    },
+    {
+      "epoch": 3.48,
+      "learning_rate": 0.0013179776828764294,
+      "loss": 0.3344,
+      "step": 25380
+    },
+    {
+      "epoch": 3.48,
+      "learning_rate": 0.001317426642788263,
+      "loss": 0.3563,
+      "step": 25390
+    },
+    {
+      "epoch": 3.48,
+      "learning_rate": 0.0013168756027000965,
+      "loss": 0.2916,
+      "step": 25400
+    },
+    {
+      "epoch": 3.48,
+      "eval_loss": 0.2820914089679718,
+      "eval_runtime": 1082.9639,
+      "eval_samples_per_second": 10.788,
+      "eval_steps_per_second": 1.349,
+      "eval_wer": 0.5866687892295097,
+      "step": 25400
+    },
+    {
+      "epoch": 3.48,
+      "learning_rate": 0.00131632456261193,
+      "loss": 0.3472,
+      "step": 25410
+    },
+    {
+      "epoch": 3.48,
+      "learning_rate": 0.0013157735225237637,
+      "loss": 0.3439,
+      "step": 25420
+    },
+    {
+      "epoch": 3.48,
+      "learning_rate": 0.0013152224824355971,
+      "loss": 0.3368,
+      "step": 25430
+    },
+    {
+      "epoch": 3.49,
+      "learning_rate": 0.001314671442347431,
+      "loss": 0.369,
+      "step": 25440
+    },
+    {
+      "epoch": 3.49,
+      "learning_rate": 0.0013141204022592643,
+      "loss": 0.2785,
+      "step": 25450
+    },
+    {
+      "epoch": 3.49,
+      "learning_rate": 0.001313569362171098,
+      "loss": 0.3453,
+      "step": 25460
+    },
+    {
+      "epoch": 3.49,
+      "learning_rate": 0.0013130183220829315,
+      "loss": 0.3067,
+      "step": 25470
+    },
+    {
+      "epoch": 3.49,
+      "learning_rate": 0.001312467281994765,
+      "loss": 0.3378,
+      "step": 25480
+    },
+    {
+      "epoch": 3.49,
+      "learning_rate": 0.0013119162419065987,
+      "loss": 0.3443,
+      "step": 25490
+    },
+    {
+      "epoch": 3.49,
+      "learning_rate": 0.0013113652018184323,
+      "loss": 0.2438,
+      "step": 25500
+    },
+    {
+      "epoch": 3.49,
+      "eval_loss": 0.26588869094848633,
+      "eval_runtime": 1065.3407,
+      "eval_samples_per_second": 10.966,
+      "eval_steps_per_second": 1.371,
+      "eval_wer": 0.5658259801691986,
+      "step": 25500
+    },
+    {
+      "epoch": 3.49,
+      "learning_rate": 0.0013108141617302659,
+      "loss": 0.36,
+      "step": 25510
+    },
+    {
+      "epoch": 3.5,
+      "learning_rate": 0.0013102631216420995,
+      "loss": 0.3464,
+      "step": 25520
+    },
+    {
+      "epoch": 3.5,
+      "learning_rate": 0.001309712081553933,
+      "loss": 0.3083,
+      "step": 25530
+    },
+    {
+      "epoch": 3.5,
+      "learning_rate": 0.0013091610414657667,
+      "loss": 0.3428,
+      "step": 25540
+    },
+    {
+      "epoch": 3.5,
+      "learning_rate": 0.0013086100013776003,
+      "loss": 0.2786,
+      "step": 25550
+    },
+    {
+      "epoch": 3.5,
+      "learning_rate": 0.0013080589612894339,
+      "loss": 0.34,
+      "step": 25560
+    },
+    {
+      "epoch": 3.5,
+      "learning_rate": 0.0013075079212012675,
+      "loss": 0.346,
+      "step": 25570
+    },
+    {
+      "epoch": 3.5,
+      "learning_rate": 0.001306956881113101,
+      "loss": 0.3433,
+      "step": 25580
+    },
+    {
+      "epoch": 3.51,
+      "learning_rate": 0.0013064058410249347,
+      "loss": 0.3372,
+      "step": 25590
+    },
+    {
+      "epoch": 3.51,
+      "learning_rate": 0.001305854800936768,
+      "loss": 0.3299,
+      "step": 25600
+    },
+    {
+      "epoch": 3.51,
+      "eval_loss": 0.2781725823879242,
+      "eval_runtime": 1068.6136,
+      "eval_samples_per_second": 10.933,
+      "eval_steps_per_second": 1.367,
+      "eval_wer": 0.5854521058855635,
+      "step": 25600
+    },
+    {
+      "epoch": 3.51,
+      "learning_rate": 0.0013053037608486018,
+      "loss": 0.3525,
+      "step": 25610
+    },
+    {
+      "epoch": 3.51,
+      "learning_rate": 0.0013047527207604352,
+      "loss": 0.3272,
+      "step": 25620
+    },
+    {
+      "epoch": 3.51,
+      "learning_rate": 0.001304201680672269,
+      "loss": 0.3229,
+      "step": 25630
+    },
+    {
+      "epoch": 3.51,
+      "learning_rate": 0.0013036506405841024,
+      "loss": 0.4146,
+      "step": 25640
+    },
+    {
+      "epoch": 3.51,
+      "learning_rate": 0.001303099600495936,
+      "loss": 0.3182,
+      "step": 25650
+    },
+    {
+      "epoch": 3.52,
+      "learning_rate": 0.0013025485604077696,
+      "loss": 0.3443,
+      "step": 25660
+    },
+    {
+      "epoch": 3.52,
+      "learning_rate": 0.0013019975203196032,
+      "loss": 0.3564,
+      "step": 25670
+    },
+    {
+      "epoch": 3.52,
+      "learning_rate": 0.001301446480231437,
+      "loss": 0.3588,
+      "step": 25680
+    },
+    {
+      "epoch": 3.52,
+      "learning_rate": 0.0013008954401432704,
+      "loss": 0.3267,
+      "step": 25690
+    },
+    {
+      "epoch": 3.52,
+      "learning_rate": 0.0013003444000551042,
+      "loss": 0.3268,
+      "step": 25700
+    },
+    {
+      "epoch": 3.52,
+      "eval_loss": 0.28613343834877014,
+      "eval_runtime": 1069.5103,
+      "eval_samples_per_second": 10.924,
+      "eval_steps_per_second": 1.366,
+      "eval_wer": 0.5814495588101519,
+      "step": 25700
+    },
+    {
+      "epoch": 3.52,
+      "learning_rate": 0.0012997933599669376,
+      "loss": 0.3581,
+      "step": 25710
+    },
+    {
+      "epoch": 3.52,
+      "learning_rate": 0.0012992423198787712,
+      "loss": 0.2963,
+      "step": 25720
+    },
+    {
+      "epoch": 3.53,
+      "learning_rate": 0.0012986912797906048,
+      "loss": 0.3162,
+      "step": 25730
+    },
+    {
+      "epoch": 3.53,
+      "learning_rate": 0.0012981402397024384,
+      "loss": 0.3295,
+      "step": 25740
+    },
+    {
+      "epoch": 3.53,
+      "learning_rate": 0.0012976443036230886,
+      "loss": 0.3185,
+      "step": 25750
+    },
+    {
+      "epoch": 3.53,
+      "learning_rate": 0.0012970932635349222,
+      "loss": 0.3742,
+      "step": 25760
+    },
+    {
+      "epoch": 3.53,
+      "learning_rate": 0.0012965422234467558,
+      "loss": 0.3541,
+      "step": 25770
+    },
+    {
+      "epoch": 3.53,
+      "learning_rate": 0.0012959911833585894,
+      "loss": 0.2967,
+      "step": 25780
+    },
+    {
+      "epoch": 3.53,
+      "learning_rate": 0.001295440143270423,
+      "loss": 0.3476,
+      "step": 25790
+    },
+    {
+      "epoch": 3.53,
+      "learning_rate": 0.0012948891031822566,
+      "loss": 0.2712,
+      "step": 25800
+    },
+    {
+      "epoch": 3.53,
+      "eval_loss": 0.28026676177978516,
+      "eval_runtime": 1065.6019,
+      "eval_samples_per_second": 10.964,
+      "eval_steps_per_second": 1.371,
+      "eval_wer": 0.5814040753206586,
+      "step": 25800
+    },
+    {
+      "epoch": 3.54,
+      "learning_rate": 0.0012943380630940902,
+      "loss": 0.3388,
+      "step": 25810
+    },
+    {
+      "epoch": 3.54,
+      "learning_rate": 0.0012937870230059236,
+      "loss": 0.3656,
+      "step": 25820
+    },
+    {
+      "epoch": 3.54,
+      "learning_rate": 0.0012932359829177574,
+      "loss": 0.333,
+      "step": 25830
+    },
+    {
+      "epoch": 3.54,
+      "learning_rate": 0.0012926849428295908,
+      "loss": 0.3281,
+      "step": 25840
+    },
+    {
+      "epoch": 3.54,
+      "learning_rate": 0.0012921339027414246,
+      "loss": 0.3265,
+      "step": 25850
+    },
+    {
+      "epoch": 3.54,
+      "learning_rate": 0.001291582862653258,
+      "loss": 0.3484,
+      "step": 25860
+    },
+    {
+      "epoch": 3.54,
+      "learning_rate": 0.0012910318225650916,
+      "loss": 0.3423,
+      "step": 25870
+    },
+    {
+      "epoch": 3.55,
+      "learning_rate": 0.0012904807824769252,
+      "loss": 0.3204,
+      "step": 25880
+    },
+    {
+      "epoch": 3.55,
+      "learning_rate": 0.0012899297423887587,
+      "loss": 0.4087,
+      "step": 25890
+    },
+    {
+      "epoch": 3.55,
+      "learning_rate": 0.0012893787023005923,
+      "loss": 0.3433,
+      "step": 25900
+    },
+    {
+      "epoch": 3.55,
+      "eval_loss": 0.2783382534980774,
+      "eval_runtime": 1070.8569,
+      "eval_samples_per_second": 10.91,
+      "eval_steps_per_second": 1.364,
+      "eval_wer": 0.5810402074047121,
+      "step": 25900
+    },
+    {
+      "epoch": 3.55,
+      "learning_rate": 0.001288827662212426,
+      "loss": 0.3658,
+      "step": 25910
+    },
+    {
+      "epoch": 3.55,
+      "learning_rate": 0.0012882766221242598,
+      "loss": 0.3448,
+      "step": 25920
+    },
+    {
+      "epoch": 3.55,
+      "learning_rate": 0.0012877255820360931,
+      "loss": 0.3006,
+      "step": 25930
+    },
+    {
+      "epoch": 3.55,
+      "learning_rate": 0.0012871745419479267,
+      "loss": 0.3893,
+      "step": 25940
+    },
+    {
+      "epoch": 3.56,
+      "learning_rate": 0.0012866235018597603,
+      "loss": 0.2934,
+      "step": 25950
+    },
+    {
+      "epoch": 3.56,
+      "learning_rate": 0.001286072461771594,
+      "loss": 0.3352,
+      "step": 25960
+    },
+    {
+      "epoch": 3.56,
+      "learning_rate": 0.0012855214216834275,
+      "loss": 0.4038,
+      "step": 25970
+    },
+    {
+      "epoch": 3.56,
+      "learning_rate": 0.0012849703815952611,
+      "loss": 0.3604,
+      "step": 25980
+    },
+    {
+      "epoch": 3.56,
+      "learning_rate": 0.0012844193415070947,
+      "loss": 0.3471,
+      "step": 25990
+    },
+    {
+      "epoch": 3.56,
+      "learning_rate": 0.0012838683014189283,
+      "loss": 0.3374,
+      "step": 26000
+    },
+    {
+      "epoch": 3.56,
+      "eval_loss": 0.2748092710971832,
+      "eval_runtime": 1064.9581,
+      "eval_samples_per_second": 10.97,
+      "eval_steps_per_second": 1.372,
+      "eval_wer": 0.5731147093605021,
+      "step": 26000
+    },
+    {
+      "epoch": 3.56,
+      "learning_rate": 0.0012833172613307617,
+      "loss": 0.3539,
+      "step": 26010
+    },
+    {
+      "epoch": 3.56,
+      "learning_rate": 0.0012827662212425955,
+      "loss": 0.367,
+      "step": 26020
+    },
+    {
+      "epoch": 3.57,
+      "learning_rate": 0.0012822151811544289,
+      "loss": 0.3308,
+      "step": 26030
+    },
+    {
+      "epoch": 3.57,
+      "learning_rate": 0.0012816641410662627,
+      "loss": 0.3129,
+      "step": 26040
+    },
+    {
+      "epoch": 3.57,
+      "learning_rate": 0.001281113100978096,
+      "loss": 0.3088,
+      "step": 26050
+    },
+    {
+      "epoch": 3.57,
+      "learning_rate": 0.0012805620608899297,
+      "loss": 0.373,
+      "step": 26060
+    },
+    {
+      "epoch": 3.57,
+      "learning_rate": 0.0012800110208017635,
+      "loss": 0.3417,
+      "step": 26070
+    },
+    {
+      "epoch": 3.57,
+      "learning_rate": 0.0012794599807135969,
+      "loss": 0.3431,
+      "step": 26080
+    },
+    {
+      "epoch": 3.57,
+      "learning_rate": 0.0012789089406254307,
+      "loss": 0.3663,
+      "step": 26090
+    },
+    {
+      "epoch": 3.58,
+      "learning_rate": 0.001278357900537264,
+      "loss": 0.3253,
+      "step": 26100
+    },
+    {
+      "epoch": 3.58,
+      "eval_loss": 0.2813658118247986,
+      "eval_runtime": 1067.18,
+      "eval_samples_per_second": 10.948,
+      "eval_steps_per_second": 1.369,
+      "eval_wer": 0.5847584826707906,
+      "step": 26100
+    },
+    {
+      "epoch": 3.58,
+      "learning_rate": 0.0012778068604490979,
+      "loss": 0.3831,
+      "step": 26110
+    },
+    {
+      "epoch": 3.58,
+      "learning_rate": 0.0012772558203609312,
+      "loss": 0.29,
+      "step": 26120
+    },
+    {
+      "epoch": 3.58,
+      "learning_rate": 0.0012767047802727648,
+      "loss": 0.3629,
+      "step": 26130
+    },
+    {
+      "epoch": 3.58,
+      "learning_rate": 0.0012761537401845984,
+      "loss": 0.3558,
+      "step": 26140
+    },
+    {
+      "epoch": 3.58,
+      "learning_rate": 0.001275602700096432,
+      "loss": 0.3459,
+      "step": 26150
+    },
+    {
+      "epoch": 3.58,
+      "learning_rate": 0.0012750516600082656,
+      "loss": 0.3461,
+      "step": 26160
+    },
+    {
+      "epoch": 3.59,
+      "learning_rate": 0.0012745006199200992,
+      "loss": 0.327,
+      "step": 26170
+    },
+    {
+      "epoch": 3.59,
+      "learning_rate": 0.0012739495798319328,
+      "loss": 0.3268,
+      "step": 26180
+    },
+    {
+      "epoch": 3.59,
+      "learning_rate": 0.0012733985397437664,
+      "loss": 0.3473,
+      "step": 26190
+    },
+    {
+      "epoch": 3.59,
+      "learning_rate": 0.0012728474996555998,
+      "loss": 0.2877,
+      "step": 26200
+    },
+    {
+      "epoch": 3.59,
+      "eval_loss": 0.2731098532676697,
+      "eval_runtime": 1069.823,
+      "eval_samples_per_second": 10.92,
+      "eval_steps_per_second": 1.366,
+      "eval_wer": 0.5704539252251433,
+      "step": 26200
+    },
+    {
+      "epoch": 3.59,
+      "learning_rate": 0.0012722964595674336,
+      "loss": 0.3762,
+      "step": 26210
+    },
+    {
+      "epoch": 3.59,
+      "learning_rate": 0.0012717454194792672,
+      "loss": 0.3472,
+      "step": 26220
+    },
+    {
+      "epoch": 3.59,
+      "learning_rate": 0.0012711943793911008,
+      "loss": 0.3558,
+      "step": 26230
+    },
+    {
+      "epoch": 3.6,
+      "learning_rate": 0.0012706433393029344,
+      "loss": 0.3574,
+      "step": 26240
+    },
+    {
+      "epoch": 3.6,
+      "learning_rate": 0.0012700922992147678,
+      "loss": 0.3276,
+      "step": 26250
+    },
+    {
+      "epoch": 3.6,
+      "learning_rate": 0.0012695412591266016,
+      "loss": 0.3149,
+      "step": 26260
+    },
+    {
+      "epoch": 3.6,
+      "learning_rate": 0.001268990219038435,
+      "loss": 0.3303,
+      "step": 26270
+    },
+    {
+      "epoch": 3.6,
+      "learning_rate": 0.0012684391789502688,
+      "loss": 0.3379,
+      "step": 26280
+    },
+    {
+      "epoch": 3.6,
+      "learning_rate": 0.0012678881388621022,
+      "loss": 0.3485,
+      "step": 26290
+    },
+    {
+      "epoch": 3.6,
+      "learning_rate": 0.001267337098773936,
+      "loss": 0.3398,
+      "step": 26300
+    },
+    {
+      "epoch": 3.6,
+      "eval_loss": 0.2784402072429657,
+      "eval_runtime": 1069.7894,
+      "eval_samples_per_second": 10.921,
+      "eval_steps_per_second": 1.366,
+      "eval_wer": 0.5902847266442282,
+      "step": 26300
+    },
+    {
+      "epoch": 3.6,
+      "learning_rate": 0.0012667860586857693,
+      "loss": 0.3699,
+      "step": 26310
+    },
+    {
+      "epoch": 3.61,
+      "learning_rate": 0.001266235018597603,
+      "loss": 0.3024,
+      "step": 26320
+    },
+    {
+      "epoch": 3.61,
+      "learning_rate": 0.0012656839785094365,
+      "loss": 0.3098,
+      "step": 26330
+    },
+    {
+      "epoch": 3.61,
+      "learning_rate": 0.0012651329384212701,
+      "loss": 0.3379,
+      "step": 26340
+    },
+    {
+      "epoch": 3.61,
+      "learning_rate": 0.0012645818983331037,
+      "loss": 0.3289,
+      "step": 26350
+    },
+    {
+      "epoch": 3.61,
+      "learning_rate": 0.0012640308582449373,
+      "loss": 0.3574,
+      "step": 26360
+    },
+    {
+      "epoch": 3.61,
+      "learning_rate": 0.001263479818156771,
+      "loss": 0.339,
+      "step": 26370
+    },
+    {
+      "epoch": 3.61,
+      "learning_rate": 0.0012629287780686045,
+      "loss": 0.3399,
+      "step": 26380
+    },
+    {
+      "epoch": 3.62,
+      "learning_rate": 0.0012623777379804381,
+      "loss": 0.3199,
+      "step": 26390
+    },
+    {
+      "epoch": 3.62,
+      "learning_rate": 0.0012618266978922717,
+      "loss": 0.3244,
+      "step": 26400
+    },
+    {
+      "epoch": 3.62,
+      "eval_loss": 0.2925315797328949,
+      "eval_runtime": 1067.0088,
+      "eval_samples_per_second": 10.949,
+      "eval_steps_per_second": 1.369,
+      "eval_wer": 0.5857022650777768,
+      "step": 26400
+    },
+    {
+      "epoch": 3.62,
+      "learning_rate": 0.0012612756578041053,
+      "loss": 0.4002,
+      "step": 26410
+    },
+    {
+      "epoch": 3.62,
+      "learning_rate": 0.001260724617715939,
+      "loss": 0.3188,
+      "step": 26420
+    },
+    {
+      "epoch": 3.62,
+      "learning_rate": 0.0012601735776277725,
+      "loss": 0.3391,
+      "step": 26430
+    },
+    {
+      "epoch": 3.62,
+      "learning_rate": 0.0012596225375396059,
+      "loss": 0.3425,
+      "step": 26440
+    },
+    {
+      "epoch": 3.62,
+      "learning_rate": 0.0012590714974514397,
+      "loss": 0.3165,
+      "step": 26450
+    },
+    {
+      "epoch": 3.63,
+      "learning_rate": 0.001258520457363273,
+      "loss": 0.3585,
+      "step": 26460
+    },
+    {
+      "epoch": 3.63,
+      "learning_rate": 0.0012579694172751069,
+      "loss": 0.3201,
+      "step": 26470
+    },
+    {
+      "epoch": 3.63,
+      "learning_rate": 0.0012574183771869403,
+      "loss": 0.3172,
+      "step": 26480
+    },
+    {
+      "epoch": 3.63,
+      "learning_rate": 0.001256867337098774,
+      "loss": 0.312,
+      "step": 26490
+    },
+    {
+      "epoch": 3.63,
+      "learning_rate": 0.0012563162970106075,
+      "loss": 0.3231,
+      "step": 26500
+    },
+    {
+      "epoch": 3.63,
+      "eval_loss": 0.26204758882522583,
+      "eval_runtime": 1067.117,
+      "eval_samples_per_second": 10.948,
+      "eval_steps_per_second": 1.369,
+      "eval_wer": 0.5660533976166652,
+      "step": 26500
+    },
+    {
+      "epoch": 3.63,
+      "learning_rate": 0.001255765256922441,
+      "loss": 0.3811,
+      "step": 26510
+    },
+    {
+      "epoch": 3.63,
+      "learning_rate": 0.0012552142168342747,
+      "loss": 0.439,
+      "step": 26520
+    },
+    {
+      "epoch": 3.63,
+      "learning_rate": 0.0012546631767461082,
+      "loss": 0.3631,
+      "step": 26530
+    },
+    {
+      "epoch": 3.64,
+      "learning_rate": 0.001254112136657942,
+      "loss": 0.3584,
+      "step": 26540
+    },
+    {
+      "epoch": 3.64,
+      "learning_rate": 0.0012535610965697754,
+      "loss": 0.3487,
+      "step": 26550
+    },
+    {
+      "epoch": 3.64,
+      "learning_rate": 0.0012530100564816093,
+      "loss": 0.3404,
+      "step": 26560
+    },
+    {
+      "epoch": 3.64,
+      "learning_rate": 0.0012524590163934426,
+      "loss": 0.3422,
+      "step": 26570
+    },
+    {
+      "epoch": 3.64,
+      "learning_rate": 0.0012519079763052762,
+      "loss": 0.3225,
+      "step": 26580
+    },
+    {
+      "epoch": 3.64,
+      "learning_rate": 0.0012513569362171098,
+      "loss": 0.3606,
+      "step": 26590
+    },
+    {
+      "epoch": 3.64,
+      "learning_rate": 0.0012508058961289434,
+      "loss": 0.2607,
+      "step": 26600
+    },
+    {
+      "epoch": 3.64,
+      "eval_loss": 0.2568303048610687,
+      "eval_runtime": 1068.4594,
+      "eval_samples_per_second": 10.934,
+      "eval_steps_per_second": 1.367,
+      "eval_wer": 0.5645524424633858,
+      "step": 26600
+    },
+    {
+      "epoch": 3.65,
+      "learning_rate": 0.001250254856040777,
+      "loss": 0.3369,
+      "step": 26610
+    },
+    {
+      "epoch": 3.65,
+      "learning_rate": 0.0012497038159526106,
+      "loss": 0.3271,
+      "step": 26620
+    },
+    {
+      "epoch": 3.65,
+      "learning_rate": 0.001249152775864444,
+      "loss": 0.3337,
+      "step": 26630
+    },
+    {
+      "epoch": 3.65,
+      "learning_rate": 0.0012486017357762778,
+      "loss": 0.3486,
+      "step": 26640
+    },
+    {
+      "epoch": 3.65,
+      "learning_rate": 0.0012480506956881112,
+      "loss": 0.2697,
+      "step": 26650
+    },
+    {
+      "epoch": 3.65,
+      "learning_rate": 0.001247499655599945,
+      "loss": 0.3307,
+      "step": 26660
+    },
+    {
+      "epoch": 3.65,
+      "learning_rate": 0.0012469486155117784,
+      "loss": 0.3767,
+      "step": 26670
+    },
+    {
+      "epoch": 3.66,
+      "learning_rate": 0.0012463975754236122,
+      "loss": 0.3103,
+      "step": 26680
+    },
+    {
+      "epoch": 3.66,
+      "learning_rate": 0.0012458465353354458,
+      "loss": 0.3233,
+      "step": 26690
+    },
+    {
+      "epoch": 3.66,
+      "learning_rate": 0.0012452954952472792,
+      "loss": 0.306,
+      "step": 26700
+    },
+    {
+      "epoch": 3.66,
+      "eval_loss": 0.26539793610572815,
+      "eval_runtime": 1066.531,
+      "eval_samples_per_second": 10.954,
+      "eval_steps_per_second": 1.37,
+      "eval_wer": 0.5638474483762395,
+      "step": 26700
+    },
+    {
+      "epoch": 3.66,
+      "learning_rate": 0.001244744455159113,
+      "loss": 0.3709,
+      "step": 26710
+    },
+    {
+      "epoch": 3.66,
+      "learning_rate": 0.0012441934150709464,
+      "loss": 0.3206,
+      "step": 26720
+    },
+    {
+      "epoch": 3.66,
+      "learning_rate": 0.0012436423749827802,
+      "loss": 0.3145,
+      "step": 26730
+    },
+    {
+      "epoch": 3.66,
+      "learning_rate": 0.0012430913348946135,
+      "loss": 0.3225,
+      "step": 26740
+    },
+    {
+      "epoch": 3.66,
+      "learning_rate": 0.0012425402948064474,
+      "loss": 0.2283,
+      "step": 26750
+    },
+    {
+      "epoch": 3.67,
+      "learning_rate": 0.0012419892547182807,
+      "loss": 0.3597,
+      "step": 26760
+    },
+    {
+      "epoch": 3.67,
+      "learning_rate": 0.0012414382146301143,
+      "loss": 0.3249,
+      "step": 26770
+    },
+    {
+      "epoch": 3.67,
+      "learning_rate": 0.001240887174541948,
+      "loss": 0.3453,
+      "step": 26780
+    },
+    {
+      "epoch": 3.67,
+      "learning_rate": 0.0012403361344537815,
+      "loss": 0.3126,
+      "step": 26790
+    },
+    {
+      "epoch": 3.67,
+      "learning_rate": 0.0012397850943656151,
+      "loss": 0.3182,
+      "step": 26800
+    },
+    {
+      "epoch": 3.67,
+      "eval_loss": 0.2721990942955017,
+      "eval_runtime": 1067.2255,
+      "eval_samples_per_second": 10.947,
+      "eval_steps_per_second": 1.369,
+      "eval_wer": 0.5665878286182116,
+      "step": 26800
+    },
+    {
+      "epoch": 3.67,
+      "learning_rate": 0.0012392340542774487,
+      "loss": 0.3371,
+      "step": 26810
+    },
+    {
+      "epoch": 3.67,
+      "learning_rate": 0.001238683014189282,
+      "loss": 0.3201,
+      "step": 26820
+    },
+    {
+      "epoch": 3.68,
+      "learning_rate": 0.001238131974101116,
+      "loss": 0.323,
+      "step": 26830
+    },
+    {
+      "epoch": 3.68,
+      "learning_rate": 0.0012375809340129495,
+      "loss": 0.3187,
+      "step": 26840
+    },
+    {
+      "epoch": 3.68,
+      "learning_rate": 0.001237029893924783,
+      "loss": 0.2758,
+      "step": 26850
+    },
+    {
+      "epoch": 3.68,
+      "learning_rate": 0.0012364788538366167,
+      "loss": 0.3637,
+      "step": 26860
+    },
+    {
+      "epoch": 3.68,
+      "learning_rate": 0.0012359278137484503,
+      "loss": 0.3118,
+      "step": 26870
+    },
+    {
+      "epoch": 3.68,
+      "learning_rate": 0.001235376773660284,
+      "loss": 0.3056,
+      "step": 26880
+    },
+    {
+      "epoch": 3.68,
+      "learning_rate": 0.0012348257335721173,
+      "loss": 0.3306,
+      "step": 26890
+    },
+    {
+      "epoch": 3.69,
+      "learning_rate": 0.001234274693483951,
+      "loss": 0.295,
+      "step": 26900
+    },
+    {
+      "epoch": 3.69,
+      "eval_loss": 0.24868972599506378,
+      "eval_runtime": 1068.2731,
+      "eval_samples_per_second": 10.936,
+      "eval_steps_per_second": 1.368,
+      "eval_wer": 0.5453242972800874,
+      "step": 26900
+    },
+    {
+      "epoch": 3.69,
+      "learning_rate": 0.0012337236533957845,
+      "loss": 0.3203,
+      "step": 26910
+    },
+    {
+      "epoch": 3.69,
+      "learning_rate": 0.0012331726133076183,
+      "loss": 0.3447,
+      "step": 26920
+    },
+    {
+      "epoch": 3.69,
+      "learning_rate": 0.0012326215732194517,
+      "loss": 0.3227,
+      "step": 26930
+    },
+    {
+      "epoch": 3.69,
+      "learning_rate": 0.0012320705331312855,
+      "loss": 0.3622,
+      "step": 26940
+    },
+    {
+      "epoch": 3.69,
+      "learning_rate": 0.0012315194930431188,
+      "loss": 0.255,
+      "step": 26950
+    },
+    {
+      "epoch": 3.69,
+      "learning_rate": 0.0012309684529549524,
+      "loss": 0.3568,
+      "step": 26960
+    },
+    {
+      "epoch": 3.7,
+      "learning_rate": 0.001230417412866786,
+      "loss": 0.3563,
+      "step": 26970
+    },
+    {
+      "epoch": 3.7,
+      "learning_rate": 0.0012298663727786196,
+      "loss": 0.336,
+      "step": 26980
+    },
+    {
+      "epoch": 3.7,
+      "learning_rate": 0.0012293153326904532,
+      "loss": 0.3536,
+      "step": 26990
+    },
+    {
+      "epoch": 3.7,
+      "learning_rate": 0.0012287642926022868,
+      "loss": 0.2382,
+      "step": 27000
+    },
+    {
+      "epoch": 3.7,
+      "eval_loss": 0.2566361725330353,
+      "eval_runtime": 1067.7438,
+      "eval_samples_per_second": 10.942,
+      "eval_steps_per_second": 1.368,
+      "eval_wer": 0.5525675429818976,
+      "step": 27000
+    },
+    {
+      "epoch": 3.7,
+      "learning_rate": 0.0012282132525141204,
+      "loss": 0.339,
+      "step": 27010
+    },
+    {
+      "epoch": 3.7,
+      "learning_rate": 0.001227662212425954,
+      "loss": 0.2919,
+      "step": 27020
+    },
+    {
+      "epoch": 3.7,
+      "learning_rate": 0.0012271111723377876,
+      "loss": 0.3293,
+      "step": 27030
+    },
+    {
+      "epoch": 3.7,
+      "learning_rate": 0.0012265601322496212,
+      "loss": 0.3202,
+      "step": 27040
+    },
+    {
+      "epoch": 3.71,
+      "learning_rate": 0.0012260090921614548,
+      "loss": 0.3298,
+      "step": 27050
+    },
+    {
+      "epoch": 3.71,
+      "learning_rate": 0.0012254580520732884,
+      "loss": 0.3139,
+      "step": 27060
+    },
+    {
+      "epoch": 3.71,
+      "learning_rate": 0.001224907011985122,
+      "loss": 0.3309,
+      "step": 27070
+    },
+    {
+      "epoch": 3.71,
+      "learning_rate": 0.0012243559718969554,
+      "loss": 0.3677,
+      "step": 27080
+    },
+    {
+      "epoch": 3.71,
+      "learning_rate": 0.0012238049318087892,
+      "loss": 0.3082,
+      "step": 27090
+    },
+    {
+      "epoch": 3.71,
+      "learning_rate": 0.0012232538917206226,
+      "loss": 0.2456,
+      "step": 27100
+    },
+    {
+      "epoch": 3.71,
+      "eval_loss": 0.2532599866390228,
+      "eval_runtime": 1069.9679,
+      "eval_samples_per_second": 10.919,
+      "eval_steps_per_second": 1.365,
+      "eval_wer": 0.5636086600563995,
+      "step": 27100
+    },
+    {
+      "epoch": 3.71,
+      "learning_rate": 0.0012227028516324564,
+      "loss": 0.3481,
+      "step": 27110
+    },
+    {
+      "epoch": 3.72,
+      "learning_rate": 0.0012221518115442898,
+      "loss": 0.3239,
+      "step": 27120
+    },
+    {
+      "epoch": 3.72,
+      "learning_rate": 0.0012216007714561236,
+      "loss": 0.3444,
+      "step": 27130
+    },
+    {
+      "epoch": 3.72,
+      "learning_rate": 0.001221049731367957,
+      "loss": 0.3264,
+      "step": 27140
+    },
+    {
+      "epoch": 3.72,
+      "learning_rate": 0.0012204986912797906,
+      "loss": 0.3217,
+      "step": 27150
+    },
+    {
+      "epoch": 3.72,
+      "learning_rate": 0.0012199476511916244,
+      "loss": 0.3234,
+      "step": 27160
+    },
+    {
+      "epoch": 3.72,
+      "learning_rate": 0.0012193966111034577,
+      "loss": 0.3534,
+      "step": 27170
+    },
+    {
+      "epoch": 3.72,
+      "learning_rate": 0.0012188455710152916,
+      "loss": 0.3025,
+      "step": 27180
+    },
+    {
+      "epoch": 3.73,
+      "learning_rate": 0.001218294530927125,
+      "loss": 0.3307,
+      "step": 27190
+    },
+    {
+      "epoch": 3.73,
+      "learning_rate": 0.0012177434908389585,
+      "loss": 0.2887,
+      "step": 27200
+    },
+    {
+      "epoch": 3.73,
+      "eval_loss": 0.2620231509208679,
+      "eval_runtime": 1067.4582,
+      "eval_samples_per_second": 10.945,
+      "eval_steps_per_second": 1.369,
+      "eval_wer": 0.574320021832075,
+      "step": 27200
+    },
+    {
+      "epoch": 3.73,
+      "learning_rate": 0.0012171924507507921,
+      "loss": 0.3144,
+      "step": 27210
+    },
+    {
+      "epoch": 3.73,
+      "learning_rate": 0.0012166414106626257,
+      "loss": 0.6724,
+      "step": 27220
+    },
+    {
+      "epoch": 3.73,
+      "learning_rate": 0.0012160903705744593,
+      "loss": 0.3389,
+      "step": 27230
+    },
+    {
+      "epoch": 3.73,
+      "learning_rate": 0.001215539330486293,
+      "loss": 0.3493,
+      "step": 27240
+    },
+    {
+      "epoch": 3.73,
+      "learning_rate": 0.0012149882903981265,
+      "loss": 0.2589,
+      "step": 27250
+    },
+    {
+      "epoch": 3.73,
+      "learning_rate": 0.0012144372503099601,
+      "loss": 0.3524,
+      "step": 27260
+    },
+    {
+      "epoch": 3.74,
+      "learning_rate": 0.0012138862102217935,
+      "loss": 0.3049,
+      "step": 27270
+    },
+    {
+      "epoch": 3.74,
+      "learning_rate": 0.0012133351701336273,
+      "loss": 0.4057,
+      "step": 27280
+    },
+    {
+      "epoch": 3.74,
+      "learning_rate": 0.0012127841300454607,
+      "loss": 0.3603,
+      "step": 27290
+    },
+    {
+      "epoch": 3.74,
+      "learning_rate": 0.0012122330899572945,
+      "loss": 0.2991,
+      "step": 27300
+    },
+    {
+      "epoch": 3.74,
+      "eval_loss": 0.2583196461200714,
+      "eval_runtime": 1073.0186,
+      "eval_samples_per_second": 10.888,
+      "eval_steps_per_second": 1.362,
+      "eval_wer": 0.5569566997180023,
+      "step": 27300
+    },
+    {
+      "epoch": 3.74,
+      "learning_rate": 0.001211682049869128,
+      "loss": 0.4051,
+      "step": 27310
+    },
+    {
+      "epoch": 3.74,
+      "learning_rate": 0.0012111310097809617,
+      "loss": 0.3809,
+      "step": 27320
+    },
+    {
+      "epoch": 3.74,
+      "learning_rate": 0.0012105799696927953,
+      "loss": 0.2851,
+      "step": 27330
+    },
+    {
+      "epoch": 3.75,
+      "learning_rate": 0.0012100289296046287,
+      "loss": 0.3258,
+      "step": 27340
+    },
+    {
+      "epoch": 3.75,
+      "learning_rate": 0.0012094778895164625,
+      "loss": 0.3581,
+      "step": 27350
+    },
+    {
+      "epoch": 3.75,
+      "learning_rate": 0.0012089268494282959,
+      "loss": 0.3338,
+      "step": 27360
+    },
+    {
+      "epoch": 3.75,
+      "learning_rate": 0.0012083758093401297,
+      "loss": 0.3379,
+      "step": 27370
+    },
+    {
+      "epoch": 3.75,
+      "learning_rate": 0.001207824769251963,
+      "loss": 0.3199,
+      "step": 27380
+    },
+    {
+      "epoch": 3.75,
+      "learning_rate": 0.0012072737291637966,
+      "loss": 0.335,
+      "step": 27390
+    },
+    {
+      "epoch": 3.75,
+      "learning_rate": 0.0012067226890756302,
+      "loss": 0.2838,
+      "step": 27400
+    },
+    {
+      "epoch": 3.75,
+      "eval_loss": 0.2538304030895233,
+      "eval_runtime": 1066.1018,
+      "eval_samples_per_second": 10.959,
+      "eval_steps_per_second": 1.37,
+      "eval_wer": 0.5534089875375239,
+      "step": 27400
+    },
+    {
+      "epoch": 3.76,
+      "learning_rate": 0.0012061716489874638,
+      "loss": 0.3819,
+      "step": 27410
+    },
+    {
+      "epoch": 3.76,
+      "learning_rate": 0.0012056206088992974,
+      "loss": 0.3417,
+      "step": 27420
+    },
+    {
+      "epoch": 3.76,
+      "learning_rate": 0.001205069568811131,
+      "loss": 0.3206,
+      "step": 27430
+    },
+    {
+      "epoch": 3.76,
+      "learning_rate": 0.0012045185287229646,
+      "loss": 0.3604,
+      "step": 27440
+    },
+    {
+      "epoch": 3.76,
+      "learning_rate": 0.0012039674886347982,
+      "loss": 0.3106,
+      "step": 27450
+    },
+    {
+      "epoch": 3.76,
+      "learning_rate": 0.0012034164485466318,
+      "loss": 0.3627,
+      "step": 27460
+    },
+    {
+      "epoch": 3.76,
+      "learning_rate": 0.0012028654084584654,
+      "loss": 0.3233,
+      "step": 27470
+    },
+    {
+      "epoch": 3.76,
+      "learning_rate": 0.001202314368370299,
+      "loss": 0.3271,
+      "step": 27480
+    },
+    {
+      "epoch": 3.77,
+      "learning_rate": 0.0012017633282821326,
+      "loss": 0.3398,
+      "step": 27490
+    },
+    {
+      "epoch": 3.77,
+      "learning_rate": 0.0012012122881939662,
+      "loss": 0.2766,
+      "step": 27500
+    },
+    {
+      "epoch": 3.77,
+      "eval_loss": 0.252188503742218,
+      "eval_runtime": 1072.102,
+      "eval_samples_per_second": 10.897,
+      "eval_steps_per_second": 1.363,
+      "eval_wer": 0.5561038842900027,
+      "step": 27500
+    },
+    {
+      "epoch": 3.77,
+      "learning_rate": 0.0012006612481057996,
+      "loss": 0.3093,
+      "step": 27510
+    },
+    {
+      "epoch": 3.77,
+      "learning_rate": 0.0012001102080176334,
+      "loss": 0.3045,
+      "step": 27520
+    },
+    {
+      "epoch": 3.77,
+      "learning_rate": 0.0011995591679294668,
+      "loss": 0.2827,
+      "step": 27530
+    },
+    {
+      "epoch": 3.77,
+      "learning_rate": 0.0011990081278413006,
+      "loss": 0.3241,
+      "step": 27540
+    },
+    {
+      "epoch": 3.77,
+      "learning_rate": 0.001198457087753134,
+      "loss": 0.2667,
+      "step": 27550
+    },
+    {
+      "epoch": 3.78,
+      "learning_rate": 0.0011979060476649678,
+      "loss": 0.3389,
+      "step": 27560
+    },
+    {
+      "epoch": 3.78,
+      "learning_rate": 0.0011973550075768012,
+      "loss": 0.3472,
+      "step": 27570
+    },
+    {
+      "epoch": 3.78,
+      "learning_rate": 0.0011968039674886348,
+      "loss": 0.343,
+      "step": 27580
+    },
+    {
+      "epoch": 3.78,
+      "learning_rate": 0.0011962529274004683,
+      "loss": 0.3084,
+      "step": 27590
+    },
+    {
+      "epoch": 3.78,
+      "learning_rate": 0.001195701887312302,
+      "loss": 0.2814,
+      "step": 27600
+    },
+    {
+      "epoch": 3.78,
+      "eval_loss": 0.2532261610031128,
+      "eval_runtime": 1069.6545,
+      "eval_samples_per_second": 10.922,
+      "eval_steps_per_second": 1.366,
+      "eval_wer": 0.5572978258892022,
+      "step": 27600
+    },
+    {
+      "epoch": 3.78,
+      "learning_rate": 0.0011951508472241355,
+      "loss": 0.3283,
+      "step": 27610
+    },
+    {
+      "epoch": 3.78,
+      "learning_rate": 0.0011945998071359691,
+      "loss": 0.3484,
+      "step": 27620
+    },
+    {
+      "epoch": 3.79,
+      "learning_rate": 0.001194048767047803,
+      "loss": 0.3305,
+      "step": 27630
+    },
+    {
+      "epoch": 3.79,
+      "learning_rate": 0.0011934977269596363,
+      "loss": 0.3333,
+      "step": 27640
+    },
+    {
+      "epoch": 3.79,
+      "learning_rate": 0.00119294668687147,
+      "loss": 0.3103,
+      "step": 27650
+    },
+    {
+      "epoch": 3.79,
+      "learning_rate": 0.0011923956467833035,
+      "loss": 0.3062,
+      "step": 27660
+    },
+    {
+      "epoch": 3.79,
+      "learning_rate": 0.0011918446066951371,
+      "loss": 0.3249,
+      "step": 27670
+    },
+    {
+      "epoch": 3.79,
+      "learning_rate": 0.0011912935666069707,
+      "loss": 0.3139,
+      "step": 27680
+    },
+    {
+      "epoch": 3.79,
+      "learning_rate": 0.0011907425265188043,
+      "loss": 0.3055,
+      "step": 27690
+    },
+    {
+      "epoch": 3.8,
+      "learning_rate": 0.0011901914864306377,
+      "loss": 0.2471,
+      "step": 27700
+    },
+    {
+      "epoch": 3.8,
+      "eval_loss": 0.24518433213233948,
+      "eval_runtime": 1067.1426,
+      "eval_samples_per_second": 10.948,
+      "eval_steps_per_second": 1.369,
+      "eval_wer": 0.5452219594287274,
+      "step": 27700
+    },
+    {
+      "epoch": 3.8,
+      "learning_rate": 0.0011896404463424715,
+      "loss": 0.3234,
+      "step": 27710
+    },
+    {
+      "epoch": 3.8,
+      "learning_rate": 0.0011890894062543049,
+      "loss": 0.305,
+      "step": 27720
+    },
+    {
+      "epoch": 3.8,
+      "learning_rate": 0.0011885383661661387,
+      "loss": 0.326,
+      "step": 27730
+    },
+    {
+      "epoch": 3.8,
+      "learning_rate": 0.001187987326077972,
+      "loss": 0.3429,
+      "step": 27740
+    },
+    {
+      "epoch": 3.8,
+      "learning_rate": 0.0011874362859898059,
+      "loss": 0.2691,
+      "step": 27750
+    },
+    {
+      "epoch": 3.8,
+      "learning_rate": 0.0011868852459016393,
+      "loss": 0.3143,
+      "step": 27760
+    },
+    {
+      "epoch": 3.8,
+      "learning_rate": 0.0011863342058134729,
+      "loss": 0.2925,
+      "step": 27770
+    },
+    {
+      "epoch": 3.81,
+      "learning_rate": 0.0011857831657253067,
+      "loss": 0.3215,
+      "step": 27780
+    },
+    {
+      "epoch": 3.81,
+      "learning_rate": 0.00118523212563714,
+      "loss": 0.3475,
+      "step": 27790
+    },
+    {
+      "epoch": 3.81,
+      "learning_rate": 0.0011846810855489739,
+      "loss": 0.2888,
+      "step": 27800
+    },
+    {
+      "epoch": 3.81,
+      "eval_loss": 0.24451225996017456,
+      "eval_runtime": 1069.2393,
+      "eval_samples_per_second": 10.926,
+      "eval_steps_per_second": 1.366,
+      "eval_wer": 0.5364891294460111,
+      "step": 27800
+    },
+    {
+      "epoch": 3.81,
+      "learning_rate": 0.0011841300454608072,
+      "loss": 0.3148,
+      "step": 27810
+    },
+    {
+      "epoch": 3.81,
+      "learning_rate": 0.001183579005372641,
+      "loss": 0.3244,
+      "step": 27820
+    },
+    {
+      "epoch": 3.81,
+      "learning_rate": 0.0011830279652844744,
+      "loss": 0.2991,
+      "step": 27830
+    },
+    {
+      "epoch": 3.81,
+      "learning_rate": 0.001182476925196308,
+      "loss": 0.3144,
+      "step": 27840
+    },
+    {
+      "epoch": 3.82,
+      "learning_rate": 0.0011819258851081416,
+      "loss": 0.2727,
+      "step": 27850
+    },
+    {
+      "epoch": 3.82,
+      "learning_rate": 0.0011813748450199752,
+      "loss": 0.3143,
+      "step": 27860
+    },
+    {
+      "epoch": 3.82,
+      "learning_rate": 0.0011808238049318088,
+      "loss": 0.3245,
+      "step": 27870
+    },
+    {
+      "epoch": 3.82,
+      "learning_rate": 0.0011802727648436424,
+      "loss": 0.3108,
+      "step": 27880
+    },
+    {
+      "epoch": 3.82,
+      "learning_rate": 0.0011797217247554758,
+      "loss": 0.3053,
+      "step": 27890
+    },
+    {
+      "epoch": 3.82,
+      "learning_rate": 0.0011791706846673096,
+      "loss": 0.2647,
+      "step": 27900
+    },
+    {
+      "epoch": 3.82,
+      "eval_loss": 0.23315469920635223,
+      "eval_runtime": 1082.9778,
+      "eval_samples_per_second": 10.788,
+      "eval_steps_per_second": 1.349,
+      "eval_wer": 0.5323842445192395,
+      "step": 27900
+    },
+    {
+      "epoch": 3.82,
+      "learning_rate": 0.001178619644579143,
+      "loss": 0.3081,
+      "step": 27910
+    },
+    {
+      "epoch": 3.83,
+      "learning_rate": 0.0011780686044909768,
+      "loss": 0.2713,
+      "step": 27920
+    },
+    {
+      "epoch": 3.83,
+      "learning_rate": 0.0011775175644028104,
+      "loss": 0.343,
+      "step": 27930
+    },
+    {
+      "epoch": 3.83,
+      "learning_rate": 0.001176966524314644,
+      "loss": 0.3248,
+      "step": 27940
+    },
+    {
+      "epoch": 3.83,
+      "learning_rate": 0.0011764154842264776,
+      "loss": 0.3068,
+      "step": 27950
+    },
+    {
+      "epoch": 3.83,
+      "learning_rate": 0.001175864444138311,
+      "loss": 0.3447,
+      "step": 27960
+    },
+    {
+      "epoch": 3.83,
+      "learning_rate": 0.0011753134040501448,
+      "loss": 0.319,
+      "step": 27970
+    },
+    {
+      "epoch": 3.83,
+      "learning_rate": 0.0011747623639619782,
+      "loss": 0.3253,
+      "step": 27980
+    },
+    {
+      "epoch": 3.83,
+      "learning_rate": 0.001174211323873812,
+      "loss": 0.3236,
+      "step": 27990
+    },
+    {
+      "epoch": 3.84,
+      "learning_rate": 0.0011736602837856454,
+      "loss": 0.2567,
+      "step": 28000
+    },
+    {
+      "epoch": 3.84,
+      "eval_loss": 0.2438521683216095,
+      "eval_runtime": 1069.4852,
+      "eval_samples_per_second": 10.924,
+      "eval_steps_per_second": 1.366,
+      "eval_wer": 0.5467683980715,
+      "step": 28000
+    },
+    {
+      "epoch": 3.84,
+      "learning_rate": 0.0011731092436974792,
+      "loss": 0.3105,
+      "step": 28010
+    },
+    {
+      "epoch": 3.84,
+      "learning_rate": 0.0011725582036093125,
+      "loss": 0.3208,
+      "step": 28020
+    },
+    {
+      "epoch": 3.84,
+      "learning_rate": 0.0011720071635211461,
+      "loss": 0.3073,
+      "step": 28030
+    },
+    {
+      "epoch": 3.84,
+      "learning_rate": 0.0011714561234329797,
+      "loss": 0.3307,
+      "step": 28040
+    },
+    {
+      "epoch": 3.84,
+      "learning_rate": 0.0011709050833448133,
+      "loss": 0.3112,
+      "step": 28050
+    },
+    {
+      "epoch": 3.84,
+      "learning_rate": 0.001170354043256647,
+      "loss": 0.3081,
+      "step": 28060
+    },
+    {
+      "epoch": 3.85,
+      "learning_rate": 0.0011698030031684805,
+      "loss": 0.3657,
+      "step": 28070
+    },
+    {
+      "epoch": 3.85,
+      "learning_rate": 0.001169251963080314,
+      "loss": 0.3278,
+      "step": 28080
+    },
+    {
+      "epoch": 3.85,
+      "learning_rate": 0.0011687009229921477,
+      "loss": 0.3352,
+      "step": 28090
+    },
+    {
+      "epoch": 3.85,
+      "learning_rate": 0.0011681498829039813,
+      "loss": 0.2848,
+      "step": 28100
+    },
+    {
+      "epoch": 3.85,
+      "eval_loss": 0.2438541203737259,
+      "eval_runtime": 1068.6205,
+      "eval_samples_per_second": 10.933,
+      "eval_steps_per_second": 1.367,
+      "eval_wer": 0.5375011370872373,
+      "step": 28100
+    },
+    {
+      "epoch": 3.85,
+      "learning_rate": 0.001167598842815815,
+      "loss": 0.3009,
+      "step": 28110
+    },
+    {
+      "epoch": 3.85,
+      "learning_rate": 0.0011670478027276485,
+      "loss": 0.302,
+      "step": 28120
+    },
+    {
+      "epoch": 3.85,
+      "learning_rate": 0.001166496762639482,
+      "loss": 0.317,
+      "step": 28130
+    },
+    {
+      "epoch": 3.86,
+      "learning_rate": 0.0011659457225513157,
+      "loss": 0.3168,
+      "step": 28140
+    },
+    {
+      "epoch": 3.86,
+      "learning_rate": 0.001165394682463149,
+      "loss": 0.2808,
+      "step": 28150
+    },
+    {
+      "epoch": 3.86,
+      "learning_rate": 0.001164843642374983,
+      "loss": 0.3826,
+      "step": 28160
+    },
+    {
+      "epoch": 3.86,
+      "learning_rate": 0.0011642926022868163,
+      "loss": 0.3249,
+      "step": 28170
+    },
+    {
+      "epoch": 3.86,
+      "learning_rate": 0.00116374156219865,
+      "loss": 0.3486,
+      "step": 28180
+    },
+    {
+      "epoch": 3.86,
+      "learning_rate": 0.0011631905221104835,
+      "loss": 0.3224,
+      "step": 28190
+    },
+    {
+      "epoch": 3.86,
+      "learning_rate": 0.0011626394820223173,
+      "loss": 0.2445,
+      "step": 28200
+    },
+    {
+      "epoch": 3.86,
+      "eval_loss": 0.2523077726364136,
+      "eval_runtime": 1067.8475,
+      "eval_samples_per_second": 10.941,
+      "eval_steps_per_second": 1.368,
+      "eval_wer": 0.5485422541617393,
+      "step": 28200
+    },
+    {
+      "epoch": 3.86,
+      "learning_rate": 0.0011620884419341507,
+      "loss": 0.3488,
+      "step": 28210
+    },
+    {
+      "epoch": 3.87,
+      "learning_rate": 0.0011615374018459843,
+      "loss": 0.3418,
+      "step": 28220
+    },
+    {
+      "epoch": 3.87,
+      "learning_rate": 0.0011609863617578178,
+      "loss": 0.3509,
+      "step": 28230
+    },
+    {
+      "epoch": 3.87,
+      "learning_rate": 0.0011604353216696514,
+      "loss": 0.3226,
+      "step": 28240
+    },
+    {
+      "epoch": 3.87,
+      "learning_rate": 0.0011598842815814853,
+      "loss": 0.3014,
+      "step": 28250
+    },
+    {
+      "epoch": 3.87,
+      "learning_rate": 0.0011593332414933186,
+      "loss": 0.3587,
+      "step": 28260
+    },
+    {
+      "epoch": 3.87,
+      "learning_rate": 0.0011587822014051522,
+      "loss": 0.2953,
+      "step": 28270
+    },
+    {
+      "epoch": 3.87,
+      "learning_rate": 0.0011582311613169858,
+      "loss": 0.3083,
+      "step": 28280
+    },
+    {
+      "epoch": 3.88,
+      "learning_rate": 0.0011576801212288194,
+      "loss": 0.3611,
+      "step": 28290
+    },
+    {
+      "epoch": 3.88,
+      "learning_rate": 0.001157129081140653,
+      "loss": 0.2836,
+      "step": 28300
+    },
+    {
+      "epoch": 3.88,
+      "eval_loss": 0.24756070971488953,
+      "eval_runtime": 1068.7059,
+      "eval_samples_per_second": 10.932,
+      "eval_steps_per_second": 1.367,
+      "eval_wer": 0.5408441735649959,
+      "step": 28300
+    },
+    {
+      "epoch": 3.88,
+      "learning_rate": 0.0011565780410524866,
+      "loss": 0.3184,
+      "step": 28310
+    },
+    {
+      "epoch": 3.88,
+      "learning_rate": 0.0011560270009643202,
+      "loss": 0.3036,
+      "step": 28320
+    },
+    {
+      "epoch": 3.88,
+      "learning_rate": 0.0011554759608761538,
+      "loss": 0.2999,
+      "step": 28330
+    },
+    {
+      "epoch": 3.88,
+      "learning_rate": 0.0011549249207879872,
+      "loss": 0.3528,
+      "step": 28340
+    },
+    {
+      "epoch": 3.88,
+      "learning_rate": 0.001154373880699821,
+      "loss": 0.2687,
+      "step": 28350
+    },
+    {
+      "epoch": 3.89,
+      "learning_rate": 0.0011538228406116544,
+      "loss": 0.3107,
+      "step": 28360
+    },
+    {
+      "epoch": 3.89,
+      "learning_rate": 0.0011532718005234882,
+      "loss": 0.2859,
+      "step": 28370
+    },
+    {
+      "epoch": 3.89,
+      "learning_rate": 0.0011527207604353216,
+      "loss": 0.3088,
+      "step": 28380
+    },
+    {
+      "epoch": 3.89,
+      "learning_rate": 0.0011521697203471554,
+      "loss": 0.3308,
+      "step": 28390
+    },
+    {
+      "epoch": 3.89,
+      "learning_rate": 0.001151618680258989,
+      "loss": 0.2441,
+      "step": 28400
+    },
+    {
+      "epoch": 3.89,
+      "eval_loss": 0.25136733055114746,
+      "eval_runtime": 1068.4242,
+      "eval_samples_per_second": 10.935,
+      "eval_steps_per_second": 1.367,
+      "eval_wer": 0.5511007004457382,
+      "step": 28400
+    },
+    {
+      "epoch": 3.89,
+      "learning_rate": 0.0011510676401708224,
+      "loss": 0.3444,
+      "step": 28410
+    },
+    {
+      "epoch": 3.89,
+      "learning_rate": 0.0011505166000826562,
+      "loss": 0.3037,
+      "step": 28420
+    },
+    {
+      "epoch": 3.9,
+      "learning_rate": 0.0011499655599944896,
+      "loss": 0.2953,
+      "step": 28430
+    },
+    {
+      "epoch": 3.9,
+      "learning_rate": 0.0011494145199063234,
+      "loss": 0.3166,
+      "step": 28440
+    },
+    {
+      "epoch": 3.9,
+      "learning_rate": 0.0011488634798181567,
+      "loss": 0.2876,
+      "step": 28450
+    },
+    {
+      "epoch": 3.9,
+      "learning_rate": 0.0011483124397299903,
+      "loss": 0.3421,
+      "step": 28460
+    },
+    {
+      "epoch": 3.9,
+      "learning_rate": 0.001147761399641824,
+      "loss": 0.3143,
+      "step": 28470
+    },
+    {
+      "epoch": 3.9,
+      "learning_rate": 0.0011472103595536575,
+      "loss": 0.2886,
+      "step": 28480
+    },
+    {
+      "epoch": 3.9,
+      "learning_rate": 0.0011466593194654911,
+      "loss": 0.3132,
+      "step": 28490
+    },
+    {
+      "epoch": 3.9,
+      "learning_rate": 0.0011461082793773247,
+      "loss": 0.272,
+      "step": 28500
+    },
+    {
+      "epoch": 3.9,
+      "eval_loss": 0.23404917120933533,
+      "eval_runtime": 1068.9432,
+      "eval_samples_per_second": 10.929,
+      "eval_steps_per_second": 1.367,
+      "eval_wer": 0.53125852815428,
+      "step": 28500
+    },
+    {
+      "epoch": 3.91,
+      "learning_rate": 0.0011455572392891583,
+      "loss": 0.3248,
+      "step": 28510
+    },
+    {
+      "epoch": 3.91,
+      "learning_rate": 0.001145006199200992,
+      "loss": 0.4786,
+      "step": 28520
+    },
+    {
+      "epoch": 3.91,
+      "learning_rate": 0.0011444551591128253,
+      "loss": 0.3434,
+      "step": 28530
+    },
+    {
+      "epoch": 3.91,
+      "learning_rate": 0.0011439041190246591,
+      "loss": 0.3396,
+      "step": 28540
+    },
+    {
+      "epoch": 3.91,
+      "learning_rate": 0.0011433530789364927,
+      "loss": 0.2799,
+      "step": 28550
+    },
+    {
+      "epoch": 3.91,
+      "learning_rate": 0.0011428020388483263,
+      "loss": 0.3457,
+      "step": 28560
+    },
+    {
+      "epoch": 3.91,
+      "learning_rate": 0.00114225099876016,
+      "loss": 0.2819,
+      "step": 28570
+    },
+    {
+      "epoch": 3.92,
+      "learning_rate": 0.0011416999586719935,
+      "loss": 0.3529,
+      "step": 28580
+    },
+    {
+      "epoch": 3.92,
+      "learning_rate": 0.001141148918583827,
+      "loss": 0.3095,
+      "step": 28590
+    },
+    {
+      "epoch": 3.92,
+      "learning_rate": 0.0011405978784956605,
+      "loss": 0.2746,
+      "step": 28600
+    },
+    {
+      "epoch": 3.92,
+      "eval_loss": 0.24538134038448334,
+      "eval_runtime": 1071.0226,
+      "eval_samples_per_second": 10.908,
+      "eval_steps_per_second": 1.364,
+      "eval_wer": 0.5393887019012099,
+      "step": 28600
+    },
+    {
+      "epoch": 3.92,
+      "learning_rate": 0.0011400468384074943,
+      "loss": 0.3045,
+      "step": 28610
+    },
+    {
+      "epoch": 3.92,
+      "learning_rate": 0.0011394957983193277,
+      "loss": 0.29,
+      "step": 28620
+    },
+    {
+      "epoch": 3.92,
+      "learning_rate": 0.0011389447582311615,
+      "loss": 0.3284,
+      "step": 28630
+    },
+    {
+      "epoch": 3.92,
+      "learning_rate": 0.0011383937181429949,
+      "loss": 0.2969,
+      "step": 28640
+    },
+    {
+      "epoch": 3.93,
+      "learning_rate": 0.0011378426780548285,
+      "loss": 0.2074,
+      "step": 28650
+    },
+    {
+      "epoch": 3.93,
+      "learning_rate": 0.001137291637966662,
+      "loss": 0.3065,
+      "step": 28660
+    },
+    {
+      "epoch": 3.93,
+      "learning_rate": 0.0011367405978784956,
+      "loss": 0.3035,
+      "step": 28670
+    },
+    {
+      "epoch": 3.93,
+      "learning_rate": 0.0011361895577903292,
+      "loss": 0.3024,
+      "step": 28680
+    },
+    {
+      "epoch": 3.93,
+      "learning_rate": 0.0011356385177021628,
+      "loss": 0.304,
+      "step": 28690
+    },
+    {
+      "epoch": 3.93,
+      "learning_rate": 0.0011350874776139964,
+      "loss": 0.2637,
+      "step": 28700
+    },
+    {
+      "epoch": 3.93,
+      "eval_loss": 0.2483680248260498,
+      "eval_runtime": 1070.3934,
+      "eval_samples_per_second": 10.915,
+      "eval_steps_per_second": 1.365,
+      "eval_wer": 0.5437323751478214,
+      "step": 28700
+    },
+    {
+      "epoch": 3.93,
+      "learning_rate": 0.00113453643752583,
+      "loss": 0.3471,
+      "step": 28710
+    },
+    {
+      "epoch": 3.93,
+      "learning_rate": 0.0011339853974376636,
+      "loss": 0.3005,
+      "step": 28720
+    },
+    {
+      "epoch": 3.94,
+      "learning_rate": 0.0011334343573494972,
+      "loss": 0.2926,
+      "step": 28730
+    },
+    {
+      "epoch": 3.94,
+      "learning_rate": 0.0011328833172613308,
+      "loss": 0.2962,
+      "step": 28740
+    },
+    {
+      "epoch": 3.94,
+      "learning_rate": 0.0011323322771731644,
+      "loss": 0.2729,
+      "step": 28750
+    },
+    {
+      "epoch": 3.94,
+      "learning_rate": 0.001131781237084998,
+      "loss": 0.3089,
+      "step": 28760
+    },
+    {
+      "epoch": 3.94,
+      "learning_rate": 0.001131285301005648,
+      "loss": 0.675,
+      "step": 28770
+    },
+    {
+      "epoch": 3.94,
+      "learning_rate": 0.0011307342609174818,
+      "loss": 0.2865,
+      "step": 28780
+    },
+    {
+      "epoch": 3.94,
+      "learning_rate": 0.0011301832208293154,
+      "loss": 0.3381,
+      "step": 28790
+    },
+    {
+      "epoch": 3.95,
+      "learning_rate": 0.001129632180741149,
+      "loss": 0.2889,
+      "step": 28800
+    },
+    {
+      "epoch": 3.95,
+      "eval_loss": 0.2344142496585846,
+      "eval_runtime": 1071.1925,
+      "eval_samples_per_second": 10.907,
+      "eval_steps_per_second": 1.364,
+      "eval_wer": 0.5312016737924133,
+      "step": 28800
+    },
+    {
+      "epoch": 3.95,
+      "learning_rate": 0.0011290811406529826,
+      "loss": 0.3536,
+      "step": 28810
+    },
+    {
+      "epoch": 3.95,
+      "learning_rate": 0.001128530100564816,
+      "loss": 0.3155,
+      "step": 28820
+    },
+    {
+      "epoch": 3.95,
+      "learning_rate": 0.0011279790604766498,
+      "loss": 0.3223,
+      "step": 28830
+    },
+    {
+      "epoch": 3.95,
+      "learning_rate": 0.0011274280203884832,
+      "loss": 0.3146,
+      "step": 28840
+    },
+    {
+      "epoch": 3.95,
+      "learning_rate": 0.001126876980300317,
+      "loss": 0.3045,
+      "step": 28850
+    },
+    {
+      "epoch": 3.95,
+      "learning_rate": 0.0011263259402121504,
+      "loss": 0.2882,
+      "step": 28860
+    },
+    {
+      "epoch": 3.96,
+      "learning_rate": 0.001125774900123984,
+      "loss": 0.3351,
+      "step": 28870
+    },
+    {
+      "epoch": 3.96,
+      "learning_rate": 0.0011252238600358176,
+      "loss": 0.3198,
+      "step": 28880
+    },
+    {
+      "epoch": 3.96,
+      "learning_rate": 0.0011246728199476512,
+      "loss": 0.3304,
+      "step": 28890
+    },
+    {
+      "epoch": 3.96,
+      "learning_rate": 0.0011241217798594848,
+      "loss": 0.2733,
+      "step": 28900
+    },
+    {
+      "epoch": 3.96,
+      "eval_loss": 0.22612608969211578,
+      "eval_runtime": 1071.7429,
+      "eval_samples_per_second": 10.901,
+      "eval_steps_per_second": 1.363,
+      "eval_wer": 0.5313836077503866,
+      "step": 28900
+    },
+    {
+      "epoch": 3.96,
+      "learning_rate": 0.0011235707397713184,
+      "loss": 0.3116,
+      "step": 28910
+    },
+    {
+      "epoch": 3.96,
+      "learning_rate": 0.001123019699683152,
+      "loss": 0.3035,
+      "step": 28920
+    },
+    {
+      "epoch": 3.96,
+      "learning_rate": 0.0011224686595949856,
+      "loss": 0.3106,
+      "step": 28930
+    },
+    {
+      "epoch": 3.96,
+      "learning_rate": 0.0011219176195068192,
+      "loss": 0.3258,
+      "step": 28940
+    },
+    {
+      "epoch": 3.97,
+      "learning_rate": 0.0011213665794186528,
+      "loss": 0.2736,
+      "step": 28950
+    },
+    {
+      "epoch": 3.97,
+      "learning_rate": 0.0011208155393304864,
+      "loss": 0.4772,
+      "step": 28960
+    },
+    {
+      "epoch": 3.97,
+      "learning_rate": 0.00112026449924232,
+      "loss": 0.3224,
+      "step": 28970
+    },
+    {
+      "epoch": 3.97,
+      "learning_rate": 0.0011197134591541535,
+      "loss": 0.3326,
+      "step": 28980
+    },
+    {
+      "epoch": 3.97,
+      "learning_rate": 0.0011191624190659871,
+      "loss": 0.3336,
+      "step": 28990
+    },
+    {
+      "epoch": 3.97,
+      "learning_rate": 0.0011186113789778207,
+      "loss": 0.3967,
+      "step": 29000
+    },
+    {
+      "epoch": 3.97,
+      "eval_loss": 0.2547242045402527,
+      "eval_runtime": 1074.0049,
+      "eval_samples_per_second": 10.878,
+      "eval_steps_per_second": 1.36,
+      "eval_wer": 0.5666560538524515,
+      "step": 29000
+    },
+    {
+      "epoch": 3.97,
+      "learning_rate": 0.0011180603388896541,
+      "loss": 0.3259,
+      "step": 29010
+    },
+    {
+      "epoch": 3.98,
+      "learning_rate": 0.001117509298801488,
+      "loss": 0.3088,
+      "step": 29020
+    },
+    {
+      "epoch": 3.98,
+      "learning_rate": 0.0011169582587133213,
+      "loss": 0.3543,
+      "step": 29030
+    },
+    {
+      "epoch": 3.98,
+      "learning_rate": 0.0011164072186251551,
+      "loss": 0.2982,
+      "step": 29040
+    },
+    {
+      "epoch": 3.98,
+      "learning_rate": 0.0011158561785369885,
+      "loss": 0.3279,
+      "step": 29050
+    },
+    {
+      "epoch": 3.98,
+      "learning_rate": 0.001115305138448822,
+      "loss": 0.3661,
+      "step": 29060
+    },
+    {
+      "epoch": 3.98,
+      "learning_rate": 0.0011147540983606557,
+      "loss": 0.3191,
+      "step": 29070
+    },
+    {
+      "epoch": 3.98,
+      "learning_rate": 0.0011142030582724893,
+      "loss": 0.3071,
+      "step": 29080
+    },
+    {
+      "epoch": 3.99,
+      "learning_rate": 0.0011136520181843229,
+      "loss": 0.283,
+      "step": 29090
+    },
+    {
+      "epoch": 3.99,
+      "learning_rate": 0.0011131009780961565,
+      "loss": 0.3052,
+      "step": 29100
+    },
+    {
+      "epoch": 3.99,
+      "eval_loss": 0.2226630598306656,
+      "eval_runtime": 1070.7558,
+      "eval_samples_per_second": 10.911,
+      "eval_steps_per_second": 1.364,
+      "eval_wer": 0.5270171927590285,
+      "step": 29100
+    },
+    {
+      "epoch": 3.99,
+      "learning_rate": 0.0011125499380079903,
+      "loss": 0.3172,
+      "step": 29110
+    },
+    {
+      "epoch": 3.99,
+      "learning_rate": 0.0011119988979198237,
+      "loss": 0.2935,
+      "step": 29120
+    },
+    {
+      "epoch": 3.99,
+      "learning_rate": 0.0011114478578316573,
+      "loss": 0.3062,
+      "step": 29130
+    },
+    {
+      "epoch": 3.99,
+      "learning_rate": 0.0011108968177434909,
+      "loss": 0.3183,
+      "step": 29140
+    },
+    {
+      "epoch": 3.99,
+      "learning_rate": 0.0011103457776553245,
+      "loss": 0.2189,
+      "step": 29150
+    },
+    {
+      "epoch": 4.0,
+      "learning_rate": 0.001109794737567158,
+      "loss": 0.3421,
+      "step": 29160
+    },
+    {
+      "epoch": 4.0,
+      "learning_rate": 0.0011092436974789917,
+      "loss": 0.32,
+      "step": 29170
+    },
+    {
+      "epoch": 4.0,
+      "learning_rate": 0.0011086926573908253,
+      "loss": 0.3366,
+      "step": 29180
+    },
+    {
+      "epoch": 4.0,
+      "learning_rate": 0.0011081416173026588,
+      "loss": 0.3298,
+      "step": 29190
+    },
+    {
+      "epoch": 4.0,
+      "learning_rate": 0.0011075905772144922,
+      "loss": 0.2849,
+      "step": 29200
+    },
+    {
+      "epoch": 4.0,
+      "eval_loss": 0.22722077369689941,
+      "eval_runtime": 1073.3888,
+      "eval_samples_per_second": 10.884,
+      "eval_steps_per_second": 1.361,
+      "eval_wer": 0.527176384972255,
+      "step": 29200
+    },
+    {
+      "epoch": 4.0,
+      "learning_rate": 0.001107039537126326,
+      "loss": 0.2872,
+      "step": 29210
+    },
+    {
+      "epoch": 4.0,
+      "learning_rate": 0.0011064884970381594,
+      "loss": 0.2918,
+      "step": 29220
+    },
+    {
+      "epoch": 4.0,
+      "learning_rate": 0.0011059374569499932,
+      "loss": 0.2771,
+      "step": 29230
+    },
+    {
+      "epoch": 4.01,
+      "learning_rate": 0.0011053864168618266,
+      "loss": 0.2785,
+      "step": 29240
+    },
+    {
+      "epoch": 4.01,
+      "learning_rate": 0.0011048353767736602,
+      "loss": 0.2167,
+      "step": 29250
+    },
+    {
+      "epoch": 4.01,
+      "learning_rate": 0.001104284336685494,
+      "loss": 0.2893,
+      "step": 29260
+    },
+    {
+      "epoch": 4.01,
+      "learning_rate": 0.0011037332965973274,
+      "loss": 0.2919,
+      "step": 29270
+    },
+    {
+      "epoch": 4.01,
+      "learning_rate": 0.0011031822565091612,
+      "loss": 0.296,
+      "step": 29280
+    },
+    {
+      "epoch": 4.01,
+      "learning_rate": 0.0011026312164209946,
+      "loss": 0.2736,
+      "step": 29290
+    },
+    {
+      "epoch": 4.01,
+      "learning_rate": 0.0011020801763328284,
+      "loss": 0.2446,
+      "step": 29300
+    },
+    {
+      "epoch": 4.01,
+      "eval_loss": 0.2285657823085785,
+      "eval_runtime": 1072.1425,
+      "eval_samples_per_second": 10.897,
+      "eval_steps_per_second": 1.363,
+      "eval_wer": 0.5139065769125807,
+      "step": 29300
+    },
+    {
+      "epoch": 4.02,
+      "learning_rate": 0.0011015291362446618,
+      "loss": 0.2961,
+      "step": 29310
+    },
+    {
+      "epoch": 4.02,
+      "learning_rate": 0.0011009780961564954,
+      "loss": 0.273,
+      "step": 29320
+    },
+    {
+      "epoch": 4.02,
+      "learning_rate": 0.001100427056068329,
+      "loss": 0.2699,
+      "step": 29330
+    },
+    {
+      "epoch": 4.02,
+      "learning_rate": 0.0010998760159801626,
+      "loss": 0.2957,
+      "step": 29340
+    },
+    {
+      "epoch": 4.02,
+      "learning_rate": 0.0010993249758919962,
+      "loss": 0.2886,
+      "step": 29350
+    },
+    {
+      "epoch": 4.02,
+      "learning_rate": 0.0010987739358038298,
+      "loss": 0.2772,
+      "step": 29360
+    },
+    {
+      "epoch": 4.02,
+      "learning_rate": 0.0010982228957156631,
+      "loss": 0.262,
+      "step": 29370
+    },
+    {
+      "epoch": 4.03,
+      "learning_rate": 0.001097671855627497,
+      "loss": 0.2831,
+      "step": 29380
+    },
+    {
+      "epoch": 4.03,
+      "learning_rate": 0.0010971208155393303,
+      "loss": 0.306,
+      "step": 29390
+    },
+    {
+      "epoch": 4.03,
+      "learning_rate": 0.0010965697754511641,
+      "loss": 0.2553,
+      "step": 29400
+    },
+    {
+      "epoch": 4.03,
+      "eval_loss": 0.22941261529922485,
+      "eval_runtime": 1070.2316,
+      "eval_samples_per_second": 10.916,
+      "eval_steps_per_second": 1.365,
+      "eval_wer": 0.5188983898844719,
+      "step": 29400
+    },
+    {
+      "epoch": 4.03,
+      "learning_rate": 0.0010960187353629977,
+      "loss": 0.2984,
+      "step": 29410
+    },
+    {
+      "epoch": 4.03,
+      "learning_rate": 0.0010954676952748313,
+      "loss": 0.281,
+      "step": 29420
+    },
+    {
+      "epoch": 4.03,
+      "learning_rate": 0.001094916655186665,
+      "loss": 0.2998,
+      "step": 29430
+    },
+    {
+      "epoch": 4.03,
+      "learning_rate": 0.0010943656150984983,
+      "loss": 0.3084,
+      "step": 29440
+    },
+    {
+      "epoch": 4.03,
+      "learning_rate": 0.0010938145750103321,
+      "loss": 0.2145,
+      "step": 29450
+    },
+    {
+      "epoch": 4.04,
+      "learning_rate": 0.0010932635349221655,
+      "loss": 0.2962,
+      "step": 29460
+    },
+    {
+      "epoch": 4.04,
+      "learning_rate": 0.0010927124948339993,
+      "loss": 0.2765,
+      "step": 29470
+    },
+    {
+      "epoch": 4.04,
+      "learning_rate": 0.0010921614547458327,
+      "loss": 0.3469,
+      "step": 29480
+    },
+    {
+      "epoch": 4.04,
+      "learning_rate": 0.0010916104146576665,
+      "loss": 0.3012,
+      "step": 29490
+    },
+    {
+      "epoch": 4.04,
+      "learning_rate": 0.0010910593745695,
+      "loss": 0.2948,
+      "step": 29500
+    },
+    {
+      "epoch": 4.04,
+      "eval_loss": 0.24770690500736237,
+      "eval_runtime": 1074.9321,
+      "eval_samples_per_second": 10.869,
+      "eval_steps_per_second": 1.359,
+      "eval_wer": 0.5320999727099063,
+      "step": 29500
+    },
+    {
+      "epoch": 4.04,
+      "learning_rate": 0.0010905083344813335,
+      "loss": 0.3397,
+      "step": 29510
+    },
+    {
+      "epoch": 4.04,
+      "learning_rate": 0.001089957294393167,
+      "loss": 0.289,
+      "step": 29520
+    },
+    {
+      "epoch": 4.05,
+      "learning_rate": 0.0010894062543050007,
+      "loss": 0.2754,
+      "step": 29530
+    },
+    {
+      "epoch": 4.05,
+      "learning_rate": 0.0010888552142168343,
+      "loss": 0.2716,
+      "step": 29540
+    },
+    {
+      "epoch": 4.05,
+      "learning_rate": 0.0010883041741286679,
+      "loss": 0.2384,
+      "step": 29550
+    },
+    {
+      "epoch": 4.05,
+      "learning_rate": 0.0010877531340405017,
+      "loss": 0.2854,
+      "step": 29560
+    },
+    {
+      "epoch": 4.05,
+      "learning_rate": 0.001087202093952335,
+      "loss": 0.2661,
+      "step": 29570
+    },
+    {
+      "epoch": 4.05,
+      "learning_rate": 0.0010866510538641687,
+      "loss": 0.2801,
+      "step": 29580
+    },
+    {
+      "epoch": 4.05,
+      "learning_rate": 0.0010861000137760023,
+      "loss": 0.3028,
+      "step": 29590
+    },
+    {
+      "epoch": 4.06,
+      "learning_rate": 0.0010855489736878359,
+      "loss": 0.2541,
+      "step": 29600
+    },
+    {
+      "epoch": 4.06,
+      "eval_loss": 0.23023104667663574,
+      "eval_runtime": 1070.2055,
+      "eval_samples_per_second": 10.917,
+      "eval_steps_per_second": 1.365,
+      "eval_wer": 0.5202628945692713,
+      "step": 29600
+    },
+    {
+      "epoch": 4.06,
+      "learning_rate": 0.0010849979335996694,
+      "loss": 0.3204,
+      "step": 29610
+    },
+    {
+      "epoch": 4.06,
+      "learning_rate": 0.001084446893511503,
+      "loss": 0.359,
+      "step": 29620
+    },
+    {
+      "epoch": 4.06,
+      "learning_rate": 0.0010838958534233364,
+      "loss": 0.2887,
+      "step": 29630
+    },
+    {
+      "epoch": 4.06,
+      "learning_rate": 0.0010833448133351702,
+      "loss": 0.3224,
+      "step": 29640
+    },
+    {
+      "epoch": 4.06,
+      "learning_rate": 0.0010827937732470036,
+      "loss": 0.217,
+      "step": 29650
+    },
+    {
+      "epoch": 4.06,
+      "learning_rate": 0.0010822427331588374,
+      "loss": 0.3291,
+      "step": 29660
+    },
+    {
+      "epoch": 4.06,
+      "learning_rate": 0.0010816916930706708,
+      "loss": 0.3016,
+      "step": 29670
+    },
+    {
+      "epoch": 4.07,
+      "learning_rate": 0.0010811406529825046,
+      "loss": 0.2989,
+      "step": 29680
+    },
+    {
+      "epoch": 4.07,
+      "learning_rate": 0.001080589612894338,
+      "loss": 0.288,
+      "step": 29690
+    },
+    {
+      "epoch": 4.07,
+      "learning_rate": 0.0010800385728061716,
+      "loss": 0.2361,
+      "step": 29700
+    },
+    {
+      "epoch": 4.07,
+      "eval_loss": 0.2227802574634552,
+      "eval_runtime": 1071.0563,
+      "eval_samples_per_second": 10.908,
+      "eval_steps_per_second": 1.364,
+      "eval_wer": 0.5249135813699627,
+      "step": 29700
+    },
+    {
+      "epoch": 4.07,
+      "learning_rate": 0.0010794875327180052,
+      "loss": 0.2916,
+      "step": 29710
+    },
+    {
+      "epoch": 4.07,
+      "learning_rate": 0.0010789364926298388,
+      "loss": 0.2732,
+      "step": 29720
+    },
+    {
+      "epoch": 4.07,
+      "learning_rate": 0.0010783854525416726,
+      "loss": 0.2611,
+      "step": 29730
+    },
+    {
+      "epoch": 4.07,
+      "learning_rate": 0.001077834412453506,
+      "loss": 0.2754,
+      "step": 29740
+    },
+    {
+      "epoch": 4.08,
+      "learning_rate": 0.0010772833723653398,
+      "loss": 0.31,
+      "step": 29750
+    },
+    {
+      "epoch": 4.08,
+      "learning_rate": 0.0010767323322771732,
+      "loss": 0.3034,
+      "step": 29760
+    },
+    {
+      "epoch": 4.08,
+      "learning_rate": 0.0010761812921890068,
+      "loss": 0.2755,
+      "step": 29770
+    },
+    {
+      "epoch": 4.08,
+      "learning_rate": 0.0010756302521008404,
+      "loss": 0.2709,
+      "step": 29780
+    },
+    {
+      "epoch": 4.08,
+      "learning_rate": 0.001075079212012674,
+      "loss": 0.2498,
+      "step": 29790
+    },
+    {
+      "epoch": 4.08,
+      "learning_rate": 0.0010745281719245076,
+      "loss": 0.3058,
+      "step": 29800
+    },
+    {
+      "epoch": 4.08,
+      "eval_loss": 0.22715097665786743,
+      "eval_runtime": 1069.6627,
+      "eval_samples_per_second": 10.922,
+      "eval_steps_per_second": 1.366,
+      "eval_wer": 0.5197739470572182,
+      "step": 29800
+    },
+    {
+      "epoch": 4.08,
+      "learning_rate": 0.0010739771318363412,
+      "loss": 0.2843,
+      "step": 29810
+    },
+    {
+      "epoch": 4.09,
+      "learning_rate": 0.0010734260917481745,
+      "loss": 0.2602,
+      "step": 29820
+    },
+    {
+      "epoch": 4.09,
+      "learning_rate": 0.0010728750516600083,
+      "loss": 0.2751,
+      "step": 29830
+    },
+    {
+      "epoch": 4.09,
+      "learning_rate": 0.0010723240115718417,
+      "loss": 0.2916,
+      "step": 29840
+    },
+    {
+      "epoch": 4.09,
+      "learning_rate": 0.0010717729714836755,
+      "loss": 0.2294,
+      "step": 29850
+    },
+    {
+      "epoch": 4.09,
+      "learning_rate": 0.001071221931395509,
+      "loss": 0.2999,
+      "step": 29860
+    },
+    {
+      "epoch": 4.09,
+      "learning_rate": 0.0010706708913073427,
+      "loss": 0.292,
+      "step": 29870
+    },
+    {
+      "epoch": 4.09,
+      "learning_rate": 0.0010701198512191763,
+      "loss": 0.2779,
+      "step": 29880
+    },
+    {
+      "epoch": 4.1,
+      "learning_rate": 0.0010695688111310097,
+      "loss": 0.2968,
+      "step": 29890
+    },
+    {
+      "epoch": 4.1,
+      "learning_rate": 0.0010690177710428435,
+      "loss": 0.2584,
+      "step": 29900
+    },
+    {
+      "epoch": 4.1,
+      "eval_loss": 0.22250257432460785,
+      "eval_runtime": 1070.6635,
+      "eval_samples_per_second": 10.912,
+      "eval_steps_per_second": 1.365,
+      "eval_wer": 0.5136336759756208,
+      "step": 29900
+    },
+    {
+      "epoch": 4.1,
+      "learning_rate": 0.001068466730954677,
+      "loss": 0.3114,
+      "step": 29910
+    },
+    {
+      "epoch": 4.1,
+      "learning_rate": 0.0010679156908665107,
+      "loss": 0.2968,
+      "step": 29920
+    },
+    {
+      "epoch": 4.1,
+      "learning_rate": 0.001067364650778344,
+      "loss": 0.2848,
+      "step": 29930
+    },
+    {
+      "epoch": 4.1,
+      "learning_rate": 0.0010668136106901777,
+      "loss": 0.3141,
+      "step": 29940
+    },
+    {
+      "epoch": 4.1,
+      "learning_rate": 0.0010662625706020113,
+      "loss": 0.2172,
+      "step": 29950
+    },
+    {
+      "epoch": 4.1,
+      "learning_rate": 0.0010657115305138449,
+      "loss": 0.2817,
+      "step": 29960
+    },
+    {
+      "epoch": 4.11,
+      "learning_rate": 0.0010651604904256785,
+      "loss": 0.3085,
+      "step": 29970
+    },
+    {
+      "epoch": 4.11,
+      "learning_rate": 0.001064609450337512,
+      "loss": 0.2936,
+      "step": 29980
+    },
+    {
+      "epoch": 4.11,
+      "learning_rate": 0.0010640584102493457,
+      "loss": 0.2891,
+      "step": 29990
+    },
+    {
+      "epoch": 4.11,
+      "learning_rate": 0.0010635073701611793,
+      "loss": 0.2686,
+      "step": 30000
+    },
+    {
+      "epoch": 4.11,
+      "eval_loss": 0.2352893352508545,
+      "eval_runtime": 1070.3822,
+      "eval_samples_per_second": 10.915,
+      "eval_steps_per_second": 1.365,
+      "eval_wer": 0.5285295187846811,
+      "step": 30000
+    },
+    {
+      "epoch": 4.11,
+      "learning_rate": 0.0010629563300730126,
+      "loss": 0.2848,
+      "step": 30010
+    },
+    {
+      "epoch": 4.11,
+      "learning_rate": 0.0010624052899848465,
+      "loss": 0.3282,
+      "step": 30020
+    },
+    {
+      "epoch": 4.11,
+      "learning_rate": 0.00106185424989668,
+      "loss": 0.2998,
+      "step": 30030
+    },
+    {
+      "epoch": 4.12,
+      "learning_rate": 0.0010613032098085136,
+      "loss": 0.3141,
+      "step": 30040
+    },
+    {
+      "epoch": 4.12,
+      "learning_rate": 0.0010607521697203472,
+      "loss": 0.2444,
+      "step": 30050
+    },
+    {
+      "epoch": 4.12,
+      "learning_rate": 0.0010602011296321808,
+      "loss": 0.308,
+      "step": 30060
+    },
+    {
+      "epoch": 4.12,
+      "learning_rate": 0.0010596500895440144,
+      "loss": 0.2924,
+      "step": 30070
+    },
+    {
+      "epoch": 4.12,
+      "learning_rate": 0.0010590990494558478,
+      "loss": 0.2786,
+      "step": 30080
+    },
+    {
+      "epoch": 4.12,
+      "learning_rate": 0.0010585480093676816,
+      "loss": 0.2649,
+      "step": 30090
+    },
+    {
+      "epoch": 4.12,
+      "learning_rate": 0.001057996969279515,
+      "loss": 0.2262,
+      "step": 30100
+    },
+    {
+      "epoch": 4.12,
+      "eval_loss": 0.2103297859430313,
+      "eval_runtime": 1072.036,
+      "eval_samples_per_second": 10.898,
+      "eval_steps_per_second": 1.363,
+      "eval_wer": 0.4910170108250705,
+      "step": 30100
+    },
+    {
+      "epoch": 4.13,
+      "learning_rate": 0.0010574459291913488,
+      "loss": 0.3038,
+      "step": 30110
+    },
+    {
+      "epoch": 4.13,
+      "learning_rate": 0.0010568948891031822,
+      "loss": 0.2855,
+      "step": 30120
+    },
+    {
+      "epoch": 4.13,
+      "learning_rate": 0.0010563438490150158,
+      "loss": 0.282,
+      "step": 30130
+    },
+    {
+      "epoch": 4.13,
+      "learning_rate": 0.0010557928089268494,
+      "loss": 0.3044,
+      "step": 30140
+    },
+    {
+      "epoch": 4.13,
+      "learning_rate": 0.001055241768838683,
+      "loss": 0.3406,
+      "step": 30150
+    },
+    {
+      "epoch": 4.13,
+      "learning_rate": 0.0010546907287505166,
+      "loss": 0.3028,
+      "step": 30160
+    },
+    {
+      "epoch": 4.13,
+      "learning_rate": 0.0010541396886623502,
+      "loss": 0.3001,
+      "step": 30170
+    },
+    {
+      "epoch": 4.13,
+      "learning_rate": 0.0010535886485741838,
+      "loss": 0.2854,
+      "step": 30180
+    },
+    {
+      "epoch": 4.14,
+      "learning_rate": 0.0010530376084860174,
+      "loss": 0.2647,
+      "step": 30190
+    },
+    {
+      "epoch": 4.14,
+      "learning_rate": 0.001052486568397851,
+      "loss": 0.211,
+      "step": 30200
+    },
+    {
+      "epoch": 4.14,
+      "eval_loss": 0.21408416330814362,
+      "eval_runtime": 1071.6223,
+      "eval_samples_per_second": 10.902,
+      "eval_steps_per_second": 1.363,
+      "eval_wer": 0.49725961975802785,
+      "step": 30200
+    },
+    {
+      "epoch": 4.14,
+      "learning_rate": 0.0010519355283096846,
+      "loss": 0.303,
+      "step": 30210
+    },
+    {
+      "epoch": 4.14,
+      "learning_rate": 0.0010513844882215182,
+      "loss": 0.2747,
+      "step": 30220
+    },
+    {
+      "epoch": 4.14,
+      "learning_rate": 0.0010508334481333518,
+      "loss": 0.2795,
+      "step": 30230
+    },
+    {
+      "epoch": 4.14,
+      "learning_rate": 0.0010502824080451854,
+      "loss": 0.2587,
+      "step": 30240
+    },
+    {
+      "epoch": 4.14,
+      "learning_rate": 0.001049731367957019,
+      "loss": 0.2523,
+      "step": 30250
+    },
+    {
+      "epoch": 4.15,
+      "learning_rate": 0.0010491803278688525,
+      "loss": 0.2652,
+      "step": 30260
+    },
+    {
+      "epoch": 4.15,
+      "learning_rate": 0.001048629287780686,
+      "loss": 0.2757,
+      "step": 30270
+    },
+    {
+      "epoch": 4.15,
+      "learning_rate": 0.0010480782476925197,
+      "loss": 0.2658,
+      "step": 30280
+    },
+    {
+      "epoch": 4.15,
+      "learning_rate": 0.0010475272076043531,
+      "loss": 0.278,
+      "step": 30290
+    },
+    {
+      "epoch": 4.15,
+      "learning_rate": 0.001046976167516187,
+      "loss": 0.2396,
+      "step": 30300
+    },
+    {
+      "epoch": 4.15,
+      "eval_loss": 0.22242088615894318,
+      "eval_runtime": 1071.2354,
+      "eval_samples_per_second": 10.906,
+      "eval_steps_per_second": 1.364,
+      "eval_wer": 0.5072432457018102,
+      "step": 30300
+    },
+    {
+      "epoch": 4.15,
+      "learning_rate": 0.0010464251274280203,
+      "loss": 0.2818,
+      "step": 30310
+    },
+    {
+      "epoch": 4.15,
+      "learning_rate": 0.001045874087339854,
+      "loss": 0.2691,
+      "step": 30320
+    },
+    {
+      "epoch": 4.16,
+      "learning_rate": 0.0010453230472516875,
+      "loss": 0.263,
+      "step": 30330
+    },
+    {
+      "epoch": 4.16,
+      "learning_rate": 0.001044772007163521,
+      "loss": 0.2934,
+      "step": 30340
+    },
+    {
+      "epoch": 4.16,
+      "learning_rate": 0.001044220967075355,
+      "loss": 0.2531,
+      "step": 30350
+    },
+    {
+      "epoch": 4.16,
+      "learning_rate": 0.0010436699269871883,
+      "loss": 0.2494,
+      "step": 30360
+    },
+    {
+      "epoch": 4.16,
+      "learning_rate": 0.001043118886899022,
+      "loss": 0.2884,
+      "step": 30370
+    },
+    {
+      "epoch": 4.16,
+      "learning_rate": 0.0010425678468108555,
+      "loss": 0.3275,
+      "step": 30380
+    },
+    {
+      "epoch": 4.16,
+      "learning_rate": 0.001042016806722689,
+      "loss": 0.2684,
+      "step": 30390
+    },
+    {
+      "epoch": 4.16,
+      "learning_rate": 0.0010414657666345227,
+      "loss": 0.2576,
+      "step": 30400
+    },
+    {
+      "epoch": 4.16,
+      "eval_loss": 0.22411681711673737,
+      "eval_runtime": 1073.3614,
+      "eval_samples_per_second": 10.884,
+      "eval_steps_per_second": 1.361,
+      "eval_wer": 0.5104725734558355,
+      "step": 30400
+    },
+    {
+      "epoch": 4.17,
+      "learning_rate": 0.0010409147265463563,
+      "loss": 0.2704,
+      "step": 30410
+    },
+    {
+      "epoch": 4.17,
+      "learning_rate": 0.0010403636864581899,
+      "loss": 0.2862,
+      "step": 30420
+    },
+    {
+      "epoch": 4.17,
+      "learning_rate": 0.0010398126463700235,
+      "loss": 0.3088,
+      "step": 30430
+    },
+    {
+      "epoch": 4.17,
+      "learning_rate": 0.001039261606281857,
+      "loss": 0.3237,
+      "step": 30440
+    },
+    {
+      "epoch": 4.17,
+      "learning_rate": 0.0010387105661936907,
+      "loss": 0.2388,
+      "step": 30450
+    },
+    {
+      "epoch": 4.17,
+      "learning_rate": 0.001038159526105524,
+      "loss": 0.298,
+      "step": 30460
+    },
+    {
+      "epoch": 4.17,
+      "learning_rate": 0.0010376084860173578,
+      "loss": 0.2728,
+      "step": 30470
+    },
+    {
+      "epoch": 4.18,
+      "learning_rate": 0.0010370574459291912,
+      "loss": 0.2586,
+      "step": 30480
+    },
+    {
+      "epoch": 4.18,
+      "learning_rate": 0.001036506405841025,
+      "loss": 0.2575,
+      "step": 30490
+    },
+    {
+      "epoch": 4.18,
+      "learning_rate": 0.0010359553657528586,
+      "loss": 0.2326,
+      "step": 30500
+    },
+    {
+      "epoch": 4.18,
+      "eval_loss": 0.21918047964572906,
+      "eval_runtime": 1075.9687,
+      "eval_samples_per_second": 10.858,
+      "eval_steps_per_second": 1.358,
+      "eval_wer": 0.5034681160738652,
+      "step": 30500
+    },
+    {
+      "epoch": 4.18,
+      "learning_rate": 0.001035404325664692,
+      "loss": 0.2796,
+      "step": 30510
+    },
+    {
+      "epoch": 4.18,
+      "learning_rate": 0.0010348532855765258,
+      "loss": 0.3096,
+      "step": 30520
+    },
+    {
+      "epoch": 4.18,
+      "learning_rate": 0.0010343022454883592,
+      "loss": 0.2976,
+      "step": 30530
+    },
+    {
+      "epoch": 4.18,
+      "learning_rate": 0.001033751205400193,
+      "loss": 0.2713,
+      "step": 30540
+    },
+    {
+      "epoch": 4.19,
+      "learning_rate": 0.0010332001653120264,
+      "loss": 0.2523,
+      "step": 30550
+    },
+    {
+      "epoch": 4.19,
+      "learning_rate": 0.0010326491252238602,
+      "loss": 0.3196,
+      "step": 30560
+    },
+    {
+      "epoch": 4.19,
+      "learning_rate": 0.0010320980851356936,
+      "loss": 0.303,
+      "step": 30570
+    },
+    {
+      "epoch": 4.19,
+      "learning_rate": 0.0010315470450475272,
+      "loss": 0.2669,
+      "step": 30580
+    },
+    {
+      "epoch": 4.19,
+      "learning_rate": 0.0010309960049593608,
+      "loss": 0.2876,
+      "step": 30590
+    },
+    {
+      "epoch": 4.19,
+      "learning_rate": 0.0010304449648711944,
+      "loss": 0.2458,
+      "step": 30600
+    },
+    {
+      "epoch": 4.19,
+      "eval_loss": 0.21522223949432373,
+      "eval_runtime": 1072.5424,
+      "eval_samples_per_second": 10.893,
+      "eval_steps_per_second": 1.362,
+      "eval_wer": 0.5008300736832529,
+      "step": 30600
+    },
+    {
+      "epoch": 4.19,
+      "learning_rate": 0.001029893924783028,
+      "loss": 0.2986,
+      "step": 30610
+    },
+    {
+      "epoch": 4.2,
+      "learning_rate": 0.0010293428846948616,
+      "loss": 0.2728,
+      "step": 30620
+    },
+    {
+      "epoch": 4.2,
+      "learning_rate": 0.0010287918446066952,
+      "loss": 0.2647,
+      "step": 30630
+    },
+    {
+      "epoch": 4.2,
+      "learning_rate": 0.0010282408045185288,
+      "loss": 0.3169,
+      "step": 30640
+    },
+    {
+      "epoch": 4.2,
+      "learning_rate": 0.0010276897644303624,
+      "loss": 0.253,
+      "step": 30650
+    },
+    {
+      "epoch": 4.2,
+      "learning_rate": 0.001027138724342196,
+      "loss": 0.2856,
+      "step": 30660
+    },
+    {
+      "epoch": 4.2,
+      "learning_rate": 0.0010265876842540296,
+      "loss": 0.271,
+      "step": 30670
+    },
+    {
+      "epoch": 4.2,
+      "learning_rate": 0.0010260366441658631,
+      "loss": 0.279,
+      "step": 30680
+    },
+    {
+      "epoch": 4.2,
+      "learning_rate": 0.0010254856040776967,
+      "loss": 0.2772,
+      "step": 30690
+    },
+    {
+      "epoch": 4.21,
+      "learning_rate": 0.0010249345639895301,
+      "loss": 0.1914,
+      "step": 30700
+    },
+    {
+      "epoch": 4.21,
+      "eval_loss": 0.21186842024326324,
+      "eval_runtime": 1076.1657,
+      "eval_samples_per_second": 10.856,
+      "eval_steps_per_second": 1.358,
+      "eval_wer": 0.5024447375602656,
+      "step": 30700
+    },
+    {
+      "epoch": 4.21,
+      "learning_rate": 0.001024383523901364,
+      "loss": 0.3521,
+      "step": 30710
+    },
+    {
+      "epoch": 4.21,
+      "learning_rate": 0.0010238324838131973,
+      "loss": 0.2388,
+      "step": 30720
+    },
+    {
+      "epoch": 4.21,
+      "learning_rate": 0.0010232814437250311,
+      "loss": 0.2749,
+      "step": 30730
+    },
+    {
+      "epoch": 4.21,
+      "learning_rate": 0.0010227304036368645,
+      "loss": 0.3075,
+      "step": 30740
+    },
+    {
+      "epoch": 4.21,
+      "learning_rate": 0.0010221793635486983,
+      "loss": 0.2289,
+      "step": 30750
+    },
+    {
+      "epoch": 4.21,
+      "learning_rate": 0.0010216283234605317,
+      "loss": 0.2917,
+      "step": 30760
+    },
+    {
+      "epoch": 4.22,
+      "learning_rate": 0.0010210772833723653,
+      "loss": 0.2804,
+      "step": 30770
+    },
+    {
+      "epoch": 4.22,
+      "learning_rate": 0.0010205262432841989,
+      "loss": 0.2851,
+      "step": 30780
+    },
+    {
+      "epoch": 4.22,
+      "learning_rate": 0.0010199752031960325,
+      "loss": 0.2815,
+      "step": 30790
+    },
+    {
+      "epoch": 4.22,
+      "learning_rate": 0.001019424163107866,
+      "loss": 0.2418,
+      "step": 30800
+    },
+    {
+      "epoch": 4.22,
+      "eval_loss": 0.21286386251449585,
+      "eval_runtime": 1073.7098,
+      "eval_samples_per_second": 10.881,
+      "eval_steps_per_second": 1.361,
+      "eval_wer": 0.5073910670426636,
+      "step": 30800
+    },
+    {
+      "epoch": 4.22,
+      "learning_rate": 0.0010188731230196997,
+      "loss": 0.3113,
+      "step": 30810
+    },
+    {
+      "epoch": 4.22,
+      "learning_rate": 0.0010183220829315335,
+      "loss": 0.2754,
+      "step": 30820
+    },
+    {
+      "epoch": 4.22,
+      "learning_rate": 0.0010177710428433669,
+      "loss": 0.2996,
+      "step": 30830
+    },
+    {
+      "epoch": 4.23,
+      "learning_rate": 0.0010172200027552005,
+      "loss": 0.2835,
+      "step": 30840
+    },
+    {
+      "epoch": 4.23,
+      "learning_rate": 0.001016668962667034,
+      "loss": 0.2142,
+      "step": 30850
+    },
+    {
+      "epoch": 4.23,
+      "learning_rate": 0.0010161179225788677,
+      "loss": 0.2851,
+      "step": 30860
+    },
+    {
+      "epoch": 4.23,
+      "learning_rate": 0.0010155668824907013,
+      "loss": 0.4539,
+      "step": 30870
+    },
+    {
+      "epoch": 4.23,
+      "learning_rate": 0.0010150158424025349,
+      "loss": 0.3382,
+      "step": 30880
+    },
+    {
+      "epoch": 4.23,
+      "learning_rate": 0.0010144648023143682,
+      "loss": 0.3292,
+      "step": 30890
+    },
+    {
+      "epoch": 4.23,
+      "learning_rate": 0.001013913762226202,
+      "loss": 0.2311,
+      "step": 30900
+    },
+    {
+      "epoch": 4.23,
+      "eval_loss": 0.22134087979793549,
+      "eval_runtime": 1072.5236,
+      "eval_samples_per_second": 10.893,
+      "eval_steps_per_second": 1.362,
+      "eval_wer": 0.5088351678340762,
+      "step": 30900
+    },
+    {
+      "epoch": 4.23,
+      "learning_rate": 0.0010133627221380354,
+      "loss": 0.2896,
+      "step": 30910
+    },
+    {
+      "epoch": 4.24,
+      "learning_rate": 0.0010128116820498692,
+      "loss": 0.3029,
+      "step": 30920
+    },
+    {
+      "epoch": 4.24,
+      "learning_rate": 0.0010122606419617026,
+      "loss": 0.2959,
+      "step": 30930
+    },
+    {
+      "epoch": 4.24,
+      "learning_rate": 0.0010117096018735364,
+      "loss": 0.3107,
+      "step": 30940
+    },
+    {
+      "epoch": 4.24,
+      "learning_rate": 0.0010111585617853698,
+      "loss": 0.2339,
+      "step": 30950
+    },
+    {
+      "epoch": 4.24,
+      "learning_rate": 0.0010106075216972034,
+      "loss": 0.2614,
+      "step": 30960
+    },
+    {
+      "epoch": 4.24,
+      "learning_rate": 0.0010100564816090372,
+      "loss": 0.2992,
+      "step": 30970
+    },
+    {
+      "epoch": 4.24,
+      "learning_rate": 0.0010095054415208706,
+      "loss": 0.2765,
+      "step": 30980
+    },
+    {
+      "epoch": 4.25,
+      "learning_rate": 0.0010089544014327044,
+      "loss": 0.3219,
+      "step": 30990
+    },
+    {
+      "epoch": 4.25,
+      "learning_rate": 0.0010084033613445378,
+      "loss": 0.2543,
+      "step": 31000
+    },
+    {
+      "epoch": 4.25,
+      "eval_loss": 0.20993460714817047,
+      "eval_runtime": 1074.4851,
+      "eval_samples_per_second": 10.873,
+      "eval_steps_per_second": 1.36,
+      "eval_wer": 0.49896525061402713,
+      "step": 31000
+    },
+    {
+      "epoch": 4.25,
+      "learning_rate": 0.0010078523212563716,
+      "loss": 0.3386,
+      "step": 31010
+    },
+    {
+      "epoch": 4.25,
+      "learning_rate": 0.001007301281168205,
+      "loss": 0.2709,
+      "step": 31020
+    },
+    {
+      "epoch": 4.25,
+      "learning_rate": 0.0010067502410800386,
+      "loss": 0.2606,
+      "step": 31030
+    },
+    {
+      "epoch": 4.25,
+      "learning_rate": 0.0010061992009918722,
+      "loss": 0.329,
+      "step": 31040
+    },
+    {
+      "epoch": 4.25,
+      "learning_rate": 0.0010056481609037058,
+      "loss": 0.2571,
+      "step": 31050
+    },
+    {
+      "epoch": 4.26,
+      "learning_rate": 0.0010050971208155394,
+      "loss": 0.2765,
+      "step": 31060
+    },
+    {
+      "epoch": 4.26,
+      "learning_rate": 0.001004546080727373,
+      "loss": 0.2879,
+      "step": 31070
+    },
+    {
+      "epoch": 4.26,
+      "learning_rate": 0.0010039950406392063,
+      "loss": 0.3164,
+      "step": 31080
+    },
+    {
+      "epoch": 4.26,
+      "learning_rate": 0.0010034440005510402,
+      "loss": 0.2685,
+      "step": 31090
+    },
+    {
+      "epoch": 4.26,
+      "learning_rate": 0.0010028929604628735,
+      "loss": 0.256,
+      "step": 31100
+    },
+    {
+      "epoch": 4.26,
+      "eval_loss": 0.20811431109905243,
+      "eval_runtime": 1073.2305,
+      "eval_samples_per_second": 10.886,
+      "eval_steps_per_second": 1.361,
+      "eval_wer": 0.4982829982716274,
+      "step": 31100
+    },
+    {
+      "epoch": 4.26,
+      "learning_rate": 0.0010023419203747073,
+      "loss": 0.2534,
+      "step": 31110
+    },
+    {
+      "epoch": 4.26,
+      "learning_rate": 0.001001790880286541,
+      "loss": 0.2653,
+      "step": 31120
+    },
+    {
+      "epoch": 4.26,
+      "learning_rate": 0.0010012398401983745,
+      "loss": 0.259,
+      "step": 31130
+    },
+    {
+      "epoch": 4.27,
+      "learning_rate": 0.0010006888001102081,
+      "loss": 0.2747,
+      "step": 31140
+    },
+    {
+      "epoch": 4.27,
+      "learning_rate": 0.0010001377600220415,
+      "loss": 0.2421,
+      "step": 31150
+    },
+    {
+      "epoch": 4.27,
+      "learning_rate": 0.000999641823942692,
+      "loss": 0.7671,
+      "step": 31160
+    },
+    {
+      "epoch": 4.27,
+      "learning_rate": 0.0009990907838545253,
+      "loss": 0.2536,
+      "step": 31170
+    },
+    {
+      "epoch": 4.27,
+      "learning_rate": 0.000998539743766359,
+      "loss": 0.2793,
+      "step": 31180
+    },
+    {
+      "epoch": 4.27,
+      "learning_rate": 0.0009979887036781925,
+      "loss": 0.2915,
+      "step": 31190
+    },
+    {
+      "epoch": 4.27,
+      "learning_rate": 0.0009974376635900261,
+      "loss": 0.2918,
+      "step": 31200
+    },
+    {
+      "epoch": 4.27,
+      "eval_loss": 0.2116004377603531,
+      "eval_runtime": 1073.4517,
+      "eval_samples_per_second": 10.884,
+      "eval_steps_per_second": 1.361,
+      "eval_wer": 0.5017283726007459,
+      "step": 31200
+    },
+    {
+      "epoch": 4.28,
+      "learning_rate": 0.0009968866235018597,
+      "loss": 0.2793,
+      "step": 31210
+    },
+    {
+      "epoch": 4.28,
+      "learning_rate": 0.0009963355834136933,
+      "loss": 0.5741,
+      "step": 31220
+    },
+    {
+      "epoch": 4.28,
+      "learning_rate": 0.000995784543325527,
+      "loss": 0.2891,
+      "step": 31230
+    },
+    {
+      "epoch": 4.28,
+      "learning_rate": 0.0009952335032373605,
+      "loss": 0.2775,
+      "step": 31240
+    },
+    {
+      "epoch": 4.28,
+      "learning_rate": 0.0009946824631491941,
+      "loss": 0.245,
+      "step": 31250
+    },
+    {
+      "epoch": 4.28,
+      "learning_rate": 0.0009941314230610277,
+      "loss": 0.2784,
+      "step": 31260
+    },
+    {
+      "epoch": 4.28,
+      "learning_rate": 0.0009935803829728613,
+      "loss": 0.2594,
+      "step": 31270
+    },
+    {
+      "epoch": 4.29,
+      "learning_rate": 0.000993029342884695,
+      "loss": 0.3061,
+      "step": 31280
+    },
+    {
+      "epoch": 4.29,
+      "learning_rate": 0.0009924783027965285,
+      "loss": 0.2634,
+      "step": 31290
+    },
+    {
+      "epoch": 4.29,
+      "learning_rate": 0.000991927262708362,
+      "loss": 0.2416,
+      "step": 31300
+    },
+    {
+      "epoch": 4.29,
+      "eval_loss": 0.20237690210342407,
+      "eval_runtime": 1072.316,
+      "eval_samples_per_second": 10.895,
+      "eval_steps_per_second": 1.362,
+      "eval_wer": 0.4900959701628309,
+      "step": 31300
+    },
+    {
+      "epoch": 4.29,
+      "learning_rate": 0.0009913762226201957,
+      "loss": 0.2792,
+      "step": 31310
+    },
+    {
+      "epoch": 4.29,
+      "learning_rate": 0.0009908251825320293,
+      "loss": 0.2725,
+      "step": 31320
+    },
+    {
+      "epoch": 4.29,
+      "learning_rate": 0.0009902741424438629,
+      "loss": 0.2881,
+      "step": 31330
+    },
+    {
+      "epoch": 4.29,
+      "learning_rate": 0.0009897231023556965,
+      "loss": 0.2662,
+      "step": 31340
+    },
+    {
+      "epoch": 4.3,
+      "learning_rate": 0.00098917206226753,
+      "loss": 0.2234,
+      "step": 31350
+    },
+    {
+      "epoch": 4.3,
+      "learning_rate": 0.0009886210221793635,
+      "loss": 0.2634,
+      "step": 31360
+    },
+    {
+      "epoch": 4.3,
+      "learning_rate": 0.000988069982091197,
+      "loss": 0.2639,
+      "step": 31370
+    },
+    {
+      "epoch": 4.3,
+      "learning_rate": 0.0009875189420030306,
+      "loss": 0.2853,
+      "step": 31380
+    },
+    {
+      "epoch": 4.3,
+      "learning_rate": 0.0009869679019148642,
+      "loss": 0.2577,
+      "step": 31390
+    },
+    {
+      "epoch": 4.3,
+      "learning_rate": 0.0009864168618266978,
+      "loss": 0.1784,
+      "step": 31400
+    },
+    {
+      "epoch": 4.3,
+      "eval_loss": 0.20894666016101837,
+      "eval_runtime": 1071.5541,
+      "eval_samples_per_second": 10.903,
+      "eval_steps_per_second": 1.363,
+      "eval_wer": 0.4927112708086964,
+      "step": 31400
+    },
+    {
+      "epoch": 4.3,
+      "learning_rate": 0.0009858658217385314,
+      "loss": 0.2739,
+      "step": 31410
+    },
+    {
+      "epoch": 4.3,
+      "learning_rate": 0.000985314781650365,
+      "loss": 0.2778,
+      "step": 31420
+    },
+    {
+      "epoch": 4.31,
+      "learning_rate": 0.0009847637415621986,
+      "loss": 0.2726,
+      "step": 31430
+    },
+    {
+      "epoch": 4.31,
+      "learning_rate": 0.0009842127014740322,
+      "loss": 0.2796,
+      "step": 31440
+    },
+    {
+      "epoch": 4.31,
+      "learning_rate": 0.0009836616613858658,
+      "loss": 0.2061,
+      "step": 31450
+    },
+    {
+      "epoch": 4.31,
+      "learning_rate": 0.0009831106212976994,
+      "loss": 0.2689,
+      "step": 31460
+    },
+    {
+      "epoch": 4.31,
+      "learning_rate": 0.000982559581209533,
+      "loss": 0.2602,
+      "step": 31470
+    },
+    {
+      "epoch": 4.31,
+      "learning_rate": 0.0009820085411213666,
+      "loss": 0.2685,
+      "step": 31480
+    },
+    {
+      "epoch": 4.31,
+      "learning_rate": 0.0009814575010332002,
+      "loss": 0.2537,
+      "step": 31490
+    },
+    {
+      "epoch": 4.32,
+      "learning_rate": 0.0009809064609450338,
+      "loss": 0.2046,
+      "step": 31500
+    },
+    {
+      "epoch": 4.32,
+      "eval_loss": 0.21403595805168152,
+      "eval_runtime": 1074.3148,
+      "eval_samples_per_second": 10.875,
+      "eval_steps_per_second": 1.36,
+      "eval_wer": 0.5046052033111981,
+      "step": 31500
+    },
+    {
+      "epoch": 4.32,
+      "learning_rate": 0.000980410524865684,
+      "loss": 0.9066,
+      "step": 31510
+    },
+    {
+      "epoch": 4.32,
+      "learning_rate": 0.0009798594847775176,
+      "loss": 0.2592,
+      "step": 31520
+    },
+    {
+      "epoch": 4.32,
+      "learning_rate": 0.0009793084446893512,
+      "loss": 0.2858,
+      "step": 31530
+    },
+    {
+      "epoch": 4.32,
+      "learning_rate": 0.0009787574046011848,
+      "loss": 0.2922,
+      "step": 31540
+    },
+    {
+      "epoch": 4.32,
+      "learning_rate": 0.0009782063645130184,
+      "loss": 0.2096,
+      "step": 31550
+    },
+    {
+      "epoch": 4.32,
+      "learning_rate": 0.000977655324424852,
+      "loss": 0.2717,
+      "step": 31560
+    },
+    {
+      "epoch": 4.33,
+      "learning_rate": 0.0009771042843366856,
+      "loss": 0.2572,
+      "step": 31570
+    },
+    {
+      "epoch": 4.33,
+      "learning_rate": 0.0009765532442485191,
+      "loss": 0.2516,
+      "step": 31580
+    },
+    {
+      "epoch": 4.33,
+      "learning_rate": 0.0009760022041603527,
+      "loss": 0.2365,
+      "step": 31590
+    },
+    {
+      "epoch": 4.33,
+      "learning_rate": 0.0009754511640721863,
+      "loss": 0.2546,
+      "step": 31600
+    },
+    {
+      "epoch": 4.33,
+      "eval_loss": 0.20593896508216858,
+      "eval_runtime": 1072.5559,
+      "eval_samples_per_second": 10.893,
+      "eval_steps_per_second": 1.362,
+      "eval_wer": 0.49005048667333756,
+      "step": 31600
+    },
+    {
+      "epoch": 4.33,
+      "learning_rate": 0.0009749001239840199,
+      "loss": 0.272,
+      "step": 31610
+    },
+    {
+      "epoch": 4.33,
+      "learning_rate": 0.0009743490838958534,
+      "loss": 0.2574,
+      "step": 31620
+    },
+    {
+      "epoch": 4.33,
+      "learning_rate": 0.000973798043807687,
+      "loss": 0.304,
+      "step": 31630
+    },
+    {
+      "epoch": 4.33,
+      "learning_rate": 0.0009732470037195206,
+      "loss": 0.272,
+      "step": 31640
+    },
+    {
+      "epoch": 4.34,
+      "learning_rate": 0.0009726959636313542,
+      "loss": 0.2348,
+      "step": 31650
+    },
+    {
+      "epoch": 4.34,
+      "learning_rate": 0.0009721449235431879,
+      "loss": 0.2724,
+      "step": 31660
+    },
+    {
+      "epoch": 4.34,
+      "learning_rate": 0.0009715938834550215,
+      "loss": 0.2763,
+      "step": 31670
+    },
+    {
+      "epoch": 4.34,
+      "learning_rate": 0.000971042843366855,
+      "loss": 0.2608,
+      "step": 31680
+    },
+    {
+      "epoch": 4.34,
+      "learning_rate": 0.0009704918032786885,
+      "loss": 0.2553,
+      "step": 31690
+    },
+    {
+      "epoch": 4.34,
+      "learning_rate": 0.0009699407631905221,
+      "loss": 0.2133,
+      "step": 31700
+    },
+    {
+      "epoch": 4.34,
+      "eval_loss": 0.2052864283323288,
+      "eval_runtime": 1072.3319,
+      "eval_samples_per_second": 10.895,
+      "eval_steps_per_second": 1.362,
+      "eval_wer": 0.4910966069316838,
+      "step": 31700
+    },
+    {
+      "epoch": 4.34,
+      "learning_rate": 0.0009693897231023557,
+      "loss": 0.2825,
+      "step": 31710
+    },
+    {
+      "epoch": 4.35,
+      "learning_rate": 0.0009688386830141893,
+      "loss": 0.265,
+      "step": 31720
+    },
+    {
+      "epoch": 4.35,
+      "learning_rate": 0.0009682876429260229,
+      "loss": 0.2435,
+      "step": 31730
+    },
+    {
+      "epoch": 4.35,
+      "learning_rate": 0.0009677366028378564,
+      "loss": 0.2917,
+      "step": 31740
+    },
+    {
+      "epoch": 4.35,
+      "learning_rate": 0.00096718556274969,
+      "loss": 0.2133,
+      "step": 31750
+    },
+    {
+      "epoch": 4.35,
+      "learning_rate": 0.0009666345226615236,
+      "loss": 0.2875,
+      "step": 31760
+    },
+    {
+      "epoch": 4.35,
+      "learning_rate": 0.0009660834825733572,
+      "loss": 0.2758,
+      "step": 31770
+    },
+    {
+      "epoch": 4.35,
+      "learning_rate": 0.0009655324424851908,
+      "loss": 0.2619,
+      "step": 31780
+    },
+    {
+      "epoch": 4.36,
+      "learning_rate": 0.0009649814023970244,
+      "loss": 0.2567,
+      "step": 31790
+    },
+    {
+      "epoch": 4.36,
+      "learning_rate": 0.000964430362308858,
+      "loss": 0.2166,
+      "step": 31800
+    },
+    {
+      "epoch": 4.36,
+      "eval_loss": 0.19733890891075134,
+      "eval_runtime": 1074.632,
+      "eval_samples_per_second": 10.872,
+      "eval_steps_per_second": 1.36,
+      "eval_wer": 0.47671245337942325,
+      "step": 31800
+    },
+    {
+      "epoch": 4.36,
+      "learning_rate": 0.0009638793222206915,
+      "loss": 0.2645,
+      "step": 31810
+    },
+    {
+      "epoch": 4.36,
+      "learning_rate": 0.0009633282821325252,
+      "loss": 0.277,
+      "step": 31820
+    },
+    {
+      "epoch": 4.36,
+      "learning_rate": 0.0009627772420443588,
+      "loss": 0.2559,
+      "step": 31830
+    },
+    {
+      "epoch": 4.36,
+      "learning_rate": 0.0009622262019561924,
+      "loss": 0.2842,
+      "step": 31840
+    },
+    {
+      "epoch": 4.36,
+      "learning_rate": 0.000961675161868026,
+      "loss": 0.2549,
+      "step": 31850
+    },
+    {
+      "epoch": 4.36,
+      "learning_rate": 0.0009611241217798596,
+      "loss": 0.2607,
+      "step": 31860
+    },
+    {
+      "epoch": 4.37,
+      "learning_rate": 0.0009605730816916931,
+      "loss": 0.2932,
+      "step": 31870
+    },
+    {
+      "epoch": 4.37,
+      "learning_rate": 0.0009600220416035267,
+      "loss": 0.2526,
+      "step": 31880
+    },
+    {
+      "epoch": 4.37,
+      "learning_rate": 0.0009594710015153603,
+      "loss": 0.25,
+      "step": 31890
+    },
+    {
+      "epoch": 4.37,
+      "learning_rate": 0.0009589199614271939,
+      "loss": 0.2095,
+      "step": 31900
+    },
+    {
+      "epoch": 4.37,
+      "eval_loss": 0.21638716757297516,
+      "eval_runtime": 1077.6831,
+      "eval_samples_per_second": 10.841,
+      "eval_steps_per_second": 1.356,
+      "eval_wer": 0.4954630219230419,
+      "step": 31900
+    },
+    {
+      "epoch": 4.37,
+      "learning_rate": 0.0009583689213390274,
+      "loss": 0.2993,
+      "step": 31910
+    },
+    {
+      "epoch": 4.37,
+      "learning_rate": 0.000957817881250861,
+      "loss": 0.2605,
+      "step": 31920
+    },
+    {
+      "epoch": 4.37,
+      "learning_rate": 0.0009572668411626945,
+      "loss": 0.2626,
+      "step": 31930
+    },
+    {
+      "epoch": 4.38,
+      "learning_rate": 0.0009567158010745281,
+      "loss": 0.2583,
+      "step": 31940
+    },
+    {
+      "epoch": 4.38,
+      "learning_rate": 0.0009561647609863617,
+      "loss": 0.2045,
+      "step": 31950
+    },
+    {
+      "epoch": 4.38,
+      "learning_rate": 0.0009556137208981953,
+      "loss": 0.2848,
+      "step": 31960
+    },
+    {
+      "epoch": 4.38,
+      "learning_rate": 0.0009550626808100289,
+      "loss": 0.2641,
+      "step": 31970
+    },
+    {
+      "epoch": 4.38,
+      "learning_rate": 0.0009545116407218626,
+      "loss": 0.2634,
+      "step": 31980
+    },
+    {
+      "epoch": 4.38,
+      "learning_rate": 0.0009539606006336962,
+      "loss": 0.2874,
+      "step": 31990
+    },
+    {
+      "epoch": 4.38,
+      "learning_rate": 0.0009534095605455297,
+      "loss": 0.2604,
+      "step": 32000
+    },
+    {
+      "epoch": 4.38,
+      "eval_loss": 0.20295126736164093,
+      "eval_runtime": 1075.6167,
+      "eval_samples_per_second": 10.862,
+      "eval_steps_per_second": 1.358,
+      "eval_wer": 0.4873783316656054,
+      "step": 32000
+    },
+    {
+      "epoch": 4.39,
+      "learning_rate": 0.0009528585204573633,
+      "loss": 0.3016,
+      "step": 32010
+    },
+    {
+      "epoch": 4.39,
+      "learning_rate": 0.0009523074803691969,
+      "loss": 0.2958,
+      "step": 32020
+    },
+    {
+      "epoch": 4.39,
+      "learning_rate": 0.0009517564402810305,
+      "loss": 0.2845,
+      "step": 32030
+    },
+    {
+      "epoch": 4.39,
+      "learning_rate": 0.0009512054001928641,
+      "loss": 0.2528,
+      "step": 32040
+    },
+    {
+      "epoch": 4.39,
+      "learning_rate": 0.0009506543601046977,
+      "loss": 0.2984,
+      "step": 32050
+    },
+    {
+      "epoch": 4.39,
+      "learning_rate": 0.0009501033200165312,
+      "loss": 0.2924,
+      "step": 32060
+    },
+    {
+      "epoch": 4.39,
+      "learning_rate": 0.0009495522799283648,
+      "loss": 0.2488,
+      "step": 32070
+    },
+    {
+      "epoch": 4.4,
+      "learning_rate": 0.0009490012398401984,
+      "loss": 0.2615,
+      "step": 32080
+    },
+    {
+      "epoch": 4.4,
+      "learning_rate": 0.000948450199752032,
+      "loss": 0.3123,
+      "step": 32090
+    },
+    {
+      "epoch": 4.4,
+      "learning_rate": 0.0009478991596638656,
+      "loss": 0.267,
+      "step": 32100
+    },
+    {
+      "epoch": 4.4,
+      "eval_loss": 0.20251768827438354,
+      "eval_runtime": 1072.9081,
+      "eval_samples_per_second": 10.889,
+      "eval_steps_per_second": 1.362,
+      "eval_wer": 0.49240425725461656,
+      "step": 32100
+    },
+    {
+      "epoch": 4.4,
+      "learning_rate": 0.0009473481195756992,
+      "loss": 0.2751,
+      "step": 32110
+    },
+    {
+      "epoch": 4.4,
+      "learning_rate": 0.0009467970794875326,
+      "loss": 0.2575,
+      "step": 32120
+    },
+    {
+      "epoch": 4.4,
+      "learning_rate": 0.0009462460393993663,
+      "loss": 0.2545,
+      "step": 32130
+    },
+    {
+      "epoch": 4.4,
+      "learning_rate": 0.0009456949993111999,
+      "loss": 0.2558,
+      "step": 32140
+    },
+    {
+      "epoch": 4.4,
+      "learning_rate": 0.0009451439592230335,
+      "loss": 0.2395,
+      "step": 32150
+    },
+    {
+      "epoch": 4.41,
+      "learning_rate": 0.0009445929191348671,
+      "loss": 0.267,
+      "step": 32160
+    },
+    {
+      "epoch": 4.41,
+      "learning_rate": 0.0009440418790467007,
+      "loss": 0.2756,
+      "step": 32170
+    },
+    {
+      "epoch": 4.41,
+      "learning_rate": 0.0009434908389585342,
+      "loss": 0.2685,
+      "step": 32180
+    },
+    {
+      "epoch": 4.41,
+      "learning_rate": 0.0009429397988703678,
+      "loss": 0.2969,
+      "step": 32190
+    },
+    {
+      "epoch": 4.41,
+      "learning_rate": 0.0009423887587822014,
+      "loss": 0.2053,
+      "step": 32200
+    },
+    {
+      "epoch": 4.41,
+      "eval_loss": 0.20219683647155762,
+      "eval_runtime": 1072.8538,
+      "eval_samples_per_second": 10.89,
+      "eval_steps_per_second": 1.362,
+      "eval_wer": 0.48469480578549984,
+      "step": 32200
+    },
+    {
+      "epoch": 4.41,
+      "learning_rate": 0.000941837718694035,
+      "loss": 0.2856,
+      "step": 32210
+    },
+    {
+      "epoch": 4.41,
+      "learning_rate": 0.0009412866786058686,
+      "loss": 0.2544,
+      "step": 32220
+    },
+    {
+      "epoch": 4.42,
+      "learning_rate": 0.0009407356385177022,
+      "loss": 0.248,
+      "step": 32230
+    },
+    {
+      "epoch": 4.42,
+      "learning_rate": 0.0009401845984295358,
+      "loss": 0.298,
+      "step": 32240
+    },
+    {
+      "epoch": 4.42,
+      "learning_rate": 0.0009396335583413693,
+      "loss": 0.2955,
+      "step": 32250
+    },
+    {
+      "epoch": 4.42,
+      "learning_rate": 0.0009390825182532029,
+      "loss": 0.2856,
+      "step": 32260
+    },
+    {
+      "epoch": 4.42,
+      "learning_rate": 0.0009385314781650365,
+      "loss": 0.2657,
+      "step": 32270
+    },
+    {
+      "epoch": 4.42,
+      "learning_rate": 0.0009379804380768701,
+      "loss": 0.285,
+      "step": 32280
+    },
+    {
+      "epoch": 4.42,
+      "learning_rate": 0.0009374293979887038,
+      "loss": 0.2986,
+      "step": 32290
+    },
+    {
+      "epoch": 4.43,
+      "learning_rate": 0.0009368783579005374,
+      "loss": 0.2527,
+      "step": 32300
+    },
+    {
+      "epoch": 4.43,
+      "eval_loss": 0.19755671918392181,
+      "eval_runtime": 1073.2221,
+      "eval_samples_per_second": 10.886,
+      "eval_steps_per_second": 1.361,
+      "eval_wer": 0.47881606476848904,
+      "step": 32300
+    },
+    {
+      "epoch": 4.43,
+      "learning_rate": 0.0009363273178123709,
+      "loss": 0.2673,
+      "step": 32310
+    },
+    {
+      "epoch": 4.43,
+      "learning_rate": 0.0009357762777242045,
+      "loss": 0.2667,
+      "step": 32320
+    },
+    {
+      "epoch": 4.43,
+      "learning_rate": 0.000935225237636038,
+      "loss": 0.2565,
+      "step": 32330
+    },
+    {
+      "epoch": 4.43,
+      "learning_rate": 0.0009346741975478716,
+      "loss": 0.2703,
+      "step": 32340
+    },
+    {
+      "epoch": 4.43,
+      "learning_rate": 0.0009341231574597052,
+      "loss": 0.2538,
+      "step": 32350
+    },
+    {
+      "epoch": 4.43,
+      "learning_rate": 0.0009335721173715388,
+      "loss": 0.2691,
+      "step": 32360
+    },
+    {
+      "epoch": 4.43,
+      "learning_rate": 0.0009330210772833723,
+      "loss": 0.3173,
+      "step": 32370
+    },
+    {
+      "epoch": 4.44,
+      "learning_rate": 0.0009324700371952059,
+      "loss": 0.273,
+      "step": 32380
+    },
+    {
+      "epoch": 4.44,
+      "learning_rate": 0.0009319189971070395,
+      "loss": 0.2621,
+      "step": 32390
+    },
+    {
+      "epoch": 4.44,
+      "learning_rate": 0.0009313679570188731,
+      "loss": 0.2553,
+      "step": 32400
+    },
+    {
+      "epoch": 4.44,
+      "eval_loss": 0.2012455016374588,
+      "eval_runtime": 1073.7361,
+      "eval_samples_per_second": 10.881,
+      "eval_steps_per_second": 1.361,
+      "eval_wer": 0.48026016555990175,
+      "step": 32400
+    },
+    {
+      "epoch": 4.44,
+      "learning_rate": 0.0009308169169307067,
+      "loss": 0.2952,
+      "step": 32410
+    },
+    {
+      "epoch": 4.44,
+      "learning_rate": 0.0009302658768425403,
+      "loss": 0.2893,
+      "step": 32420
+    },
+    {
+      "epoch": 4.44,
+      "learning_rate": 0.0009297148367543739,
+      "loss": 0.3214,
+      "step": 32430
+    },
+    {
+      "epoch": 4.44,
+      "learning_rate": 0.0009291637966662075,
+      "loss": 0.2771,
+      "step": 32440
+    },
+    {
+      "epoch": 4.45,
+      "learning_rate": 0.0009286127565780411,
+      "loss": 0.2284,
+      "step": 32450
+    },
+    {
+      "epoch": 4.45,
+      "learning_rate": 0.0009280617164898747,
+      "loss": 0.2941,
+      "step": 32460
+    },
+    {
+      "epoch": 4.45,
+      "learning_rate": 0.0009275106764017083,
+      "loss": 0.2479,
+      "step": 32470
+    },
+    {
+      "epoch": 4.45,
+      "learning_rate": 0.0009269596363135419,
+      "loss": 0.2394,
+      "step": 32480
+    },
+    {
+      "epoch": 4.45,
+      "learning_rate": 0.0009264085962253755,
+      "loss": 0.2308,
+      "step": 32490
+    },
+    {
+      "epoch": 4.45,
+      "learning_rate": 0.000925857556137209,
+      "loss": 0.2374,
+      "step": 32500
+    },
+    {
+      "epoch": 4.45,
+      "eval_loss": 0.19127798080444336,
+      "eval_runtime": 1073.3164,
+      "eval_samples_per_second": 10.885,
+      "eval_steps_per_second": 1.361,
+      "eval_wer": 0.47127717638497224,
+      "step": 32500
+    },
+    {
+      "epoch": 4.45,
+      "learning_rate": 0.0009253065160490426,
+      "loss": 0.2872,
+      "step": 32510
+    },
+    {
+      "epoch": 4.46,
+      "learning_rate": 0.0009247554759608762,
+      "loss": 0.2528,
+      "step": 32520
+    },
+    {
+      "epoch": 4.46,
+      "learning_rate": 0.0009242044358727098,
+      "loss": 0.2645,
+      "step": 32530
+    },
+    {
+      "epoch": 4.46,
+      "learning_rate": 0.0009236533957845434,
+      "loss": 0.2877,
+      "step": 32540
+    },
+    {
+      "epoch": 4.46,
+      "learning_rate": 0.000923102355696377,
+      "loss": 0.2691,
+      "step": 32550
+    },
+    {
+      "epoch": 4.46,
+      "learning_rate": 0.0009225513156082104,
+      "loss": 0.2627,
+      "step": 32560
+    },
+    {
+      "epoch": 4.46,
+      "learning_rate": 0.000922000275520044,
+      "loss": 0.2901,
+      "step": 32570
+    },
+    {
+      "epoch": 4.46,
+      "learning_rate": 0.0009214492354318776,
+      "loss": 0.2575,
+      "step": 32580
+    },
+    {
+      "epoch": 4.46,
+      "learning_rate": 0.0009208981953437112,
+      "loss": 0.2792,
+      "step": 32590
+    },
+    {
+      "epoch": 4.47,
+      "learning_rate": 0.0009203471552555449,
+      "loss": 0.2054,
+      "step": 32600
+    },
+    {
+      "epoch": 4.47,
+      "eval_loss": 0.194502592086792,
+      "eval_runtime": 1073.816,
+      "eval_samples_per_second": 10.88,
+      "eval_steps_per_second": 1.361,
+      "eval_wer": 0.47579141271718367,
+      "step": 32600
+    },
+    {
+      "epoch": 4.47,
+      "learning_rate": 0.0009197961151673785,
+      "loss": 0.2761,
+      "step": 32610
+    },
+    {
+      "epoch": 4.47,
+      "learning_rate": 0.0009192450750792121,
+      "loss": 0.2447,
+      "step": 32620
+    },
+    {
+      "epoch": 4.47,
+      "learning_rate": 0.0009186940349910456,
+      "loss": 0.2518,
+      "step": 32630
+    },
+    {
+      "epoch": 4.47,
+      "learning_rate": 0.0009181429949028792,
+      "loss": 0.2568,
+      "step": 32640
+    },
+    {
+      "epoch": 4.47,
+      "learning_rate": 0.0009175919548147128,
+      "loss": 0.2143,
+      "step": 32650
+    },
+    {
+      "epoch": 4.47,
+      "learning_rate": 0.0009170409147265464,
+      "loss": 0.2515,
+      "step": 32660
+    },
+    {
+      "epoch": 4.48,
+      "learning_rate": 0.00091648987463838,
+      "loss": 0.2773,
+      "step": 32670
+    },
+    {
+      "epoch": 4.48,
+      "learning_rate": 0.0009159388345502136,
+      "loss": 0.2375,
+      "step": 32680
+    },
+    {
+      "epoch": 4.48,
+      "learning_rate": 0.0009153877944620471,
+      "loss": 0.264,
+      "step": 32690
+    },
+    {
+      "epoch": 4.48,
+      "learning_rate": 0.0009148367543738807,
+      "loss": 0.195,
+      "step": 32700
+    },
+    {
+      "epoch": 4.48,
+      "eval_loss": 0.19825096428394318,
+      "eval_runtime": 1073.5486,
+      "eval_samples_per_second": 10.883,
+      "eval_steps_per_second": 1.361,
+      "eval_wer": 0.47943009187664876,
+      "step": 32700
+    },
+    {
+      "epoch": 4.48,
+      "learning_rate": 0.0009142857142857143,
+      "loss": 0.2566,
+      "step": 32710
+    },
+    {
+      "epoch": 4.48,
+      "learning_rate": 0.0009137346741975479,
+      "loss": 0.2593,
+      "step": 32720
+    },
+    {
+      "epoch": 4.48,
+      "learning_rate": 0.0009131836341093815,
+      "loss": 0.2762,
+      "step": 32730
+    },
+    {
+      "epoch": 4.49,
+      "learning_rate": 0.0009126325940212151,
+      "loss": 0.2398,
+      "step": 32740
+    },
+    {
+      "epoch": 4.49,
+      "learning_rate": 0.0009120815539330487,
+      "loss": 0.2726,
+      "step": 32750
+    },
+    {
+      "epoch": 4.49,
+      "learning_rate": 0.0009115305138448822,
+      "loss": 0.2608,
+      "step": 32760
+    },
+    {
+      "epoch": 4.49,
+      "learning_rate": 0.0009109794737567158,
+      "loss": 0.2712,
+      "step": 32770
+    },
+    {
+      "epoch": 4.49,
+      "learning_rate": 0.0009104284336685494,
+      "loss": 0.2457,
+      "step": 32780
+    },
+    {
+      "epoch": 4.49,
+      "learning_rate": 0.000909877393580383,
+      "loss": 0.2974,
+      "step": 32790
+    },
+    {
+      "epoch": 4.49,
+      "learning_rate": 0.0009093263534922166,
+      "loss": 0.2873,
+      "step": 32800
+    },
+    {
+      "epoch": 4.49,
+      "eval_loss": 0.19646403193473816,
+      "eval_runtime": 1072.5117,
+      "eval_samples_per_second": 10.893,
+      "eval_steps_per_second": 1.362,
+      "eval_wer": 0.48649140362048576,
+      "step": 32800
+    },
+    {
+      "epoch": 4.5,
+      "learning_rate": 0.0009087753134040502,
+      "loss": 0.2744,
+      "step": 32810
+    },
+    {
+      "epoch": 4.5,
+      "learning_rate": 0.0009082242733158837,
+      "loss": 0.2666,
+      "step": 32820
+    },
+    {
+      "epoch": 4.5,
+      "learning_rate": 0.0009076732332277173,
+      "loss": 0.3244,
+      "step": 32830
+    },
+    {
+      "epoch": 4.5,
+      "learning_rate": 0.0009071221931395509,
+      "loss": 0.2537,
+      "step": 32840
+    },
+    {
+      "epoch": 4.5,
+      "learning_rate": 0.0009065711530513845,
+      "loss": 0.238,
+      "step": 32850
+    },
+    {
+      "epoch": 4.5,
+      "learning_rate": 0.0009060201129632181,
+      "loss": 0.2742,
+      "step": 32860
+    },
+    {
+      "epoch": 4.5,
+      "learning_rate": 0.0009054690728750517,
+      "loss": 0.253,
+      "step": 32870
+    },
+    {
+      "epoch": 4.5,
+      "learning_rate": 0.0009049180327868852,
+      "loss": 0.7734,
+      "step": 32880
+    },
+    {
+      "epoch": 4.51,
+      "learning_rate": 0.0009043669926987188,
+      "loss": 0.2617,
+      "step": 32890
+    },
+    {
+      "epoch": 4.51,
+      "learning_rate": 0.0009038159526105524,
+      "loss": 0.2529,
+      "step": 32900
+    },
+    {
+      "epoch": 4.51,
+      "eval_loss": 0.19125251471996307,
+      "eval_runtime": 1077.34,
+      "eval_samples_per_second": 10.844,
+      "eval_steps_per_second": 1.356,
+      "eval_wer": 0.4714591103429455,
+      "step": 32900
+    },
+    {
+      "epoch": 4.51,
+      "learning_rate": 0.0009032649125223861,
+      "loss": 0.2603,
+      "step": 32910
+    },
+    {
+      "epoch": 4.51,
+      "learning_rate": 0.0009027138724342197,
+      "loss": 0.5423,
+      "step": 32920
+    },
+    {
+      "epoch": 4.51,
+      "learning_rate": 0.0009021628323460533,
+      "loss": 0.2686,
+      "step": 32930
+    },
+    {
+      "epoch": 4.51,
+      "learning_rate": 0.0009016117922578868,
+      "loss": 0.2681,
+      "step": 32940
+    },
+    {
+      "epoch": 4.51,
+      "learning_rate": 0.0009010607521697204,
+      "loss": 0.2462,
+      "step": 32950
+    },
+    {
+      "epoch": 4.52,
+      "learning_rate": 0.000900509712081554,
+      "loss": 0.3144,
+      "step": 32960
+    },
+    {
+      "epoch": 4.52,
+      "learning_rate": 0.0008999586719933875,
+      "loss": 0.2797,
+      "step": 32970
+    },
+    {
+      "epoch": 4.52,
+      "learning_rate": 0.0008994076319052211,
+      "loss": 0.2876,
+      "step": 32980
+    },
+    {
+      "epoch": 4.52,
+      "learning_rate": 0.0008988565918170547,
+      "loss": 0.2871,
+      "step": 32990
+    },
+    {
+      "epoch": 4.52,
+      "learning_rate": 0.0008983055517288882,
+      "loss": 0.2309,
+      "step": 33000
+    },
+    {
+      "epoch": 4.52,
+      "eval_loss": 0.20021037757396698,
+      "eval_runtime": 1071.8381,
+      "eval_samples_per_second": 10.9,
+      "eval_steps_per_second": 1.363,
+      "eval_wer": 0.4792026744291822,
+      "step": 33000
+    },
+    {
+      "epoch": 4.52,
+      "learning_rate": 0.0008977545116407218,
+      "loss": 0.2726,
+      "step": 33010
+    },
+    {
+      "epoch": 4.52,
+      "learning_rate": 0.0008972034715525554,
+      "loss": 0.2351,
+      "step": 33020
+    },
+    {
+      "epoch": 4.53,
+      "learning_rate": 0.000896652431464389,
+      "loss": 0.2896,
+      "step": 33030
+    },
+    {
+      "epoch": 4.53,
+      "learning_rate": 0.0008961013913762226,
+      "loss": 0.3964,
+      "step": 33040
+    },
+    {
+      "epoch": 4.53,
+      "learning_rate": 0.0008955503512880562,
+      "loss": 0.2427,
+      "step": 33050
+    },
+    {
+      "epoch": 4.53,
+      "learning_rate": 0.0008949993111998899,
+      "loss": 0.2536,
+      "step": 33060
+    },
+    {
+      "epoch": 4.53,
+      "learning_rate": 0.0008944482711117234,
+      "loss": 0.2546,
+      "step": 33070
+    },
+    {
+      "epoch": 4.53,
+      "learning_rate": 0.000893897231023557,
+      "loss": 0.2655,
+      "step": 33080
+    },
+    {
+      "epoch": 4.53,
+      "learning_rate": 0.0008933461909353906,
+      "loss": 0.2504,
+      "step": 33090
+    },
+    {
+      "epoch": 4.53,
+      "learning_rate": 0.0008927951508472242,
+      "loss": 0.1991,
+      "step": 33100
+    },
+    {
+      "epoch": 4.53,
+      "eval_loss": 0.18560616672039032,
+      "eval_runtime": 1073.4922,
+      "eval_samples_per_second": 10.883,
+      "eval_steps_per_second": 1.361,
+      "eval_wer": 0.46803647775857365,
+      "step": 33100
+    },
+    {
+      "epoch": 4.54,
+      "learning_rate": 0.0008922441107590578,
+      "loss": 0.2977,
+      "step": 33110
+    },
+    {
+      "epoch": 4.54,
+      "learning_rate": 0.0008916930706708914,
+      "loss": 0.2651,
+      "step": 33120
+    },
+    {
+      "epoch": 4.54,
+      "learning_rate": 0.0008911420305827249,
+      "loss": 0.2958,
+      "step": 33130
+    },
+    {
+      "epoch": 4.54,
+      "learning_rate": 0.0008905909904945585,
+      "loss": 0.224,
+      "step": 33140
+    },
+    {
+      "epoch": 4.54,
+      "learning_rate": 0.0008900399504063921,
+      "loss": 0.2362,
+      "step": 33150
+    },
+    {
+      "epoch": 4.54,
+      "learning_rate": 0.0008894889103182257,
+      "loss": 0.2599,
+      "step": 33160
+    },
+    {
+      "epoch": 4.54,
+      "learning_rate": 0.0008889378702300593,
+      "loss": 0.26,
+      "step": 33170
+    },
+    {
+      "epoch": 4.55,
+      "learning_rate": 0.0008883868301418929,
+      "loss": 0.2581,
+      "step": 33180
+    },
+    {
+      "epoch": 4.55,
+      "learning_rate": 0.0008878357900537263,
+      "loss": 0.2867,
+      "step": 33190
+    },
+    {
+      "epoch": 4.55,
+      "learning_rate": 0.0008872847499655599,
+      "loss": 0.2009,
+      "step": 33200
+    },
+    {
+      "epoch": 4.55,
+      "eval_loss": 0.19034399092197418,
+      "eval_runtime": 1074.7717,
+      "eval_samples_per_second": 10.87,
+      "eval_steps_per_second": 1.359,
+      "eval_wer": 0.47526835258801053,
+      "step": 33200
+    },
+    {
+      "epoch": 4.55,
+      "learning_rate": 0.0008867337098773935,
+      "loss": 0.2718,
+      "step": 33210
+    },
+    {
+      "epoch": 4.55,
+      "learning_rate": 0.0008861826697892272,
+      "loss": 0.2557,
+      "step": 33220
+    },
+    {
+      "epoch": 4.55,
+      "learning_rate": 0.0008856316297010608,
+      "loss": 0.2694,
+      "step": 33230
+    },
+    {
+      "epoch": 4.55,
+      "learning_rate": 0.0008850805896128944,
+      "loss": 0.2623,
+      "step": 33240
+    },
+    {
+      "epoch": 4.56,
+      "learning_rate": 0.000884529549524728,
+      "loss": 0.2163,
+      "step": 33250
+    },
+    {
+      "epoch": 4.56,
+      "learning_rate": 0.0008839785094365615,
+      "loss": 0.2797,
+      "step": 33260
+    },
+    {
+      "epoch": 4.56,
+      "learning_rate": 0.0008834274693483951,
+      "loss": 0.2471,
+      "step": 33270
+    },
+    {
+      "epoch": 4.56,
+      "learning_rate": 0.0008828764292602287,
+      "loss": 0.2408,
+      "step": 33280
+    },
+    {
+      "epoch": 4.56,
+      "learning_rate": 0.0008823253891720623,
+      "loss": 0.2799,
+      "step": 33290
+    },
+    {
+      "epoch": 4.56,
+      "learning_rate": 0.0008817743490838959,
+      "loss": 0.2214,
+      "step": 33300
+    },
+    {
+      "epoch": 4.56,
+      "eval_loss": 0.1914065033197403,
+      "eval_runtime": 1074.0282,
+      "eval_samples_per_second": 10.878,
+      "eval_steps_per_second": 1.36,
+      "eval_wer": 0.47091330846902574,
+      "step": 33300
+    },
+    {
+      "epoch": 4.56,
+      "learning_rate": 0.0008812233089957295,
+      "loss": 0.252,
+      "step": 33310
+    },
+    {
+      "epoch": 4.57,
+      "learning_rate": 0.000880672268907563,
+      "loss": 0.2591,
+      "step": 33320
+    },
+    {
+      "epoch": 4.57,
+      "learning_rate": 0.0008801212288193966,
+      "loss": 0.2661,
+      "step": 33330
+    },
+    {
+      "epoch": 4.57,
+      "learning_rate": 0.0008795701887312302,
+      "loss": 0.2386,
+      "step": 33340
+    },
+    {
+      "epoch": 4.57,
+      "learning_rate": 0.0008790191486430638,
+      "loss": 0.2509,
+      "step": 33350
+    },
+    {
+      "epoch": 4.57,
+      "learning_rate": 0.0008784681085548974,
+      "loss": 0.2794,
+      "step": 33360
+    },
+    {
+      "epoch": 4.57,
+      "learning_rate": 0.0008779170684667311,
+      "loss": 0.2699,
+      "step": 33370
+    },
+    {
+      "epoch": 4.57,
+      "learning_rate": 0.0008773660283785646,
+      "loss": 0.3073,
+      "step": 33380
+    },
+    {
+      "epoch": 4.57,
+      "learning_rate": 0.0008768149882903982,
+      "loss": 0.2408,
+      "step": 33390
+    },
+    {
+      "epoch": 4.58,
+      "learning_rate": 0.0008762639482022317,
+      "loss": 0.2517,
+      "step": 33400
+    },
+    {
+      "epoch": 4.58,
+      "eval_loss": 0.18687430024147034,
+      "eval_runtime": 1072.874,
+      "eval_samples_per_second": 10.889,
+      "eval_steps_per_second": 1.362,
+      "eval_wer": 0.46561448194305466,
+      "step": 33400
+    },
+    {
+      "epoch": 4.58,
+      "learning_rate": 0.0008757129081140653,
+      "loss": 0.3622,
+      "step": 33410
+    },
+    {
+      "epoch": 4.58,
+      "learning_rate": 0.0008751618680258989,
+      "loss": 0.2476,
+      "step": 33420
+    },
+    {
+      "epoch": 4.58,
+      "learning_rate": 0.0008746108279377325,
+      "loss": 0.253,
+      "step": 33430
+    },
+    {
+      "epoch": 4.58,
+      "learning_rate": 0.0008740597878495661,
+      "loss": 0.2638,
+      "step": 33440
+    },
+    {
+      "epoch": 4.58,
+      "learning_rate": 0.0008735087477613996,
+      "loss": 0.2355,
+      "step": 33450
+    },
+    {
+      "epoch": 4.58,
+      "learning_rate": 0.0008729577076732332,
+      "loss": 0.2861,
+      "step": 33460
+    },
+    {
+      "epoch": 4.59,
+      "learning_rate": 0.0008724066675850668,
+      "loss": 0.2639,
+      "step": 33470
+    },
+    {
+      "epoch": 4.59,
+      "learning_rate": 0.0008718556274969004,
+      "loss": 0.2637,
+      "step": 33480
+    },
+    {
+      "epoch": 4.59,
+      "learning_rate": 0.000871304587408734,
+      "loss": 0.293,
+      "step": 33490
+    },
+    {
+      "epoch": 4.59,
+      "learning_rate": 0.0008707535473205676,
+      "loss": 0.1969,
+      "step": 33500
+    },
+    {
+      "epoch": 4.59,
+      "eval_loss": 0.1859211027622223,
+      "eval_runtime": 1072.9471,
+      "eval_samples_per_second": 10.889,
+      "eval_steps_per_second": 1.362,
+      "eval_wer": 0.4671040662239607,
+      "step": 33500
+    },
+    {
+      "epoch": 4.59,
+      "learning_rate": 0.0008702025072324011,
+      "loss": 0.2536,
+      "step": 33510
+    },
+    {
+      "epoch": 4.59,
+      "learning_rate": 0.0008696514671442347,
+      "loss": 0.2461,
+      "step": 33520
+    },
+    {
+      "epoch": 4.59,
+      "learning_rate": 0.0008691004270560684,
+      "loss": 0.2601,
+      "step": 33530
+    },
+    {
+      "epoch": 4.6,
+      "learning_rate": 0.000868549386967902,
+      "loss": 0.3025,
+      "step": 33540
+    },
+    {
+      "epoch": 4.6,
+      "learning_rate": 0.0008679983468797356,
+      "loss": 0.2037,
+      "step": 33550
+    },
+    {
+      "epoch": 4.6,
+      "learning_rate": 0.0008674473067915692,
+      "loss": 0.2367,
+      "step": 33560
+    },
+    {
+      "epoch": 4.6,
+      "learning_rate": 0.0008668962667034027,
+      "loss": 0.2666,
+      "step": 33570
+    },
+    {
+      "epoch": 4.6,
+      "learning_rate": 0.0008663452266152363,
+      "loss": 0.2404,
+      "step": 33580
+    },
+    {
+      "epoch": 4.6,
+      "learning_rate": 0.0008657941865270699,
+      "loss": 0.2465,
+      "step": 33590
+    },
+    {
+      "epoch": 4.6,
+      "learning_rate": 0.0008652431464389035,
+      "loss": 0.202,
+      "step": 33600
+    },
+    {
+      "epoch": 4.6,
+      "eval_loss": 0.1861080378293991,
+      "eval_runtime": 1075.2294,
+      "eval_samples_per_second": 10.866,
+      "eval_steps_per_second": 1.359,
+      "eval_wer": 0.4659214954971345,
+      "step": 33600
+    },
+    {
+      "epoch": 4.6,
+      "learning_rate": 0.000864692106350737,
+      "loss": 0.2562,
+      "step": 33610
+    },
+    {
+      "epoch": 4.61,
+      "learning_rate": 0.0008641410662625706,
+      "loss": 0.2569,
+      "step": 33620
+    },
+    {
+      "epoch": 4.61,
+      "learning_rate": 0.0008635900261744041,
+      "loss": 0.2538,
+      "step": 33630
+    },
+    {
+      "epoch": 4.61,
+      "learning_rate": 0.0008630389860862377,
+      "loss": 0.3088,
+      "step": 33640
+    },
+    {
+      "epoch": 4.61,
+      "learning_rate": 0.0008624879459980713,
+      "loss": 0.2357,
+      "step": 33650
+    },
+    {
+      "epoch": 4.61,
+      "learning_rate": 0.0008619369059099049,
+      "loss": 0.2558,
+      "step": 33660
+    },
+    {
+      "epoch": 4.61,
+      "learning_rate": 0.0008613858658217385,
+      "loss": 0.2594,
+      "step": 33670
+    },
+    {
+      "epoch": 4.61,
+      "learning_rate": 0.0008608348257335721,
+      "loss": 0.232,
+      "step": 33680
+    },
+    {
+      "epoch": 4.62,
+      "learning_rate": 0.0008602837856454058,
+      "loss": 0.2697,
+      "step": 33690
+    },
+    {
+      "epoch": 4.62,
+      "learning_rate": 0.0008597327455572393,
+      "loss": 0.206,
+      "step": 33700
+    },
+    {
+      "epoch": 4.62,
+      "eval_loss": 0.1823737770318985,
+      "eval_runtime": 1074.9808,
+      "eval_samples_per_second": 10.868,
+      "eval_steps_per_second": 1.359,
+      "eval_wer": 0.45766624215409807,
+      "step": 33700
+    },
+    {
+      "epoch": 4.62,
+      "learning_rate": 0.0008591817054690729,
+      "loss": 0.2592,
+      "step": 33710
+    },
+    {
+      "epoch": 4.62,
+      "learning_rate": 0.0008586306653809065,
+      "loss": 0.287,
+      "step": 33720
+    },
+    {
+      "epoch": 4.62,
+      "learning_rate": 0.0008580796252927401,
+      "loss": 0.2514,
+      "step": 33730
+    },
+    {
+      "epoch": 4.62,
+      "learning_rate": 0.0008575285852045737,
+      "loss": 0.3138,
+      "step": 33740
+    },
+    {
+      "epoch": 4.62,
+      "learning_rate": 0.0008569775451164073,
+      "loss": 0.2281,
+      "step": 33750
+    },
+    {
+      "epoch": 4.63,
+      "learning_rate": 0.0008564265050282408,
+      "loss": 0.2645,
+      "step": 33760
+    },
+    {
+      "epoch": 4.63,
+      "learning_rate": 0.0008558754649400744,
+      "loss": 0.261,
+      "step": 33770
+    },
+    {
+      "epoch": 4.63,
+      "learning_rate": 0.000855324424851908,
+      "loss": 0.2638,
+      "step": 33780
+    },
+    {
+      "epoch": 4.63,
+      "learning_rate": 0.0008547733847637416,
+      "loss": 0.2734,
+      "step": 33790
+    },
+    {
+      "epoch": 4.63,
+      "learning_rate": 0.0008542223446755752,
+      "loss": 0.3113,
+      "step": 33800
+    },
+    {
+      "epoch": 4.63,
+      "eval_loss": 0.19135889410972595,
+      "eval_runtime": 1074.8695,
+      "eval_samples_per_second": 10.869,
+      "eval_steps_per_second": 1.359,
+      "eval_wer": 0.48792413353952513,
+      "step": 33800
+    },
+    {
+      "epoch": 4.63,
+      "learning_rate": 0.0008536713045874088,
+      "loss": 0.2827,
+      "step": 33810
+    },
+    {
+      "epoch": 4.63,
+      "learning_rate": 0.0008531202644992422,
+      "loss": 0.2793,
+      "step": 33820
+    },
+    {
+      "epoch": 4.63,
+      "learning_rate": 0.0008525692244110758,
+      "loss": 0.2686,
+      "step": 33830
+    },
+    {
+      "epoch": 4.64,
+      "learning_rate": 0.0008520181843229095,
+      "loss": 0.2876,
+      "step": 33840
+    },
+    {
+      "epoch": 4.64,
+      "learning_rate": 0.0008514671442347431,
+      "loss": 0.2192,
+      "step": 33850
+    },
+    {
+      "epoch": 4.64,
+      "learning_rate": 0.0008509161041465767,
+      "loss": 0.2721,
+      "step": 33860
+    },
+    {
+      "epoch": 4.64,
+      "learning_rate": 0.0008503650640584103,
+      "loss": 0.2738,
+      "step": 33870
+    },
+    {
+      "epoch": 4.64,
+      "learning_rate": 0.0008498140239702439,
+      "loss": 0.2674,
+      "step": 33880
+    },
+    {
+      "epoch": 4.64,
+      "learning_rate": 0.0008492629838820774,
+      "loss": 0.2833,
+      "step": 33890
+    },
+    {
+      "epoch": 4.64,
+      "learning_rate": 0.000848711943793911,
+      "loss": 0.2779,
+      "step": 33900
+    },
+    {
+      "epoch": 4.64,
+      "eval_loss": 0.19409558176994324,
+      "eval_runtime": 1076.1452,
+      "eval_samples_per_second": 10.856,
+      "eval_steps_per_second": 1.358,
+      "eval_wer": 0.4746088419903575,
+      "step": 33900
+    },
+    {
+      "epoch": 4.65,
+      "learning_rate": 0.0008481609037057446,
+      "loss": 0.2621,
+      "step": 33910
+    },
+    {
+      "epoch": 4.65,
+      "learning_rate": 0.0008476098636175782,
+      "loss": 0.2687,
+      "step": 33920
+    },
+    {
+      "epoch": 4.65,
+      "learning_rate": 0.0008470588235294118,
+      "loss": 0.2461,
+      "step": 33930
+    },
+    {
+      "epoch": 4.65,
+      "learning_rate": 0.0008465077834412454,
+      "loss": 0.2911,
+      "step": 33940
+    },
+    {
+      "epoch": 4.65,
+      "learning_rate": 0.0008459567433530789,
+      "loss": 0.2458,
+      "step": 33950
+    },
+    {
+      "epoch": 4.65,
+      "learning_rate": 0.0008454057032649125,
+      "loss": 0.2479,
+      "step": 33960
+    },
+    {
+      "epoch": 4.65,
+      "learning_rate": 0.0008448546631767461,
+      "loss": 0.2561,
+      "step": 33970
+    },
+    {
+      "epoch": 4.66,
+      "learning_rate": 0.0008443036230885797,
+      "loss": 0.247,
+      "step": 33980
+    },
+    {
+      "epoch": 4.66,
+      "learning_rate": 0.0008437525830004133,
+      "loss": 0.3028,
+      "step": 33990
+    },
+    {
+      "epoch": 4.66,
+      "learning_rate": 0.000843201542912247,
+      "loss": 0.2115,
+      "step": 34000
+    },
+    {
+      "epoch": 4.66,
+      "eval_loss": 0.18494321405887604,
+      "eval_runtime": 1077.9726,
+      "eval_samples_per_second": 10.838,
+      "eval_steps_per_second": 1.355,
+      "eval_wer": 0.46112298735558993,
+      "step": 34000
+    },
+    {
+      "epoch": 4.66,
+      "learning_rate": 0.0008426505028240805,
+      "loss": 0.3009,
+      "step": 34010
+    },
+    {
+      "epoch": 4.66,
+      "learning_rate": 0.0008420994627359141,
+      "loss": 0.2225,
+      "step": 34020
+    },
+    {
+      "epoch": 4.66,
+      "learning_rate": 0.0008415484226477477,
+      "loss": 0.2258,
+      "step": 34030
+    },
+    {
+      "epoch": 4.66,
+      "learning_rate": 0.0008409973825595812,
+      "loss": 0.2694,
+      "step": 34040
+    },
+    {
+      "epoch": 4.67,
+      "learning_rate": 0.0008404463424714148,
+      "loss": 0.2171,
+      "step": 34050
+    },
+    {
+      "epoch": 4.67,
+      "learning_rate": 0.0008398953023832484,
+      "loss": 0.2449,
+      "step": 34060
+    },
+    {
+      "epoch": 4.67,
+      "learning_rate": 0.000839344262295082,
+      "loss": 0.2374,
+      "step": 34070
+    },
+    {
+      "epoch": 4.67,
+      "learning_rate": 0.0008387932222069155,
+      "loss": 0.2854,
+      "step": 34080
+    },
+    {
+      "epoch": 4.67,
+      "learning_rate": 0.0008382421821187491,
+      "loss": 0.2536,
+      "step": 34090
+    },
+    {
+      "epoch": 4.67,
+      "learning_rate": 0.0008376911420305827,
+      "loss": 0.1859,
+      "step": 34100
+    },
+    {
+      "epoch": 4.67,
+      "eval_loss": 0.18722227215766907,
+      "eval_runtime": 1075.9188,
+      "eval_samples_per_second": 10.859,
+      "eval_steps_per_second": 1.358,
+      "eval_wer": 0.46767260984262715,
+      "step": 34100
+    },
+    {
+      "epoch": 4.67,
+      "learning_rate": 0.0008371401019424163,
+      "loss": 0.2743,
+      "step": 34110
+    },
+    {
+      "epoch": 4.67,
+      "learning_rate": 0.0008365890618542499,
+      "loss": 0.2782,
+      "step": 34120
+    },
+    {
+      "epoch": 4.68,
+      "learning_rate": 0.0008360380217660835,
+      "loss": 0.2605,
+      "step": 34130
+    },
+    {
+      "epoch": 4.68,
+      "learning_rate": 0.000835486981677917,
+      "loss": 0.2481,
+      "step": 34140
+    },
+    {
+      "epoch": 4.68,
+      "learning_rate": 0.0008349359415897507,
+      "loss": 0.2133,
+      "step": 34150
+    },
+    {
+      "epoch": 4.68,
+      "learning_rate": 0.0008343849015015843,
+      "loss": 0.2221,
+      "step": 34160
+    },
+    {
+      "epoch": 4.68,
+      "learning_rate": 0.0008338338614134179,
+      "loss": 0.2546,
+      "step": 34170
+    },
+    {
+      "epoch": 4.68,
+      "learning_rate": 0.0008332828213252515,
+      "loss": 0.2797,
+      "step": 34180
+    },
+    {
+      "epoch": 4.68,
+      "learning_rate": 0.0008327317812370851,
+      "loss": 0.2159,
+      "step": 34190
+    },
+    {
+      "epoch": 4.69,
+      "learning_rate": 0.0008321807411489186,
+      "loss": 0.2078,
+      "step": 34200
+    },
+    {
+      "epoch": 4.69,
+      "eval_loss": 0.18041867017745972,
+      "eval_runtime": 1074.7364,
+      "eval_samples_per_second": 10.871,
+      "eval_steps_per_second": 1.359,
+      "eval_wer": 0.4623965250614027,
+      "step": 34200
+    },
+    {
+      "epoch": 4.69,
+      "learning_rate": 0.0008316297010607522,
+      "loss": 0.2421,
+      "step": 34210
+    },
+    {
+      "epoch": 4.69,
+      "learning_rate": 0.0008310786609725858,
+      "loss": 0.2832,
+      "step": 34220
+    },
+    {
+      "epoch": 4.69,
+      "learning_rate": 0.0008305276208844194,
+      "loss": 0.2662,
+      "step": 34230
+    },
+    {
+      "epoch": 4.69,
+      "learning_rate": 0.000829976580796253,
+      "loss": 0.305,
+      "step": 34240
+    },
+    {
+      "epoch": 4.69,
+      "learning_rate": 0.0008294255407080865,
+      "loss": 0.1778,
+      "step": 34250
+    },
+    {
+      "epoch": 4.69,
+      "learning_rate": 0.0008288745006199201,
+      "loss": 0.256,
+      "step": 34260
+    },
+    {
+      "epoch": 4.7,
+      "learning_rate": 0.0008283234605317536,
+      "loss": 0.2916,
+      "step": 34270
+    },
+    {
+      "epoch": 4.7,
+      "learning_rate": 0.0008277724204435872,
+      "loss": 0.2567,
+      "step": 34280
+    },
+    {
+      "epoch": 4.7,
+      "learning_rate": 0.0008272213803554208,
+      "loss": 0.2227,
+      "step": 34290
+    },
+    {
+      "epoch": 4.7,
+      "learning_rate": 0.0008266703402672544,
+      "loss": 0.1957,
+      "step": 34300
+    },
+    {
+      "epoch": 4.7,
+      "eval_loss": 0.18803822994232178,
+      "eval_runtime": 1075.6337,
+      "eval_samples_per_second": 10.862,
+      "eval_steps_per_second": 1.358,
+      "eval_wer": 0.46529609751660145,
+      "step": 34300
+    },
+    {
+      "epoch": 4.7,
+      "learning_rate": 0.0008261193001790881,
+      "loss": 0.2516,
+      "step": 34310
+    },
+    {
+      "epoch": 4.7,
+      "learning_rate": 0.0008255682600909217,
+      "loss": 0.2469,
+      "step": 34320
+    },
+    {
+      "epoch": 4.7,
+      "learning_rate": 0.0008250172200027552,
+      "loss": 0.2644,
+      "step": 34330
+    },
+    {
+      "epoch": 4.7,
+      "learning_rate": 0.0008244661799145888,
+      "loss": 0.2658,
+      "step": 34340
+    },
+    {
+      "epoch": 4.71,
+      "learning_rate": 0.0008239151398264224,
+      "loss": 0.195,
+      "step": 34350
+    },
+    {
+      "epoch": 4.71,
+      "learning_rate": 0.000823364099738256,
+      "loss": 0.2328,
+      "step": 34360
+    },
+    {
+      "epoch": 4.71,
+      "learning_rate": 0.0008228130596500896,
+      "loss": 0.2341,
+      "step": 34370
+    },
+    {
+      "epoch": 4.71,
+      "learning_rate": 0.0008222620195619232,
+      "loss": 0.2721,
+      "step": 34380
+    },
+    {
+      "epoch": 4.71,
+      "learning_rate": 0.0008217109794737567,
+      "loss": 0.2694,
+      "step": 34390
+    },
+    {
+      "epoch": 4.71,
+      "learning_rate": 0.0008211599393855903,
+      "loss": 0.2222,
+      "step": 34400
+    },
+    {
+      "epoch": 4.71,
+      "eval_loss": 0.17720100283622742,
+      "eval_runtime": 1076.16,
+      "eval_samples_per_second": 10.856,
+      "eval_steps_per_second": 1.358,
+      "eval_wer": 0.45361821158919313,
+      "step": 34400
+    },
+    {
+      "epoch": 4.71,
+      "learning_rate": 0.0008206088992974239,
+      "loss": 0.2566,
+      "step": 34410
+    },
+    {
+      "epoch": 4.72,
+      "learning_rate": 0.0008200578592092575,
+      "loss": 0.2499,
+      "step": 34420
+    },
+    {
+      "epoch": 4.72,
+      "learning_rate": 0.0008195068191210911,
+      "loss": 0.2621,
+      "step": 34430
+    },
+    {
+      "epoch": 4.72,
+      "learning_rate": 0.0008189557790329247,
+      "loss": 0.2641,
+      "step": 34440
+    },
+    {
+      "epoch": 4.72,
+      "learning_rate": 0.0008184047389447581,
+      "loss": 0.1874,
+      "step": 34450
+    },
+    {
+      "epoch": 4.72,
+      "learning_rate": 0.0008178536988565919,
+      "loss": 0.2411,
+      "step": 34460
+    },
+    {
+      "epoch": 4.72,
+      "learning_rate": 0.0008173026587684254,
+      "loss": 0.258,
+      "step": 34470
+    },
+    {
+      "epoch": 4.72,
+      "learning_rate": 0.000816751618680259,
+      "loss": 0.2409,
+      "step": 34480
+    },
+    {
+      "epoch": 4.73,
+      "learning_rate": 0.0008162005785920926,
+      "loss": 0.2226,
+      "step": 34490
+    },
+    {
+      "epoch": 4.73,
+      "learning_rate": 0.0008156495385039262,
+      "loss": 0.2334,
+      "step": 34500
+    },
+    {
+      "epoch": 4.73,
+      "eval_loss": 0.1845354586839676,
+      "eval_runtime": 1077.361,
+      "eval_samples_per_second": 10.844,
+      "eval_steps_per_second": 1.356,
+      "eval_wer": 0.4648753752387883,
+      "step": 34500
+    },
+    {
+      "epoch": 4.73,
+      "learning_rate": 0.0008150984984157598,
+      "loss": 0.2558,
+      "step": 34510
+    },
+    {
+      "epoch": 4.73,
+      "learning_rate": 0.0008145474583275933,
+      "loss": 0.2533,
+      "step": 34520
+    },
+    {
+      "epoch": 4.73,
+      "learning_rate": 0.0008139964182394269,
+      "loss": 0.2615,
+      "step": 34530
+    },
+    {
+      "epoch": 4.73,
+      "learning_rate": 0.0008134453781512605,
+      "loss": 0.2285,
+      "step": 34540
+    },
+    {
+      "epoch": 4.73,
+      "learning_rate": 0.0008128943380630941,
+      "loss": 0.2017,
+      "step": 34550
+    },
+    {
+      "epoch": 4.73,
+      "learning_rate": 0.0008123432979749277,
+      "loss": 0.274,
+      "step": 34560
+    },
+    {
+      "epoch": 4.74,
+      "learning_rate": 0.0008117922578867613,
+      "loss": 0.2355,
+      "step": 34570
+    },
+    {
+      "epoch": 4.74,
+      "learning_rate": 0.0008112412177985948,
+      "loss": 0.2247,
+      "step": 34580
+    },
+    {
+      "epoch": 4.74,
+      "learning_rate": 0.0008106901777104284,
+      "loss": 0.2536,
+      "step": 34590
+    },
+    {
+      "epoch": 4.74,
+      "learning_rate": 0.000810139137622262,
+      "loss": 0.1754,
+      "step": 34600
+    },
+    {
+      "epoch": 4.74,
+      "eval_loss": 0.17293119430541992,
+      "eval_runtime": 1076.4306,
+      "eval_samples_per_second": 10.853,
+      "eval_steps_per_second": 1.357,
+      "eval_wer": 0.45042299645228784,
+      "step": 34600
+    },
+    {
+      "epoch": 4.74,
+      "learning_rate": 0.0008095880975340956,
+      "loss": 0.2131,
+      "step": 34610
+    },
+    {
+      "epoch": 4.74,
+      "learning_rate": 0.0008090370574459293,
+      "loss": 0.2517,
+      "step": 34620
+    },
+    {
+      "epoch": 4.74,
+      "learning_rate": 0.0008084860173577629,
+      "loss": 0.2462,
+      "step": 34630
+    },
+    {
+      "epoch": 4.75,
+      "learning_rate": 0.0008079349772695964,
+      "loss": 0.2369,
+      "step": 34640
+    },
+    {
+      "epoch": 4.75,
+      "learning_rate": 0.00080738393718143,
+      "loss": 0.2082,
+      "step": 34650
+    },
+    {
+      "epoch": 4.75,
+      "learning_rate": 0.0008068328970932636,
+      "loss": 0.2506,
+      "step": 34660
+    },
+    {
+      "epoch": 4.75,
+      "learning_rate": 0.0008062818570050972,
+      "loss": 0.2861,
+      "step": 34670
+    },
+    {
+      "epoch": 4.75,
+      "learning_rate": 0.0008057308169169307,
+      "loss": 0.2682,
+      "step": 34680
+    },
+    {
+      "epoch": 4.75,
+      "learning_rate": 0.0008051797768287643,
+      "loss": 0.2178,
+      "step": 34690
+    },
+    {
+      "epoch": 4.75,
+      "learning_rate": 0.0008046287367405979,
+      "loss": 0.2025,
+      "step": 34700
+    },
+    {
+      "epoch": 4.75,
+      "eval_loss": 0.17688369750976562,
+      "eval_runtime": 1079.0414,
+      "eval_samples_per_second": 10.827,
+      "eval_steps_per_second": 1.354,
+      "eval_wer": 0.45119621577367414,
+      "step": 34700
+    },
+    {
+      "epoch": 4.76,
+      "learning_rate": 0.0008040776966524314,
+      "loss": 0.2678,
+      "step": 34710
+    },
+    {
+      "epoch": 4.76,
+      "learning_rate": 0.000803526656564265,
+      "loss": 0.2359,
+      "step": 34720
+    },
+    {
+      "epoch": 4.76,
+      "learning_rate": 0.0008029756164760986,
+      "loss": 0.2565,
+      "step": 34730
+    },
+    {
+      "epoch": 4.76,
+      "learning_rate": 0.0008024245763879322,
+      "loss": 0.2542,
+      "step": 34740
+    },
+    {
+      "epoch": 4.76,
+      "learning_rate": 0.0008018735362997658,
+      "loss": 0.2594,
+      "step": 34750
+    },
+    {
+      "epoch": 4.76,
+      "learning_rate": 0.0008013224962115994,
+      "loss": 0.2785,
+      "step": 34760
+    },
+    {
+      "epoch": 4.76,
+      "learning_rate": 0.000800771456123433,
+      "loss": 0.2593,
+      "step": 34770
+    },
+    {
+      "epoch": 4.77,
+      "learning_rate": 0.0008002204160352666,
+      "loss": 0.2261,
+      "step": 34780
+    },
+    {
+      "epoch": 4.77,
+      "learning_rate": 0.0007996693759471002,
+      "loss": 0.2469,
+      "step": 34790
+    },
+    {
+      "epoch": 4.77,
+      "learning_rate": 0.0007991183358589338,
+      "loss": 0.213,
+      "step": 34800
+    },
+    {
+      "epoch": 4.77,
+      "eval_loss": 0.1807287633419037,
+      "eval_runtime": 1078.8226,
+      "eval_samples_per_second": 10.829,
+      "eval_steps_per_second": 1.354,
+      "eval_wer": 0.4550509415082325,
+      "step": 34800
+    },
+    {
+      "epoch": 4.77,
+      "learning_rate": 0.0007985672957707674,
+      "loss": 0.2579,
+      "step": 34810
+    },
+    {
+      "epoch": 4.77,
+      "learning_rate": 0.000798016255682601,
+      "loss": 0.2447,
+      "step": 34820
+    },
+    {
+      "epoch": 4.77,
+      "learning_rate": 0.0007974652155944345,
+      "loss": 0.2393,
+      "step": 34830
+    },
+    {
+      "epoch": 4.77,
+      "learning_rate": 0.0007969141755062681,
+      "loss": 0.2391,
+      "step": 34840
+    },
+    {
+      "epoch": 4.77,
+      "learning_rate": 0.0007963631354181017,
+      "loss": 0.2349,
+      "step": 34850
+    },
+    {
+      "epoch": 4.78,
+      "learning_rate": 0.0007958120953299353,
+      "loss": 0.2394,
+      "step": 34860
+    },
+    {
+      "epoch": 4.78,
+      "learning_rate": 0.0007952610552417689,
+      "loss": 0.3017,
+      "step": 34870
+    },
+    {
+      "epoch": 4.78,
+      "learning_rate": 0.0007947100151536025,
+      "loss": 0.2627,
+      "step": 34880
+    },
+    {
+      "epoch": 4.78,
+      "learning_rate": 0.000794158975065436,
+      "loss": 0.2343,
+      "step": 34890
+    },
+    {
+      "epoch": 4.78,
+      "learning_rate": 0.0007936079349772695,
+      "loss": 0.2255,
+      "step": 34900
+    },
+    {
+      "epoch": 4.78,
+      "eval_loss": 0.17419004440307617,
+      "eval_runtime": 1075.8125,
+      "eval_samples_per_second": 10.86,
+      "eval_steps_per_second": 1.358,
+      "eval_wer": 0.44785317929591556,
+      "step": 34900
+    },
+    {
+      "epoch": 4.78,
+      "learning_rate": 0.0007930568948891031,
+      "loss": 0.2435,
+      "step": 34910
+    },
+    {
+      "epoch": 4.78,
+      "learning_rate": 0.0007925058548009367,
+      "loss": 0.2103,
+      "step": 34920
+    },
+    {
+      "epoch": 4.79,
+      "learning_rate": 0.0007919548147127704,
+      "loss": 0.2349,
+      "step": 34930
+    },
+    {
+      "epoch": 4.79,
+      "learning_rate": 0.000791403774624604,
+      "loss": 0.2514,
+      "step": 34940
+    },
+    {
+      "epoch": 4.79,
+      "learning_rate": 0.0007908527345364376,
+      "loss": 0.2193,
+      "step": 34950
+    },
+    {
+      "epoch": 4.79,
+      "learning_rate": 0.0007903016944482711,
+      "loss": 0.2479,
+      "step": 34960
+    },
+    {
+      "epoch": 4.79,
+      "learning_rate": 0.0007897506543601047,
+      "loss": 0.3231,
+      "step": 34970
+    },
+    {
+      "epoch": 4.79,
+      "learning_rate": 0.0007891996142719383,
+      "loss": 0.2856,
+      "step": 34980
+    },
+    {
+      "epoch": 4.79,
+      "learning_rate": 0.0007886485741837719,
+      "loss": 0.3306,
+      "step": 34990
+    },
+    {
+      "epoch": 4.8,
+      "learning_rate": 0.0007880975340956055,
+      "loss": 0.1698,
+      "step": 35000
+    },
+    {
+      "epoch": 4.8,
+      "eval_loss": 0.18027128279209137,
+      "eval_runtime": 1076.8553,
+      "eval_samples_per_second": 10.849,
+      "eval_steps_per_second": 1.357,
+      "eval_wer": 0.4584394614754844,
+      "step": 35000
+    },
+    {
+      "epoch": 4.8,
+      "learning_rate": 0.0007875464940074391,
+      "loss": 0.2592,
+      "step": 35010
+    },
+    {
+      "epoch": 4.8,
+      "learning_rate": 0.0007869954539192726,
+      "loss": 0.2388,
+      "step": 35020
+    },
+    {
+      "epoch": 4.8,
+      "learning_rate": 0.0007864444138311062,
+      "loss": 0.2443,
+      "step": 35030
+    },
+    {
+      "epoch": 4.8,
+      "learning_rate": 0.0007858933737429398,
+      "loss": 0.2447,
+      "step": 35040
+    },
+    {
+      "epoch": 4.8,
+      "learning_rate": 0.0007853423336547734,
+      "loss": 0.2149,
+      "step": 35050
+    },
+    {
+      "epoch": 4.8,
+      "learning_rate": 0.000784791293566607,
+      "loss": 0.2469,
+      "step": 35060
+    },
+    {
+      "epoch": 4.8,
+      "learning_rate": 0.0007842402534784406,
+      "loss": 0.2432,
+      "step": 35070
+    },
+    {
+      "epoch": 4.81,
+      "learning_rate": 0.0007836892133902743,
+      "loss": 0.2347,
+      "step": 35080
+    },
+    {
+      "epoch": 4.81,
+      "learning_rate": 0.0007831381733021078,
+      "loss": 0.2517,
+      "step": 35090
+    },
+    {
+      "epoch": 4.81,
+      "learning_rate": 0.000782642237222758,
+      "loss": 0.2254,
+      "step": 35100
+    },
+    {
+      "epoch": 4.81,
+      "eval_loss": 0.1773955523967743,
+      "eval_runtime": 1075.7471,
+      "eval_samples_per_second": 10.86,
+      "eval_steps_per_second": 1.358,
+      "eval_wer": 0.45095742745383427,
+      "step": 35100
+    },
+    {
+      "epoch": 4.81,
+      "learning_rate": 0.0007820911971345916,
+      "loss": 0.2428,
+      "step": 35110
+    },
+    {
+      "epoch": 4.81,
+      "learning_rate": 0.0007815401570464251,
+      "loss": 0.2633,
+      "step": 35120
+    },
+    {
+      "epoch": 4.81,
+      "learning_rate": 0.0007809891169582587,
+      "loss": 0.2433,
+      "step": 35130
+    },
+    {
+      "epoch": 4.81,
+      "learning_rate": 0.0007804380768700923,
+      "loss": 0.4021,
+      "step": 35140
+    },
+    {
+      "epoch": 4.82,
+      "learning_rate": 0.0007798870367819259,
+      "loss": 0.2329,
+      "step": 35150
+    },
+    {
+      "epoch": 4.82,
+      "learning_rate": 0.0007793359966937596,
+      "loss": 0.4482,
+      "step": 35160
+    },
+    {
+      "epoch": 4.82,
+      "learning_rate": 0.0007787849566055932,
+      "loss": 0.2283,
+      "step": 35170
+    },
+    {
+      "epoch": 4.82,
+      "learning_rate": 0.0007782339165174267,
+      "loss": 0.2507,
+      "step": 35180
+    },
+    {
+      "epoch": 4.82,
+      "learning_rate": 0.0007776828764292602,
+      "loss": 0.3008,
+      "step": 35190
+    },
+    {
+      "epoch": 4.82,
+      "learning_rate": 0.0007771318363410938,
+      "loss": 0.1913,
+      "step": 35200
+    },
+    {
+      "epoch": 4.82,
+      "eval_loss": 0.17416377365589142,
+      "eval_runtime": 1078.7506,
+      "eval_samples_per_second": 10.83,
+      "eval_steps_per_second": 1.354,
+      "eval_wer": 0.4504457381970345,
+      "step": 35200
+    },
+    {
+      "epoch": 4.82,
+      "learning_rate": 0.0007765807962529274,
+      "loss": 0.2326,
+      "step": 35210
+    },
+    {
+      "epoch": 4.83,
+      "learning_rate": 0.000776029756164761,
+      "loss": 0.2581,
+      "step": 35220
+    },
+    {
+      "epoch": 4.83,
+      "learning_rate": 0.0007754787160765946,
+      "loss": 0.2576,
+      "step": 35230
+    },
+    {
+      "epoch": 4.83,
+      "learning_rate": 0.0007749276759884281,
+      "loss": 0.2404,
+      "step": 35240
+    },
+    {
+      "epoch": 4.83,
+      "learning_rate": 0.0007743766359002617,
+      "loss": 0.2253,
+      "step": 35250
+    },
+    {
+      "epoch": 4.83,
+      "learning_rate": 0.0007738255958120953,
+      "loss": 0.2515,
+      "step": 35260
+    },
+    {
+      "epoch": 4.83,
+      "learning_rate": 0.0007732745557239289,
+      "loss": 0.2445,
+      "step": 35270
+    },
+    {
+      "epoch": 4.83,
+      "learning_rate": 0.0007727235156357625,
+      "loss": 0.229,
+      "step": 35280
+    },
+    {
+      "epoch": 4.83,
+      "learning_rate": 0.0007721724755475961,
+      "loss": 0.2607,
+      "step": 35290
+    },
+    {
+      "epoch": 4.84,
+      "learning_rate": 0.0007716214354594297,
+      "loss": 0.2123,
+      "step": 35300
+    },
+    {
+      "epoch": 4.84,
+      "eval_loss": 0.17729370296001434,
+      "eval_runtime": 1075.6086,
+      "eval_samples_per_second": 10.862,
+      "eval_steps_per_second": 1.358,
+      "eval_wer": 0.4591217138178841,
+      "step": 35300
+    },
+    {
+      "epoch": 4.84,
+      "learning_rate": 0.0007710703953712632,
+      "loss": 0.2676,
+      "step": 35310
+    },
+    {
+      "epoch": 4.84,
+      "learning_rate": 0.0007705193552830969,
+      "loss": 0.2164,
+      "step": 35320
+    },
+    {
+      "epoch": 4.84,
+      "learning_rate": 0.0007699683151949305,
+      "loss": 0.2781,
+      "step": 35330
+    },
+    {
+      "epoch": 4.84,
+      "learning_rate": 0.0007694172751067641,
+      "loss": 0.2259,
+      "step": 35340
+    },
+    {
+      "epoch": 4.84,
+      "learning_rate": 0.0007688662350185977,
+      "loss": 0.2302,
+      "step": 35350
+    },
+    {
+      "epoch": 4.84,
+      "learning_rate": 0.0007683151949304313,
+      "loss": 0.2488,
+      "step": 35360
+    },
+    {
+      "epoch": 4.85,
+      "learning_rate": 0.0007677641548422648,
+      "loss": 0.2637,
+      "step": 35370
+    },
+    {
+      "epoch": 4.85,
+      "learning_rate": 0.0007672131147540984,
+      "loss": 0.278,
+      "step": 35380
+    },
+    {
+      "epoch": 4.85,
+      "learning_rate": 0.000766662074665932,
+      "loss": 0.2443,
+      "step": 35390
+    },
+    {
+      "epoch": 4.85,
+      "learning_rate": 0.0007661110345777655,
+      "loss": 0.2402,
+      "step": 35400
+    },
+    {
+      "epoch": 4.85,
+      "eval_loss": 0.17681477963924408,
+      "eval_runtime": 1076.2385,
+      "eval_samples_per_second": 10.855,
+      "eval_steps_per_second": 1.358,
+      "eval_wer": 0.4526516874374602,
+      "step": 35400
+    },
+    {
+      "epoch": 4.85,
+      "learning_rate": 0.0007655599944895991,
+      "loss": 0.2394,
+      "step": 35410
+    },
+    {
+      "epoch": 4.85,
+      "learning_rate": 0.0007650089544014327,
+      "loss": 0.2668,
+      "step": 35420
+    },
+    {
+      "epoch": 4.85,
+      "learning_rate": 0.0007644579143132662,
+      "loss": 0.2264,
+      "step": 35430
+    },
+    {
+      "epoch": 4.86,
+      "learning_rate": 0.0007639068742250998,
+      "loss": 0.2216,
+      "step": 35440
+    },
+    {
+      "epoch": 4.86,
+      "learning_rate": 0.0007633558341369334,
+      "loss": 0.2096,
+      "step": 35450
+    },
+    {
+      "epoch": 4.86,
+      "learning_rate": 0.000762804794048767,
+      "loss": 0.2367,
+      "step": 35460
+    },
+    {
+      "epoch": 4.86,
+      "learning_rate": 0.0007622537539606007,
+      "loss": 0.245,
+      "step": 35470
+    },
+    {
+      "epoch": 4.86,
+      "learning_rate": 0.0007617027138724343,
+      "loss": 0.2641,
+      "step": 35480
+    },
+    {
+      "epoch": 4.86,
+      "learning_rate": 0.0007611516737842679,
+      "loss": 0.263,
+      "step": 35490
+    },
+    {
+      "epoch": 4.86,
+      "learning_rate": 0.0007606006336961014,
+      "loss": 0.1617,
+      "step": 35500
+    },
+    {
+      "epoch": 4.86,
+      "eval_loss": 0.17165154218673706,
+      "eval_runtime": 1075.8966,
+      "eval_samples_per_second": 10.859,
+      "eval_steps_per_second": 1.358,
+      "eval_wer": 0.4483762394250887,
+      "step": 35500
+    },
+    {
+      "epoch": 4.87,
+      "learning_rate": 0.000760049593607935,
+      "loss": 0.238,
+      "step": 35510
+    },
+    {
+      "epoch": 4.87,
+      "learning_rate": 0.0007594985535197686,
+      "loss": 0.2235,
+      "step": 35520
+    },
+    {
+      "epoch": 4.87,
+      "learning_rate": 0.0007589475134316022,
+      "loss": 0.225,
+      "step": 35530
+    },
+    {
+      "epoch": 4.87,
+      "learning_rate": 0.0007583964733434358,
+      "loss": 0.2573,
+      "step": 35540
+    },
+    {
+      "epoch": 4.87,
+      "learning_rate": 0.0007578454332552694,
+      "loss": 0.2144,
+      "step": 35550
+    },
+    {
+      "epoch": 4.87,
+      "learning_rate": 0.0007572943931671029,
+      "loss": 0.2365,
+      "step": 35560
+    },
+    {
+      "epoch": 4.87,
+      "learning_rate": 0.0007567433530789365,
+      "loss": 0.2345,
+      "step": 35570
+    },
+    {
+      "epoch": 4.87,
+      "learning_rate": 0.0007561923129907701,
+      "loss": 0.2251,
+      "step": 35580
+    },
+    {
+      "epoch": 4.88,
+      "learning_rate": 0.0007556412729026037,
+      "loss": 0.237,
+      "step": 35590
+    },
+    {
+      "epoch": 4.88,
+      "learning_rate": 0.0007550902328144373,
+      "loss": 0.2387,
+      "step": 35600
+    },
+    {
+      "epoch": 4.88,
+      "eval_loss": 0.1713322252035141,
+      "eval_runtime": 1076.1019,
+      "eval_samples_per_second": 10.857,
+      "eval_steps_per_second": 1.358,
+      "eval_wer": 0.4540389338670063,
+      "step": 35600
+    },
+    {
+      "epoch": 4.88,
+      "learning_rate": 0.0007545391927262708,
+      "loss": 0.2418,
+      "step": 35610
+    },
+    {
+      "epoch": 4.88,
+      "learning_rate": 0.0007539881526381043,
+      "loss": 0.2347,
+      "step": 35620
+    },
+    {
+      "epoch": 4.88,
+      "learning_rate": 0.000753437112549938,
+      "loss": 0.2442,
+      "step": 35630
+    },
+    {
+      "epoch": 4.88,
+      "learning_rate": 0.0007528860724617716,
+      "loss": 0.2603,
+      "step": 35640
+    },
+    {
+      "epoch": 4.88,
+      "learning_rate": 0.0007523350323736052,
+      "loss": 0.2088,
+      "step": 35650
+    },
+    {
+      "epoch": 4.89,
+      "learning_rate": 0.0007517839922854388,
+      "loss": 0.2388,
+      "step": 35660
+    },
+    {
+      "epoch": 4.89,
+      "learning_rate": 0.0007512329521972724,
+      "loss": 0.2114,
+      "step": 35670
+    },
+    {
+      "epoch": 4.89,
+      "learning_rate": 0.000750681912109106,
+      "loss": 0.2274,
+      "step": 35680
+    },
+    {
+      "epoch": 4.89,
+      "learning_rate": 0.0007501308720209395,
+      "loss": 0.2166,
+      "step": 35690
+    },
+    {
+      "epoch": 4.89,
+      "learning_rate": 0.0007495798319327731,
+      "loss": 0.1844,
+      "step": 35700
+    },
+    {
+      "epoch": 4.89,
+      "eval_loss": 0.16468322277069092,
+      "eval_runtime": 1076.9302,
+      "eval_samples_per_second": 10.848,
+      "eval_steps_per_second": 1.357,
+      "eval_wer": 0.4385972891840262,
+      "step": 35700
+    },
+    {
+      "epoch": 4.89,
+      "learning_rate": 0.0007490287918446067,
+      "loss": 0.2446,
+      "step": 35710
+    },
+    {
+      "epoch": 4.89,
+      "learning_rate": 0.0007484777517564403,
+      "loss": 0.2565,
+      "step": 35720
+    },
+    {
+      "epoch": 4.9,
+      "learning_rate": 0.0007479267116682739,
+      "loss": 0.2506,
+      "step": 35730
+    },
+    {
+      "epoch": 4.9,
+      "learning_rate": 0.0007473756715801075,
+      "loss": 0.2446,
+      "step": 35740
+    },
+    {
+      "epoch": 4.9,
+      "learning_rate": 0.000746824631491941,
+      "loss": 0.1664,
+      "step": 35750
+    },
+    {
+      "epoch": 4.9,
+      "learning_rate": 0.0007462735914037746,
+      "loss": 0.2449,
+      "step": 35760
+    },
+    {
+      "epoch": 4.9,
+      "learning_rate": 0.0007457225513156082,
+      "loss": 0.263,
+      "step": 35770
+    },
+    {
+      "epoch": 4.9,
+      "learning_rate": 0.0007451715112274419,
+      "loss": 0.2509,
+      "step": 35780
+    },
+    {
+      "epoch": 4.9,
+      "learning_rate": 0.0007446204711392755,
+      "loss": 0.248,
+      "step": 35790
+    },
+    {
+      "epoch": 4.9,
+      "learning_rate": 0.0007440694310511091,
+      "loss": 0.2113,
+      "step": 35800
+    },
+    {
+      "epoch": 4.9,
+      "eval_loss": 0.1760196089744568,
+      "eval_runtime": 1074.8553,
+      "eval_samples_per_second": 10.869,
+      "eval_steps_per_second": 1.359,
+      "eval_wer": 0.45797325570817793,
+      "step": 35800
+    },
+    {
+      "epoch": 4.91,
+      "learning_rate": 0.0007435183909629426,
+      "loss": 0.2739,
+      "step": 35810
+    },
+    {
+      "epoch": 4.91,
+      "learning_rate": 0.0007429673508747762,
+      "loss": 0.2427,
+      "step": 35820
+    },
+    {
+      "epoch": 4.91,
+      "learning_rate": 0.0007424163107866097,
+      "loss": 0.238,
+      "step": 35830
+    },
+    {
+      "epoch": 4.91,
+      "learning_rate": 0.0007418652706984433,
+      "loss": 0.2491,
+      "step": 35840
+    },
+    {
+      "epoch": 4.91,
+      "learning_rate": 0.0007413142306102769,
+      "loss": 0.2266,
+      "step": 35850
+    },
+    {
+      "epoch": 4.91,
+      "learning_rate": 0.0007407631905221105,
+      "loss": 0.2379,
+      "step": 35860
+    },
+    {
+      "epoch": 4.91,
+      "learning_rate": 0.000740212150433944,
+      "loss": 0.2451,
+      "step": 35870
+    },
+    {
+      "epoch": 4.92,
+      "learning_rate": 0.0007396611103457776,
+      "loss": 0.2293,
+      "step": 35880
+    },
+    {
+      "epoch": 4.92,
+      "learning_rate": 0.0007391100702576112,
+      "loss": 0.2429,
+      "step": 35890
+    },
+    {
+      "epoch": 4.92,
+      "learning_rate": 0.0007385590301694448,
+      "loss": 0.1892,
+      "step": 35900
+    },
+    {
+      "epoch": 4.92,
+      "eval_loss": 0.1643328070640564,
+      "eval_runtime": 1076.3974,
+      "eval_samples_per_second": 10.854,
+      "eval_steps_per_second": 1.357,
+      "eval_wer": 0.4391544619303193,
+      "step": 35900
+    },
+    {
+      "epoch": 4.92,
+      "learning_rate": 0.0007380079900812784,
+      "loss": 0.2311,
+      "step": 35910
+    },
+    {
+      "epoch": 4.92,
+      "learning_rate": 0.000737456949993112,
+      "loss": 0.2313,
+      "step": 35920
+    },
+    {
+      "epoch": 4.92,
+      "learning_rate": 0.0007369059099049456,
+      "loss": 0.2966,
+      "step": 35930
+    },
+    {
+      "epoch": 4.92,
+      "learning_rate": 0.0007363548698167792,
+      "loss": 0.257,
+      "step": 35940
+    },
+    {
+      "epoch": 4.93,
+      "learning_rate": 0.0007358038297286128,
+      "loss": 0.2491,
+      "step": 35950
+    },
+    {
+      "epoch": 4.93,
+      "learning_rate": 0.0007352527896404464,
+      "loss": 0.2261,
+      "step": 35960
+    },
+    {
+      "epoch": 4.93,
+      "learning_rate": 0.00073470174955228,
+      "loss": 0.2536,
+      "step": 35970
+    },
+    {
+      "epoch": 4.93,
+      "learning_rate": 0.0007341507094641136,
+      "loss": 0.2492,
+      "step": 35980
+    },
+    {
+      "epoch": 4.93,
+      "learning_rate": 0.0007335996693759472,
+      "loss": 0.2323,
+      "step": 35990
+    },
+    {
+      "epoch": 4.93,
+      "learning_rate": 0.0007330486292877807,
+      "loss": 0.2179,
+      "step": 36000
+    },
+    {
+      "epoch": 4.93,
+      "eval_loss": 0.16508789360523224,
+      "eval_runtime": 1077.0765,
+      "eval_samples_per_second": 10.847,
+      "eval_steps_per_second": 1.356,
+      "eval_wer": 0.44074638406258526,
+      "step": 36000
+    },
+    {
+      "epoch": 4.93,
+      "learning_rate": 0.0007324975891996143,
+      "loss": 0.2474,
+      "step": 36010
+    },
+    {
+      "epoch": 4.93,
+      "learning_rate": 0.0007319465491114479,
+      "loss": 0.2451,
+      "step": 36020
+    },
+    {
+      "epoch": 4.94,
+      "learning_rate": 0.0007313955090232815,
+      "loss": 0.2321,
+      "step": 36030
+    },
+    {
+      "epoch": 4.94,
+      "learning_rate": 0.000730844468935115,
+      "loss": 0.2274,
+      "step": 36040
+    },
+    {
+      "epoch": 4.94,
+      "learning_rate": 0.0007302934288469486,
+      "loss": 0.2106,
+      "step": 36050
+    },
+    {
+      "epoch": 4.94,
+      "learning_rate": 0.0007297423887587821,
+      "loss": 0.2547,
+      "step": 36060
+    },
+    {
+      "epoch": 4.94,
+      "learning_rate": 0.0007291913486706157,
+      "loss": 0.2259,
+      "step": 36070
+    },
+    {
+      "epoch": 4.94,
+      "learning_rate": 0.0007286403085824493,
+      "loss": 0.241,
+      "step": 36080
+    },
+    {
+      "epoch": 4.94,
+      "learning_rate": 0.0007280892684942829,
+      "loss": 0.246,
+      "step": 36090
+    },
+    {
+      "epoch": 4.95,
+      "learning_rate": 0.0007275382284061166,
+      "loss": 0.2053,
+      "step": 36100
+    },
+    {
+      "epoch": 4.95,
+      "eval_loss": 0.16630975902080536,
+      "eval_runtime": 1077.582,
+      "eval_samples_per_second": 10.842,
+      "eval_steps_per_second": 1.356,
+      "eval_wer": 0.44787592104066226,
+      "step": 36100
+    },
+    {
+      "epoch": 4.95,
+      "learning_rate": 0.0007269871883179502,
+      "loss": 0.2261,
+      "step": 36110
+    },
+    {
+      "epoch": 4.95,
+      "learning_rate": 0.0007264361482297838,
+      "loss": 0.2447,
+      "step": 36120
+    },
+    {
+      "epoch": 4.95,
+      "learning_rate": 0.0007258851081416173,
+      "loss": 0.2613,
+      "step": 36130
+    },
+    {
+      "epoch": 4.95,
+      "learning_rate": 0.0007253340680534509,
+      "loss": 0.2357,
+      "step": 36140
+    },
+    {
+      "epoch": 4.95,
+      "learning_rate": 0.0007247830279652845,
+      "loss": 0.2023,
+      "step": 36150
+    },
+    {
+      "epoch": 4.95,
+      "learning_rate": 0.0007242319878771181,
+      "loss": 0.2315,
+      "step": 36160
+    },
+    {
+      "epoch": 4.96,
+      "learning_rate": 0.0007236809477889517,
+      "loss": 0.2955,
+      "step": 36170
+    },
+    {
+      "epoch": 4.96,
+      "learning_rate": 0.0007231299077007853,
+      "loss": 0.2622,
+      "step": 36180
+    },
+    {
+      "epoch": 4.96,
+      "learning_rate": 0.0007225788676126188,
+      "loss": 0.2601,
+      "step": 36190
+    },
+    {
+      "epoch": 4.96,
+      "learning_rate": 0.0007220278275244524,
+      "loss": 0.1748,
+      "step": 36200
+    },
+    {
+      "epoch": 4.96,
+      "eval_loss": 0.17102374136447906,
+      "eval_runtime": 1075.4222,
+      "eval_samples_per_second": 10.864,
+      "eval_steps_per_second": 1.359,
+      "eval_wer": 0.4454311834803966,
+      "step": 36200
+    },
+    {
+      "epoch": 4.96,
+      "learning_rate": 0.000721476787436286,
+      "loss": 0.2384,
+      "step": 36210
+    },
+    {
+      "epoch": 4.96,
+      "learning_rate": 0.0007209257473481196,
+      "loss": 0.2349,
+      "step": 36220
+    },
+    {
+      "epoch": 4.96,
+      "learning_rate": 0.0007203747072599532,
+      "loss": 0.2295,
+      "step": 36230
+    },
+    {
+      "epoch": 4.97,
+      "learning_rate": 0.0007198236671717868,
+      "loss": 0.2338,
+      "step": 36240
+    },
+    {
+      "epoch": 4.97,
+      "learning_rate": 0.0007192726270836203,
+      "loss": 0.1865,
+      "step": 36250
+    },
+    {
+      "epoch": 4.97,
+      "learning_rate": 0.0007187215869954539,
+      "loss": 0.2266,
+      "step": 36260
+    },
+    {
+      "epoch": 4.97,
+      "learning_rate": 0.0007181705469072875,
+      "loss": 0.2301,
+      "step": 36270
+    },
+    {
+      "epoch": 4.97,
+      "learning_rate": 0.0007176195068191211,
+      "loss": 0.2052,
+      "step": 36280
+    },
+    {
+      "epoch": 4.97,
+      "learning_rate": 0.0007170684667309547,
+      "loss": 0.2526,
+      "step": 36290
+    },
+    {
+      "epoch": 4.97,
+      "learning_rate": 0.0007165174266427883,
+      "loss": 0.2233,
+      "step": 36300
+    },
+    {
+      "epoch": 4.97,
+      "eval_loss": 0.1752770096063614,
+      "eval_runtime": 1078.2407,
+      "eval_samples_per_second": 10.835,
+      "eval_steps_per_second": 1.355,
+      "eval_wer": 0.4540616756117529,
+      "step": 36300
+    },
+    {
+      "epoch": 4.97,
+      "learning_rate": 0.0007159663865546219,
+      "loss": 0.2228,
+      "step": 36310
+    },
+    {
+      "epoch": 4.98,
+      "learning_rate": 0.0007154153464664554,
+      "loss": 0.2152,
+      "step": 36320
+    },
+    {
+      "epoch": 4.98,
+      "learning_rate": 0.000714864306378289,
+      "loss": 0.2536,
+      "step": 36330
+    },
+    {
+      "epoch": 4.98,
+      "learning_rate": 0.0007143132662901226,
+      "loss": 0.2395,
+      "step": 36340
+    },
+    {
+      "epoch": 4.98,
+      "learning_rate": 0.0007137622262019562,
+      "loss": 0.1967,
+      "step": 36350
+    },
+    {
+      "epoch": 4.98,
+      "learning_rate": 0.0007132111861137898,
+      "loss": 0.2523,
+      "step": 36360
+    },
+    {
+      "epoch": 4.98,
+      "learning_rate": 0.0007126601460256234,
+      "loss": 0.2174,
+      "step": 36370
+    },
+    {
+      "epoch": 4.98,
+      "learning_rate": 0.0007121091059374569,
+      "loss": 0.2543,
+      "step": 36380
+    },
+    {
+      "epoch": 4.99,
+      "learning_rate": 0.0007115580658492905,
+      "loss": 0.2643,
+      "step": 36390
+    },
+    {
+      "epoch": 4.99,
+      "learning_rate": 0.0007110070257611241,
+      "loss": 0.1941,
+      "step": 36400
+    },
+    {
+      "epoch": 4.99,
+      "eval_loss": 0.16360482573509216,
+      "eval_runtime": 1078.9073,
+      "eval_samples_per_second": 10.829,
+      "eval_steps_per_second": 1.354,
+      "eval_wer": 0.4438051487310106,
+      "step": 36400
+    },
+    {
+      "epoch": 19.96,
+      "learning_rate": 0.0008277900220396714,
+      "loss": 8.6587,
+      "step": 36410
+    },
+    {
+      "epoch": 19.97,
+      "learning_rate": 0.0008272891204167502,
+      "loss": 3.1925,
+      "step": 36420
+    },
+    {
+      "epoch": 19.97,
+      "learning_rate": 0.0008267882187938289,
+      "loss": 3.1791,
+      "step": 36430
+    },
+    {
+      "epoch": 19.98,
+      "learning_rate": 0.0008262873171709077,
+      "loss": 3.1428,
+      "step": 36440
+    },
+    {
+      "epoch": 19.98,
+      "learning_rate": 0.0008257864155479864,
+      "loss": 3.2217,
+      "step": 36450
+    },
+    {
+      "epoch": 19.99,
+      "learning_rate": 0.0008252855139250651,
+      "loss": 3.1955,
+      "step": 36460
+    },
+    {
+      "epoch": 19.99,
+      "learning_rate": 0.0008247846123021438,
+      "loss": 3.0869,
+      "step": 36470
+    },
+    {
+      "epoch": 20.0,
+      "learning_rate": 0.0008242837106792226,
+      "loss": 3.0832,
+      "step": 36480
+    },
+    {
+      "epoch": 20.01,
+      "learning_rate": 0.0008237828090563013,
+      "loss": 3.0618,
+      "step": 36490
+    },
+    {
+      "epoch": 20.01,
+      "learning_rate": 0.0008232819074333801,
+      "loss": 3.0246,
+      "step": 36500
+    },
+    {
+      "epoch": 20.01,
+      "eval_loss": 3.0311832427978516,
+      "eval_runtime": 1157.1176,
+      "eval_samples_per_second": 10.097,
+      "eval_steps_per_second": 0.632,
+      "eval_wer": 1.0,
+      "step": 36500
+    },
+    {
+      "epoch": 20.02,
+      "learning_rate": 0.0008227810058104589,
+      "loss": 3.033,
+      "step": 36510
+    },
+    {
+      "epoch": 20.02,
+      "learning_rate": 0.0008222801041875376,
+      "loss": 3.038,
+      "step": 36520
+    },
+    {
+      "epoch": 20.03,
+      "learning_rate": 0.0008217792025646164,
+      "loss": 3.0226,
+      "step": 36530
+    },
+    {
+      "epoch": 20.03,
+      "learning_rate": 0.0008212783009416951,
+      "loss": 3.0238,
+      "step": 36540
+    },
+    {
+      "epoch": 20.04,
+      "learning_rate": 0.0008207773993187739,
+      "loss": 2.9855,
+      "step": 36550
+    },
+    {
+      "epoch": 20.04,
+      "learning_rate": 0.0008202764976958525,
+      "loss": 2.9475,
+      "step": 36560
+    },
+    {
+      "epoch": 20.05,
+      "learning_rate": 0.0008197755960729313,
+      "loss": 2.8934,
+      "step": 36570
+    },
+    {
+      "epoch": 20.05,
+      "learning_rate": 0.00081927469445001,
+      "loss": 2.7133,
+      "step": 36580
+    },
+    {
+      "epoch": 20.06,
+      "learning_rate": 0.0008187737928270888,
+      "loss": 2.4602,
+      "step": 36590
+    },
+    {
+      "epoch": 20.07,
+      "learning_rate": 0.0008182728912041675,
+      "loss": 1.9537,
+      "step": 36600
+    },
+    {
+      "epoch": 20.07,
+      "eval_loss": 1.4966574907302856,
+      "eval_runtime": 1145.6815,
+      "eval_samples_per_second": 10.197,
+      "eval_steps_per_second": 0.638,
+      "eval_wer": 0.9955767306467752,
+      "step": 36600
+    },
+    {
+      "epoch": 20.07,
+      "learning_rate": 0.0008177719895812462,
+      "loss": 1.379,
+      "step": 36610
+    },
+    {
+      "epoch": 20.08,
+      "learning_rate": 0.000817271087958325,
+      "loss": 1.0349,
+      "step": 36620
+    },
+    {
+      "epoch": 20.08,
+      "learning_rate": 0.0008167701863354038,
+      "loss": 0.8145,
+      "step": 36630
+    },
+    {
+      "epoch": 20.09,
+      "learning_rate": 0.0008162692847124825,
+      "loss": 0.6683,
+      "step": 36640
+    },
+    {
+      "epoch": 20.09,
+      "learning_rate": 0.0008157683830895613,
+      "loss": 0.5354,
+      "step": 36650
+    },
+    {
+      "epoch": 20.1,
+      "learning_rate": 0.0008152674814666399,
+      "loss": 0.4564,
+      "step": 36660
+    },
+    {
+      "epoch": 20.1,
+      "learning_rate": 0.0008147665798437187,
+      "loss": 0.3889,
+      "step": 36670
+    },
+    {
+      "epoch": 20.11,
+      "learning_rate": 0.0008142656782207974,
+      "loss": 0.3294,
+      "step": 36680
+    },
+    {
+      "epoch": 20.11,
+      "learning_rate": 0.0008137647765978762,
+      "loss": 0.4073,
+      "step": 36690
+    },
+    {
+      "epoch": 20.12,
+      "learning_rate": 0.000813263874974955,
+      "loss": 0.3565,
+      "step": 36700
+    },
+    {
+      "epoch": 20.12,
+      "eval_loss": 0.22850795090198517,
+      "eval_runtime": 1146.6746,
+      "eval_samples_per_second": 10.189,
+      "eval_steps_per_second": 0.637,
+      "eval_wer": 0.3911011552806331,
+      "step": 36700
+    },
+    {
+      "epoch": 20.13,
+      "learning_rate": 0.0008127629733520336,
+      "loss": 0.3015,
+      "step": 36710
+    },
+    {
+      "epoch": 20.13,
+      "learning_rate": 0.0008122620717291124,
+      "loss": 0.2986,
+      "step": 36720
+    },
+    {
+      "epoch": 20.14,
+      "learning_rate": 0.0008117611701061912,
+      "loss": 0.2637,
+      "step": 36730
+    },
+    {
+      "epoch": 20.14,
+      "learning_rate": 0.00081126026848327,
+      "loss": 0.2596,
+      "step": 36740
+    },
+    {
+      "epoch": 20.15,
+      "learning_rate": 0.0008107593668603487,
+      "loss": 0.27,
+      "step": 36750
+    },
+    {
+      "epoch": 20.15,
+      "learning_rate": 0.0008102584652374274,
+      "loss": 0.2689,
+      "step": 36760
+    },
+    {
+      "epoch": 20.16,
+      "learning_rate": 0.0008097575636145061,
+      "loss": 0.2568,
+      "step": 36770
+    },
+    {
+      "epoch": 20.16,
+      "learning_rate": 0.0008092566619915849,
+      "loss": 0.2198,
+      "step": 36780
+    },
+    {
+      "epoch": 20.17,
+      "learning_rate": 0.0008087557603686636,
+      "loss": 0.2771,
+      "step": 36790
+    },
+    {
+      "epoch": 20.17,
+      "learning_rate": 0.0008082548587457424,
+      "loss": 0.2553,
+      "step": 36800
+    },
+    {
+      "epoch": 20.17,
+      "eval_loss": 0.17408448457717896,
+      "eval_runtime": 1144.2309,
+      "eval_samples_per_second": 10.21,
+      "eval_steps_per_second": 0.639,
+      "eval_wer": 0.34322978258892023,
+      "step": 36800
+    },
+    {
+      "epoch": 20.18,
+      "learning_rate": 0.000807753957122821,
+      "loss": 0.2594,
+      "step": 36810
+    },
+    {
+      "epoch": 20.19,
+      "learning_rate": 0.0008072530554998998,
+      "loss": 0.2426,
+      "step": 36820
+    },
+    {
+      "epoch": 20.19,
+      "learning_rate": 0.0008067521538769785,
+      "loss": 0.2176,
+      "step": 36830
+    },
+    {
+      "epoch": 20.2,
+      "learning_rate": 0.0008062512522540573,
+      "loss": 0.2459,
+      "step": 36840
+    },
+    {
+      "epoch": 20.2,
+      "learning_rate": 0.0008057503506311361,
+      "loss": 0.2333,
+      "step": 36850
+    },
+    {
+      "epoch": 20.21,
+      "learning_rate": 0.0008052494490082148,
+      "loss": 0.2362,
+      "step": 36860
+    },
+    {
+      "epoch": 20.21,
+      "learning_rate": 0.0008047485473852935,
+      "loss": 0.237,
+      "step": 36870
+    },
+    {
+      "epoch": 20.22,
+      "learning_rate": 0.0008042476457623723,
+      "loss": 0.22,
+      "step": 36880
+    },
+    {
+      "epoch": 20.22,
+      "learning_rate": 0.000803746744139451,
+      "loss": 0.2175,
+      "step": 36890
+    },
+    {
+      "epoch": 20.23,
+      "learning_rate": 0.0008032458425165298,
+      "loss": 0.2412,
+      "step": 36900
+    },
+    {
+      "epoch": 20.23,
+      "eval_loss": 0.15863750874996185,
+      "eval_runtime": 1149.7732,
+      "eval_samples_per_second": 10.161,
+      "eval_steps_per_second": 0.636,
+      "eval_wer": 0.32414945874647505,
+      "step": 36900
+    },
+    {
+      "epoch": 20.24,
+      "learning_rate": 0.0008027449408936085,
+      "loss": 0.2285,
+      "step": 36910
+    },
+    {
+      "epoch": 20.24,
+      "learning_rate": 0.0008022440392706872,
+      "loss": 0.3659,
+      "step": 36920
+    },
+    {
+      "epoch": 20.25,
+      "learning_rate": 0.000801743137647766,
+      "loss": 0.2081,
+      "step": 36930
+    },
+    {
+      "epoch": 20.25,
+      "learning_rate": 0.0008012422360248447,
+      "loss": 0.2351,
+      "step": 36940
+    },
+    {
+      "epoch": 20.26,
+      "learning_rate": 0.0008007413344019235,
+      "loss": 0.2341,
+      "step": 36950
+    },
+    {
+      "epoch": 20.26,
+      "learning_rate": 0.0008002404327790022,
+      "loss": 0.2183,
+      "step": 36960
+    },
+    {
+      "epoch": 20.27,
+      "learning_rate": 0.000799739531156081,
+      "loss": 0.2317,
+      "step": 36970
+    },
+    {
+      "epoch": 20.27,
+      "learning_rate": 0.0007992386295331597,
+      "loss": 0.2069,
+      "step": 36980
+    },
+    {
+      "epoch": 20.28,
+      "learning_rate": 0.0007987377279102385,
+      "loss": 0.2086,
+      "step": 36990
+    },
+    {
+      "epoch": 20.28,
+      "learning_rate": 0.0007982368262873172,
+      "loss": 0.2265,
+      "step": 37000
+    },
+    {
+      "epoch": 20.28,
+      "eval_loss": 0.15294358134269714,
+      "eval_runtime": 1139.9472,
+      "eval_samples_per_second": 10.249,
+      "eval_steps_per_second": 0.641,
+      "eval_wer": 0.3040912398799236,
+      "step": 37000
+    },
+    {
+      "epoch": 20.29,
+      "learning_rate": 0.0007977359246643959,
+      "loss": 0.2067,
+      "step": 37010
+    },
+    {
+      "epoch": 20.3,
+      "learning_rate": 0.0007972350230414746,
+      "loss": 0.2325,
+      "step": 37020
+    },
+    {
+      "epoch": 20.3,
+      "learning_rate": 0.0007967341214185534,
+      "loss": 0.1954,
+      "step": 37030
+    },
+    {
+      "epoch": 20.31,
+      "learning_rate": 0.0007962332197956321,
+      "loss": 0.2175,
+      "step": 37040
+    },
+    {
+      "epoch": 20.31,
+      "learning_rate": 0.0007957323181727109,
+      "loss": 0.386,
+      "step": 37050
+    },
+    {
+      "epoch": 20.32,
+      "learning_rate": 0.0007952314165497895,
+      "loss": 0.2214,
+      "step": 37060
+    },
+    {
+      "epoch": 20.32,
+      "learning_rate": 0.0007947305149268684,
+      "loss": 0.2198,
+      "step": 37070
+    },
+    {
+      "epoch": 20.33,
+      "learning_rate": 0.0007942296133039472,
+      "loss": 0.1962,
+      "step": 37080
+    },
+    {
+      "epoch": 20.33,
+      "learning_rate": 0.0007937287116810259,
+      "loss": 0.2282,
+      "step": 37090
+    },
+    {
+      "epoch": 20.34,
+      "learning_rate": 0.0007932278100581047,
+      "loss": 0.2197,
+      "step": 37100
+    },
+    {
+      "epoch": 20.34,
+      "eval_loss": 0.14504532516002655,
+      "eval_runtime": 1115.8382,
+      "eval_samples_per_second": 10.47,
+      "eval_steps_per_second": 0.655,
+      "eval_wer": 0.29482397889566087,
+      "step": 37100
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 40128,
+  "num_train_epochs": 22,
+  "save_steps": 100,
+  "total_flos": 1.3363368748374363e+20,
+  "trial_name": null,
+  "trial_params": null
+}