diff --git "a/trainer_state.json" "b/trainer_state.json"
--- "a/trainer_state.json"
+++ "b/trainer_state.json"
@@ -1,14368 +1,24028 @@
 {
   "best_metric": null,
   "best_model_checkpoint": null,
-  "epoch": 29.885057471264368,
-  "global_step": 20800,
+  "epoch": 50.0,
+  "global_step": 34800,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
   "log_history": [
     {
       "epoch": 0.01,
-      "learning_rate": 4.8e-06,
-      "loss": 24.5479,
+      "learning_rate": 5.999999999999999e-06,
+      "loss": 0.0495,
       "step": 10
     },
     {
       "epoch": 0.03,
-      "learning_rate": 1.0799999999999998e-05,
-      "loss": 24.3336,
+      "learning_rate": 1.1999999999999999e-05,
+      "loss": 0.044,
       "step": 20
     },
     {
       "epoch": 0.04,
-      "learning_rate": 1.68e-05,
-      "loss": 23.7501,
+      "learning_rate": 1.7999999999999997e-05,
+      "loss": 0.0557,
       "step": 30
     },
     {
       "epoch": 0.06,
-      "learning_rate": 2.28e-05,
-      "loss": 22.2394,
+      "learning_rate": 2.3999999999999997e-05,
+      "loss": 0.0405,
       "step": 40
     },
     {
       "epoch": 0.07,
-      "learning_rate": 2.88e-05,
-      "loss": 22.1365,
+      "learning_rate": 2.9999999999999997e-05,
+      "loss": 0.0389,
       "step": 50
     },
     {
       "epoch": 0.09,
-      "learning_rate": 3.42e-05,
-      "loss": 20.337,
+      "learning_rate": 3.5999999999999994e-05,
+      "loss": 0.0497,
       "step": 60
     },
     {
       "epoch": 0.1,
-      "learning_rate": 4.02e-05,
-      "loss": 19.8122,
+      "learning_rate": 4.2e-05,
+      "loss": 0.0358,
       "step": 70
     },
     {
       "epoch": 0.11,
-      "learning_rate": 4.62e-05,
-      "loss": 19.0623,
+      "learning_rate": 4.7999999999999994e-05,
+      "loss": 0.0594,
       "step": 80
     },
     {
       "epoch": 0.13,
-      "learning_rate": 5.2199999999999995e-05,
-      "loss": 16.9603,
+      "learning_rate": 5.399999999999999e-05,
+      "loss": 0.0458,
       "step": 90
     },
     {
       "epoch": 0.14,
-      "learning_rate": 5.82e-05,
-      "loss": 15.8856,
+      "learning_rate": 5.9999999999999995e-05,
+      "loss": 0.0433,
       "step": 100
     },
     {
       "epoch": 0.14,
-      "eval_loss": 15.38314151763916,
-      "eval_runtime": 39.1757,
-      "eval_samples_per_second": 11.665,
-      "eval_steps_per_second": 1.481,
-      "eval_wer": 1.0,
+      "eval_loss": 0.14026756584644318,
+      "eval_runtime": 39.7591,
+      "eval_samples_per_second": 11.494,
+      "eval_steps_per_second": 1.459,
+      "eval_wer": 0.1454600210811625,
       "step": 100
     },
     {
       "epoch": 0.16,
-      "learning_rate": 6.419999999999999e-05,
-      "loss": 14.1248,
+      "learning_rate": 6.599999999999999e-05,
+      "loss": 0.0569,
       "step": 110
     },
     {
       "epoch": 0.17,
-      "learning_rate": 7.02e-05,
-      "loss": 12.634,
+      "learning_rate": 7.199999999999999e-05,
+      "loss": 0.0497,
       "step": 120
     },
     {
       "epoch": 0.19,
-      "learning_rate": 7.62e-05,
-      "loss": 10.9827,
+      "learning_rate": 7.8e-05,
+      "loss": 0.0426,
       "step": 130
     },
     {
       "epoch": 0.2,
-      "learning_rate": 8.22e-05,
-      "loss": 9.6156,
+      "learning_rate": 8.4e-05,
+      "loss": 0.0521,
       "step": 140
     },
     {
       "epoch": 0.22,
-      "learning_rate": 8.819999999999999e-05,
-      "loss": 8.3307,
+      "learning_rate": 8.999999999999999e-05,
+      "loss": 0.0375,
       "step": 150
     },
     {
       "epoch": 0.23,
-      "learning_rate": 9.419999999999999e-05,
-      "loss": 7.1401,
+      "learning_rate": 9.599999999999999e-05,
+      "loss": 0.0505,
       "step": 160
     },
     {
       "epoch": 0.24,
-      "learning_rate": 0.0001002,
-      "loss": 6.2121,
+      "learning_rate": 0.000102,
+      "loss": 0.042,
       "step": 170
     },
     {
       "epoch": 0.26,
-      "learning_rate": 0.00010619999999999998,
-      "loss": 5.5431,
+      "learning_rate": 0.00010799999999999998,
+      "loss": 0.0549,
       "step": 180
     },
     {
       "epoch": 0.27,
-      "learning_rate": 0.00011219999999999999,
-      "loss": 5.1547,
+      "learning_rate": 0.00011399999999999999,
+      "loss": 0.0558,
       "step": 190
     },
     {
       "epoch": 0.29,
-      "learning_rate": 0.0001182,
-      "loss": 4.9694,
+      "learning_rate": 0.00011999999999999999,
+      "loss": 0.0423,
       "step": 200
     },
     {
       "epoch": 0.29,
-      "eval_loss": 4.939667701721191,
-      "eval_runtime": 39.3067,
-      "eval_samples_per_second": 11.627,
-      "eval_steps_per_second": 1.476,
-      "eval_wer": 1.0,
+      "eval_loss": 0.14491483569145203,
+      "eval_runtime": 39.8939,
+      "eval_samples_per_second": 11.455,
+      "eval_steps_per_second": 1.454,
+      "eval_wer": 0.14756813732871554,
       "step": 200
     },
     {
       "epoch": 0.3,
-      "learning_rate": 0.00012419999999999998,
-      "loss": 4.8874,
+      "learning_rate": 0.00012599999999999997,
+      "loss": 0.0706,
       "step": 210
     },
     {
       "epoch": 0.32,
-      "learning_rate": 0.0001302,
-      "loss": 4.8633,
+      "learning_rate": 0.00013199999999999998,
+      "loss": 0.0403,
       "step": 220
     },
     {
       "epoch": 0.33,
-      "learning_rate": 0.0001362,
-      "loss": 4.8666,
+      "learning_rate": 0.000138,
+      "loss": 0.0578,
       "step": 230
     },
     {
       "epoch": 0.34,
-      "learning_rate": 0.0001422,
-      "loss": 4.8345,
+      "learning_rate": 0.00014399999999999998,
+      "loss": 0.0642,
       "step": 240
     },
     {
       "epoch": 0.36,
-      "learning_rate": 0.0001482,
-      "loss": 4.8414,
+      "learning_rate": 0.00015,
+      "loss": 0.054,
       "step": 250
     },
     {
       "epoch": 0.37,
-      "learning_rate": 0.00015419999999999998,
-      "loss": 4.8116,
+      "learning_rate": 0.000156,
+      "loss": 0.0576,
       "step": 260
     },
     {
       "epoch": 0.39,
-      "learning_rate": 0.0001602,
-      "loss": 4.7888,
+      "learning_rate": 0.000162,
+      "loss": 0.0668,
       "step": 270
     },
     {
       "epoch": 0.4,
-      "learning_rate": 0.0001662,
-      "loss": 4.7886,
+      "learning_rate": 0.000168,
+      "loss": 0.057,
       "step": 280
     },
     {
       "epoch": 0.42,
-      "learning_rate": 0.00017219999999999998,
-      "loss": 4.7816,
+      "learning_rate": 0.00017399999999999997,
+      "loss": 0.05,
       "step": 290
     },
     {
       "epoch": 0.43,
-      "learning_rate": 0.00017819999999999997,
-      "loss": 4.8183,
+      "learning_rate": 0.00017999999999999998,
+      "loss": 0.0603,
       "step": 300
     },
     {
       "epoch": 0.43,
-      "eval_loss": 4.77665901184082,
-      "eval_runtime": 39.6162,
-      "eval_samples_per_second": 11.536,
-      "eval_steps_per_second": 1.464,
-      "eval_wer": 1.0,
+      "eval_loss": 0.15721559524536133,
+      "eval_runtime": 39.6461,
+      "eval_samples_per_second": 11.527,
+      "eval_steps_per_second": 1.463,
+      "eval_wer": 0.15163379009185363,
       "step": 300
     },
     {
       "epoch": 0.45,
-      "learning_rate": 0.00018419999999999998,
-      "loss": 4.7999,
+      "learning_rate": 0.000186,
+      "loss": 0.0903,
       "step": 310
     },
     {
       "epoch": 0.46,
-      "learning_rate": 0.0001902,
-      "loss": 4.7751,
+      "learning_rate": 0.00019199999999999998,
+      "loss": 0.0532,
       "step": 320
     },
     {
       "epoch": 0.47,
-      "learning_rate": 0.0001962,
-      "loss": 4.7608,
+      "learning_rate": 0.000198,
+      "loss": 0.079,
       "step": 330
     },
     {
       "epoch": 0.49,
-      "learning_rate": 0.0002022,
-      "loss": 4.743,
+      "learning_rate": 0.000204,
+      "loss": 0.0732,
       "step": 340
     },
     {
       "epoch": 0.5,
-      "learning_rate": 0.00020819999999999996,
-      "loss": 4.7291,
+      "learning_rate": 0.00020999999999999998,
+      "loss": 0.0617,
       "step": 350
     },
     {
       "epoch": 0.52,
-      "learning_rate": 0.00021419999999999998,
-      "loss": 4.7118,
+      "learning_rate": 0.00021599999999999996,
+      "loss": 0.0848,
       "step": 360
     },
     {
       "epoch": 0.53,
-      "learning_rate": 0.00022019999999999999,
-      "loss": 4.7286,
+      "learning_rate": 0.00022199999999999998,
+      "loss": 0.0599,
       "step": 370
     },
     {
       "epoch": 0.55,
-      "learning_rate": 0.00022619999999999997,
-      "loss": 4.715,
+      "learning_rate": 0.00022799999999999999,
+      "loss": 0.0706,
       "step": 380
     },
     {
       "epoch": 0.56,
-      "learning_rate": 0.00023219999999999998,
-      "loss": 4.709,
+      "learning_rate": 0.000234,
+      "loss": 0.058,
       "step": 390
     },
     {
       "epoch": 0.57,
-      "learning_rate": 0.0002382,
-      "loss": 4.683,
+      "learning_rate": 0.00023999999999999998,
+      "loss": 0.0691,
       "step": 400
     },
     {
       "epoch": 0.57,
-      "eval_loss": 4.713740348815918,
-      "eval_runtime": 39.3947,
-      "eval_samples_per_second": 11.601,
-      "eval_steps_per_second": 1.472,
-      "eval_wer": 1.0,
+      "eval_loss": 0.16631954908370972,
+      "eval_runtime": 39.7215,
+      "eval_samples_per_second": 11.505,
+      "eval_steps_per_second": 1.46,
+      "eval_wer": 0.1669929227525975,
       "step": 400
     },
     {
       "epoch": 0.59,
-      "learning_rate": 0.00024419999999999997,
-      "loss": 4.7096,
+      "learning_rate": 0.00024599999999999996,
+      "loss": 0.0786,
       "step": 410
     },
     {
       "epoch": 0.6,
-      "learning_rate": 0.00025019999999999996,
-      "loss": 4.7024,
+      "learning_rate": 0.00025199999999999995,
+      "loss": 0.0908,
       "step": 420
     },
     {
       "epoch": 0.62,
-      "learning_rate": 0.0002562,
-      "loss": 4.7074,
+      "learning_rate": 0.000258,
+      "loss": 0.0918,
       "step": 430
     },
     {
       "epoch": 0.63,
-      "learning_rate": 0.0002622,
-      "loss": 4.7047,
+      "learning_rate": 0.00026399999999999997,
+      "loss": 0.0768,
       "step": 440
     },
     {
       "epoch": 0.65,
-      "learning_rate": 0.00026819999999999996,
-      "loss": 4.6906,
+      "learning_rate": 0.00027,
+      "loss": 0.0976,
       "step": 450
     },
     {
       "epoch": 0.66,
-      "learning_rate": 0.0002742,
-      "loss": 4.6973,
+      "learning_rate": 0.000276,
+      "loss": 0.1104,
       "step": 460
     },
     {
       "epoch": 0.68,
-      "learning_rate": 0.0002802,
-      "loss": 4.6773,
+      "learning_rate": 0.00028199999999999997,
+      "loss": 0.0866,
       "step": 470
     },
     {
       "epoch": 0.69,
-      "learning_rate": 0.00028619999999999996,
-      "loss": 4.6905,
+      "learning_rate": 0.00028799999999999995,
+      "loss": 0.1069,
       "step": 480
     },
     {
       "epoch": 0.7,
-      "learning_rate": 0.00029219999999999995,
-      "loss": 4.6664,
+      "learning_rate": 0.000294,
+      "loss": 0.0814,
       "step": 490
     },
     {
       "epoch": 0.72,
-      "learning_rate": 0.0002982,
-      "loss": 4.6742,
+      "learning_rate": 0.0003,
+      "loss": 0.087,
       "step": 500
     },
     {
       "epoch": 0.72,
-      "eval_loss": 4.695952415466309,
-      "eval_runtime": 39.4463,
-      "eval_samples_per_second": 11.585,
-      "eval_steps_per_second": 1.47,
-      "eval_wer": 1.0,
+      "eval_loss": 0.19561158120632172,
+      "eval_runtime": 39.5545,
+      "eval_samples_per_second": 11.554,
+      "eval_steps_per_second": 1.466,
+      "eval_wer": 0.18732118656828792,
       "step": 500
     },
     {
       "epoch": 0.73,
-      "learning_rate": 0.0002998969578017664,
-      "loss": 4.7007,
+      "learning_rate": 0.0002999125364431487,
+      "loss": 0.1204,
       "step": 510
     },
     {
       "epoch": 0.75,
-      "learning_rate": 0.0002997497546614327,
-      "loss": 4.6779,
+      "learning_rate": 0.00029982507288629735,
+      "loss": 0.0861,
       "step": 520
     },
     {
       "epoch": 0.76,
-      "learning_rate": 0.0002996025515210991,
-      "loss": 4.6749,
+      "learning_rate": 0.00029973760932944606,
+      "loss": 0.1073,
       "step": 530
     },
     {
       "epoch": 0.78,
-      "learning_rate": 0.00029945534838076544,
-      "loss": 4.6655,
+      "learning_rate": 0.0002996501457725947,
+      "loss": 0.1028,
       "step": 540
     },
     {
       "epoch": 0.79,
-      "learning_rate": 0.0002993081452404318,
-      "loss": 4.6543,
+      "learning_rate": 0.0002995626822157434,
+      "loss": 0.087,
       "step": 550
     },
     {
       "epoch": 0.8,
-      "learning_rate": 0.0002991609421000981,
-      "loss": 4.6716,
+      "learning_rate": 0.0002994752186588921,
+      "loss": 0.1206,
       "step": 560
     },
     {
       "epoch": 0.82,
-      "learning_rate": 0.00029901373895976446,
-      "loss": 4.6725,
+      "learning_rate": 0.00029938775510204076,
+      "loss": 0.0949,
       "step": 570
     },
     {
       "epoch": 0.83,
-      "learning_rate": 0.00029886653581943077,
-      "loss": 4.6846,
+      "learning_rate": 0.0002993002915451895,
+      "loss": 0.1248,
       "step": 580
     },
     {
       "epoch": 0.85,
-      "learning_rate": 0.00029871933267909713,
-      "loss": 4.6431,
+      "learning_rate": 0.0002992128279883382,
+      "loss": 0.115,
       "step": 590
     },
     {
       "epoch": 0.86,
-      "learning_rate": 0.00029857212953876344,
-      "loss": 4.6485,
+      "learning_rate": 0.00029912536443148685,
+      "loss": 0.1038,
       "step": 600
     },
     {
       "epoch": 0.86,
-      "eval_loss": 4.668909072875977,
-      "eval_runtime": 39.3362,
-      "eval_samples_per_second": 11.618,
-      "eval_steps_per_second": 1.474,
-      "eval_wer": 1.0,
+      "eval_loss": 0.2007717341184616,
+      "eval_runtime": 39.645,
+      "eval_samples_per_second": 11.527,
+      "eval_steps_per_second": 1.463,
+      "eval_wer": 0.18882698388796867,
       "step": 600
     },
     {
       "epoch": 0.88,
-      "learning_rate": 0.0002984249263984298,
-      "loss": 4.6752,
+      "learning_rate": 0.00029903790087463557,
+      "loss": 0.1188,
       "step": 610
     },
     {
       "epoch": 0.89,
-      "learning_rate": 0.00029827772325809616,
-      "loss": 4.6491,
+      "learning_rate": 0.00029895043731778423,
+      "loss": 0.097,
       "step": 620
     },
     {
       "epoch": 0.91,
-      "learning_rate": 0.00029813052011776246,
-      "loss": 4.6515,
+      "learning_rate": 0.00029886297376093294,
+      "loss": 0.097,
       "step": 630
     },
     {
       "epoch": 0.92,
-      "learning_rate": 0.0002979833169774288,
-      "loss": 4.6491,
+      "learning_rate": 0.0002987755102040816,
+      "loss": 0.1028,
       "step": 640
     },
     {
       "epoch": 0.93,
-      "learning_rate": 0.0002978361138370952,
-      "loss": 4.6027,
+      "learning_rate": 0.00029868804664723027,
+      "loss": 0.0929,
       "step": 650
     },
     {
       "epoch": 0.95,
-      "learning_rate": 0.0002976889106967615,
-      "loss": 4.6736,
+      "learning_rate": 0.000298600583090379,
+      "loss": 0.1155,
       "step": 660
     },
     {
       "epoch": 0.96,
-      "learning_rate": 0.00029754170755642785,
-      "loss": 4.6461,
+      "learning_rate": 0.00029851311953352764,
+      "loss": 0.1659,
       "step": 670
     },
     {
       "epoch": 0.98,
-      "learning_rate": 0.00029739450441609416,
-      "loss": 4.6184,
+      "learning_rate": 0.00029842565597667636,
+      "loss": 0.1001,
       "step": 680
     },
     {
       "epoch": 0.99,
-      "learning_rate": 0.0002972473012757605,
-      "loss": 4.6395,
+      "learning_rate": 0.00029833819241982507,
+      "loss": 0.1038,
       "step": 690
     },
     {
       "epoch": 1.01,
-      "learning_rate": 0.0002971000981354269,
-      "loss": 4.6174,
+      "learning_rate": 0.00029825072886297373,
+      "loss": 0.1149,
       "step": 700
     },
     {
       "epoch": 1.01,
-      "eval_loss": 4.663065433502197,
-      "eval_runtime": 39.4396,
-      "eval_samples_per_second": 11.587,
-      "eval_steps_per_second": 1.471,
-      "eval_wer": 0.9998494202680319,
+      "eval_loss": 0.21170839667320251,
+      "eval_runtime": 39.8933,
+      "eval_samples_per_second": 11.456,
+      "eval_steps_per_second": 1.454,
+      "eval_wer": 0.19876524619786176,
       "step": 700
     },
     {
       "epoch": 1.02,
-      "learning_rate": 0.0002969528949950932,
-      "loss": 4.6355,
+      "learning_rate": 0.00029816326530612245,
+      "loss": 0.1066,
       "step": 710
     },
     {
       "epoch": 1.03,
-      "learning_rate": 0.00029680569185475955,
-      "loss": 4.611,
+      "learning_rate": 0.0002980758017492711,
+      "loss": 0.118,
       "step": 720
     },
     {
       "epoch": 1.05,
-      "learning_rate": 0.0002966584887144259,
-      "loss": 4.6402,
+      "learning_rate": 0.00029798833819241977,
+      "loss": 0.1548,
       "step": 730
     },
     {
       "epoch": 1.06,
-      "learning_rate": 0.0002965112855740922,
-      "loss": 4.6334,
+      "learning_rate": 0.0002979008746355685,
+      "loss": 0.1112,
       "step": 740
     },
     {
       "epoch": 1.08,
-      "learning_rate": 0.0002963640824337586,
-      "loss": 4.6247,
+      "learning_rate": 0.00029781341107871714,
+      "loss": 0.1176,
       "step": 750
     },
     {
       "epoch": 1.09,
-      "learning_rate": 0.0002962168792934249,
-      "loss": 4.6173,
+      "learning_rate": 0.00029772594752186586,
+      "loss": 0.1084,
       "step": 760
     },
     {
       "epoch": 1.11,
-      "learning_rate": 0.00029606967615309124,
-      "loss": 4.5934,
+      "learning_rate": 0.0002976384839650146,
+      "loss": 0.1014,
       "step": 770
     },
     {
       "epoch": 1.12,
-      "learning_rate": 0.00029592247301275755,
-      "loss": 4.6392,
+      "learning_rate": 0.00029755102040816323,
+      "loss": 0.1117,
       "step": 780
     },
     {
       "epoch": 1.14,
-      "learning_rate": 0.0002957752698724239,
-      "loss": 4.5958,
+      "learning_rate": 0.00029746355685131195,
+      "loss": 0.1001,
       "step": 790
     },
     {
       "epoch": 1.15,
-      "learning_rate": 0.00029562806673209027,
-      "loss": 4.5931,
+      "learning_rate": 0.0002973760932944606,
+      "loss": 0.1031,
       "step": 800
     },
     {
       "epoch": 1.15,
-      "eval_loss": 4.590127944946289,
-      "eval_runtime": 39.464,
-      "eval_samples_per_second": 11.58,
-      "eval_steps_per_second": 1.47,
-      "eval_wer": 0.9968378256286704,
+      "eval_loss": 0.18768654763698578,
+      "eval_runtime": 39.7924,
+      "eval_samples_per_second": 11.485,
+      "eval_steps_per_second": 1.458,
+      "eval_wer": 0.18822466496009638,
       "step": 800
     },
     {
       "epoch": 1.16,
-      "learning_rate": 0.00029548086359175663,
-      "loss": 4.5797,
+      "learning_rate": 0.0002972886297376093,
+      "loss": 0.0994,
       "step": 810
     },
     {
       "epoch": 1.18,
-      "learning_rate": 0.00029533366045142293,
-      "loss": 4.5442,
+      "learning_rate": 0.000297201166180758,
+      "loss": 0.0947,
       "step": 820
     },
     {
       "epoch": 1.19,
-      "learning_rate": 0.0002951864573110893,
-      "loss": 4.5553,
+      "learning_rate": 0.00029711370262390665,
+      "loss": 0.1321,
       "step": 830
     },
     {
       "epoch": 1.21,
-      "learning_rate": 0.0002950392541707556,
-      "loss": 4.5279,
+      "learning_rate": 0.00029702623906705536,
+      "loss": 0.1205,
       "step": 840
     },
     {
       "epoch": 1.22,
-      "learning_rate": 0.00029489205103042196,
-      "loss": 4.5188,
+      "learning_rate": 0.0002969387755102041,
+      "loss": 0.1585,
       "step": 850
     },
     {
       "epoch": 1.24,
-      "learning_rate": 0.00029474484789008827,
-      "loss": 4.5404,
+      "learning_rate": 0.00029685131195335274,
+      "loss": 0.106,
       "step": 860
     },
     {
       "epoch": 1.25,
-      "learning_rate": 0.00029459764474975463,
-      "loss": 4.5379,
+      "learning_rate": 0.00029676384839650145,
+      "loss": 0.1093,
       "step": 870
     },
     {
       "epoch": 1.26,
-      "learning_rate": 0.000294450441609421,
-      "loss": 4.4347,
+      "learning_rate": 0.0002966763848396501,
+      "loss": 0.1274,
       "step": 880
     },
     {
       "epoch": 1.28,
-      "learning_rate": 0.0002943032384690873,
-      "loss": 4.4093,
+      "learning_rate": 0.00029658892128279883,
+      "loss": 0.1058,
       "step": 890
     },
     {
       "epoch": 1.29,
-      "learning_rate": 0.00029415603532875366,
-      "loss": 4.3427,
+      "learning_rate": 0.0002965014577259475,
+      "loss": 0.1333,
       "step": 900
     },
     {
       "epoch": 1.29,
-      "eval_loss": 4.386187553405762,
-      "eval_runtime": 39.5926,
-      "eval_samples_per_second": 11.543,
-      "eval_steps_per_second": 1.465,
-      "eval_wer": 1.0,
+      "eval_loss": 0.1909557580947876,
+      "eval_runtime": 39.5497,
+      "eval_samples_per_second": 11.555,
+      "eval_steps_per_second": 1.467,
+      "eval_wer": 0.19500075289865984,
       "step": 900
     },
     {
       "epoch": 1.31,
-      "learning_rate": 0.00029400883218842,
-      "loss": 4.28,
+      "learning_rate": 0.00029641399416909615,
+      "loss": 0.1212,
       "step": 910
     },
     {
       "epoch": 1.32,
-      "learning_rate": 0.0002938616290480863,
-      "loss": 4.1754,
+      "learning_rate": 0.00029632653061224487,
+      "loss": 0.1079,
       "step": 920
     },
     {
       "epoch": 1.34,
-      "learning_rate": 0.0002937144259077527,
-      "loss": 4.0524,
+      "learning_rate": 0.0002962390670553935,
+      "loss": 0.1354,
       "step": 930
     },
     {
       "epoch": 1.35,
-      "learning_rate": 0.000293567222767419,
-      "loss": 3.8665,
+      "learning_rate": 0.00029615160349854224,
+      "loss": 0.0998,
       "step": 940
     },
     {
       "epoch": 1.36,
-      "learning_rate": 0.00029342001962708535,
-      "loss": 3.786,
+      "learning_rate": 0.00029606413994169096,
+      "loss": 0.105,
       "step": 950
     },
     {
       "epoch": 1.38,
-      "learning_rate": 0.0002932728164867517,
-      "loss": 3.4787,
+      "learning_rate": 0.0002959766763848396,
+      "loss": 0.1337,
       "step": 960
     },
     {
       "epoch": 1.39,
-      "learning_rate": 0.000293125613346418,
-      "loss": 3.2242,
+      "learning_rate": 0.00029588921282798833,
+      "loss": 0.0865,
       "step": 970
     },
     {
       "epoch": 1.41,
-      "learning_rate": 0.0002929784102060844,
-      "loss": 2.9959,
+      "learning_rate": 0.000295801749271137,
+      "loss": 0.1259,
       "step": 980
     },
     {
       "epoch": 1.42,
-      "learning_rate": 0.00029283120706575074,
-      "loss": 2.8204,
+      "learning_rate": 0.0002957142857142857,
+      "loss": 0.1159,
       "step": 990
     },
     {
       "epoch": 1.44,
-      "learning_rate": 0.00029268400392541704,
-      "loss": 2.6594,
+      "learning_rate": 0.00029562682215743437,
+      "loss": 0.1086,
       "step": 1000
     },
     {
       "epoch": 1.44,
-      "eval_loss": 2.4058356285095215,
-      "eval_runtime": 39.9286,
-      "eval_samples_per_second": 11.445,
-      "eval_steps_per_second": 1.453,
-      "eval_wer": 0.9799728956482457,
+      "eval_loss": 0.19512739777565002,
+      "eval_runtime": 39.4175,
+      "eval_samples_per_second": 11.594,
+      "eval_steps_per_second": 1.471,
+      "eval_wer": 0.2017768408372233,
       "step": 1000
     },
     {
       "epoch": 1.45,
-      "learning_rate": 0.0002925368007850834,
-      "loss": 2.527,
+      "learning_rate": 0.00029553935860058303,
+      "loss": 0.119,
       "step": 1010
     },
     {
       "epoch": 1.47,
-      "learning_rate": 0.0002923895976447497,
-      "loss": 2.371,
+      "learning_rate": 0.00029545189504373174,
+      "loss": 0.1213,
       "step": 1020
     },
     {
       "epoch": 1.48,
-      "learning_rate": 0.00029224239450441607,
-      "loss": 2.271,
+      "learning_rate": 0.00029536443148688046,
+      "loss": 0.1069,
       "step": 1030
     },
     {
       "epoch": 1.49,
-      "learning_rate": 0.0002920951913640824,
-      "loss": 2.185,
+      "learning_rate": 0.0002952769679300291,
+      "loss": 0.1051,
       "step": 1040
     },
     {
       "epoch": 1.51,
-      "learning_rate": 0.00029194798822374874,
-      "loss": 2.167,
+      "learning_rate": 0.00029518950437317784,
+      "loss": 0.1337,
       "step": 1050
     },
     {
       "epoch": 1.52,
-      "learning_rate": 0.0002918007850834151,
-      "loss": 2.0004,
+      "learning_rate": 0.0002951020408163265,
+      "loss": 0.1232,
       "step": 1060
     },
     {
       "epoch": 1.54,
-      "learning_rate": 0.00029165358194308146,
-      "loss": 1.9067,
+      "learning_rate": 0.0002950145772594752,
+      "loss": 0.1016,
       "step": 1070
     },
     {
       "epoch": 1.55,
-      "learning_rate": 0.00029150637880274777,
-      "loss": 1.8721,
+      "learning_rate": 0.00029492711370262387,
+      "loss": 0.1196,
       "step": 1080
     },
     {
       "epoch": 1.57,
-      "learning_rate": 0.0002913591756624141,
-      "loss": 1.7606,
+      "learning_rate": 0.00029483965014577253,
+      "loss": 0.1081,
       "step": 1090
     },
     {
       "epoch": 1.58,
-      "learning_rate": 0.00029121197252208043,
-      "loss": 1.729,
+      "learning_rate": 0.00029475218658892125,
+      "loss": 0.1025,
       "step": 1100
     },
     {
       "epoch": 1.58,
-      "eval_loss": 1.5358420610427856,
-      "eval_runtime": 39.5025,
-      "eval_samples_per_second": 11.569,
-      "eval_steps_per_second": 1.468,
-      "eval_wer": 0.8920343321788887,
+      "eval_loss": 0.1834399253129959,
+      "eval_runtime": 39.434,
+      "eval_samples_per_second": 11.589,
+      "eval_steps_per_second": 1.471,
+      "eval_wer": 0.19469959343472368,
       "step": 1100
     },
     {
       "epoch": 1.59,
-      "learning_rate": 0.0002910647693817468,
-      "loss": 1.621,
+      "learning_rate": 0.00029466472303206996,
+      "loss": 0.1053,
       "step": 1110
     },
     {
       "epoch": 1.61,
-      "learning_rate": 0.0002909175662414131,
-      "loss": 1.6394,
+      "learning_rate": 0.0002945772594752186,
+      "loss": 0.0881,
       "step": 1120
     },
     {
       "epoch": 1.62,
-      "learning_rate": 0.00029077036310107946,
-      "loss": 1.626,
+      "learning_rate": 0.00029448979591836734,
+      "loss": 0.129,
       "step": 1130
     },
     {
       "epoch": 1.64,
-      "learning_rate": 0.0002906231599607458,
-      "loss": 1.5561,
+      "learning_rate": 0.000294402332361516,
+      "loss": 0.0928,
       "step": 1140
     },
     {
       "epoch": 1.65,
-      "learning_rate": 0.0002904759568204122,
-      "loss": 1.5221,
+      "learning_rate": 0.0002943148688046647,
+      "loss": 0.1065,
       "step": 1150
     },
     {
       "epoch": 1.67,
-      "learning_rate": 0.0002903287536800785,
-      "loss": 1.4047,
+      "learning_rate": 0.0002942274052478134,
+      "loss": 0.1062,
       "step": 1160
     },
     {
       "epoch": 1.68,
-      "learning_rate": 0.00029018155053974485,
-      "loss": 1.4521,
+      "learning_rate": 0.0002941399416909621,
+      "loss": 0.0941,
       "step": 1170
     },
     {
       "epoch": 1.7,
-      "learning_rate": 0.00029003434739941115,
-      "loss": 1.3526,
+      "learning_rate": 0.00029405247813411075,
+      "loss": 0.1117,
       "step": 1180
     },
     {
       "epoch": 1.71,
-      "learning_rate": 0.00028988714425907746,
-      "loss": 1.2765,
+      "learning_rate": 0.00029396501457725947,
+      "loss": 0.087,
       "step": 1190
     },
     {
       "epoch": 1.72,
-      "learning_rate": 0.0002897399411187438,
-      "loss": 1.3998,
+      "learning_rate": 0.0002938775510204081,
+      "loss": 0.137,
       "step": 1200
     },
     {
       "epoch": 1.72,
-      "eval_loss": 1.1598572731018066,
-      "eval_runtime": 39.5369,
-      "eval_samples_per_second": 11.559,
-      "eval_steps_per_second": 1.467,
-      "eval_wer": 0.8021382321939466,
+      "eval_loss": 0.18502399325370789,
+      "eval_runtime": 39.3835,
+      "eval_samples_per_second": 11.604,
+      "eval_steps_per_second": 1.473,
+      "eval_wer": 0.1925914771871706,
       "step": 1200
     },
     {
       "epoch": 1.74,
-      "learning_rate": 0.0002895927379784102,
-      "loss": 1.2611,
+      "learning_rate": 0.00029379008746355684,
+      "loss": 0.1106,
       "step": 1210
     },
     {
       "epoch": 1.75,
-      "learning_rate": 0.00028944553483807654,
-      "loss": 1.2765,
+      "learning_rate": 0.0002937026239067055,
+      "loss": 0.0867,
       "step": 1220
     },
     {
       "epoch": 1.77,
-      "learning_rate": 0.00028929833169774285,
-      "loss": 1.1803,
+      "learning_rate": 0.0002936151603498542,
+      "loss": 0.1207,
       "step": 1230
     },
     {
       "epoch": 1.78,
-      "learning_rate": 0.0002891511285574092,
-      "loss": 1.2348,
+      "learning_rate": 0.0002935276967930029,
+      "loss": 0.1277,
       "step": 1240
     },
     {
       "epoch": 1.8,
-      "learning_rate": 0.00028900392541707557,
-      "loss": 1.1927,
+      "learning_rate": 0.0002934402332361516,
+      "loss": 0.1125,
       "step": 1250
     },
     {
       "epoch": 1.81,
-      "learning_rate": 0.0002888567222767419,
-      "loss": 1.141,
+      "learning_rate": 0.0002933527696793003,
+      "loss": 0.1062,
       "step": 1260
     },
     {
       "epoch": 1.82,
-      "learning_rate": 0.0002887095191364082,
-      "loss": 1.2261,
+      "learning_rate": 0.0002932653061224489,
+      "loss": 0.0973,
       "step": 1270
     },
     {
       "epoch": 1.84,
-      "learning_rate": 0.00028856231599607454,
-      "loss": 1.1062,
+      "learning_rate": 0.00029317784256559763,
+      "loss": 0.124,
       "step": 1280
     },
     {
       "epoch": 1.85,
-      "learning_rate": 0.0002884151128557409,
-      "loss": 1.095,
+      "learning_rate": 0.00029309037900874634,
+      "loss": 0.0987,
       "step": 1290
     },
     {
       "epoch": 1.87,
-      "learning_rate": 0.00028826790971540726,
-      "loss": 1.1236,
+      "learning_rate": 0.000293002915451895,
+      "loss": 0.1062,
       "step": 1300
     },
     {
       "epoch": 1.87,
-      "eval_loss": 0.9830673336982727,
-      "eval_runtime": 39.4178,
-      "eval_samples_per_second": 11.594,
-      "eval_steps_per_second": 1.471,
-      "eval_wer": 0.7566631531395874,
+      "eval_loss": 0.17643560469150543,
+      "eval_runtime": 39.5145,
+      "eval_samples_per_second": 11.565,
+      "eval_steps_per_second": 1.468,
+      "eval_wer": 0.18837524469206446,
       "step": 1300
     },
     {
       "epoch": 1.88,
-      "learning_rate": 0.00028812070657507357,
-      "loss": 1.0961,
+      "learning_rate": 0.0002929154518950437,
+      "loss": 0.1073,
       "step": 1310
     },
     {
       "epoch": 1.9,
-      "learning_rate": 0.00028797350343473993,
-      "loss": 1.09,
+      "learning_rate": 0.0002928279883381924,
+      "loss": 0.0911,
       "step": 1320
     },
     {
       "epoch": 1.91,
-      "learning_rate": 0.0002878263002944063,
-      "loss": 1.0263,
+      "learning_rate": 0.0002927405247813411,
+      "loss": 0.1032,
       "step": 1330
     },
     {
       "epoch": 1.93,
-      "learning_rate": 0.0002876790971540726,
-      "loss": 1.0797,
+      "learning_rate": 0.00029265306122448976,
+      "loss": 0.1205,
       "step": 1340
     },
     {
       "epoch": 1.94,
-      "learning_rate": 0.0002875318940137389,
-      "loss": 1.0631,
+      "learning_rate": 0.00029256559766763847,
+      "loss": 0.1096,
       "step": 1350
     },
     {
       "epoch": 1.95,
-      "learning_rate": 0.00028738469087340526,
-      "loss": 0.9842,
+      "learning_rate": 0.00029247813411078713,
+      "loss": 0.11,
       "step": 1360
     },
     {
       "epoch": 1.97,
-      "learning_rate": 0.0002872374877330716,
-      "loss": 1.129,
+      "learning_rate": 0.00029239067055393585,
+      "loss": 0.1194,
       "step": 1370
     },
     {
       "epoch": 1.98,
-      "learning_rate": 0.00028709028459273793,
-      "loss": 0.9745,
+      "learning_rate": 0.0002923032069970845,
+      "loss": 0.1052,
       "step": 1380
     },
     {
       "epoch": 2.0,
-      "learning_rate": 0.0002869430814524043,
-      "loss": 0.9876,
+      "learning_rate": 0.0002922157434402332,
+      "loss": 0.1039,
       "step": 1390
     },
     {
       "epoch": 2.01,
-      "learning_rate": 0.00028679587831207065,
-      "loss": 1.0142,
+      "learning_rate": 0.0002921282798833819,
+      "loss": 0.1321,
       "step": 1400
     },
     {
       "epoch": 2.01,
-      "eval_loss": 0.8522623181343079,
-      "eval_runtime": 39.9987,
-      "eval_samples_per_second": 11.425,
-      "eval_steps_per_second": 1.45,
-      "eval_wer": 0.6712844451136877,
+      "eval_loss": 0.1891152262687683,
+      "eval_runtime": 39.7053,
+      "eval_samples_per_second": 11.51,
+      "eval_steps_per_second": 1.461,
+      "eval_wer": 0.18912814335190484,
       "step": 1400
     },
     {
       "epoch": 2.03,
-      "learning_rate": 0.000286648675171737,
-      "loss": 0.9384,
+      "learning_rate": 0.0002920408163265306,
+      "loss": 0.1047,
       "step": 1410
     },
     {
       "epoch": 2.04,
-      "learning_rate": 0.0002865014720314033,
-      "loss": 0.9115,
+      "learning_rate": 0.00029195335276967926,
+      "loss": 0.1019,
       "step": 1420
     },
     {
       "epoch": 2.05,
-      "learning_rate": 0.0002863542688910696,
-      "loss": 0.9601,
+      "learning_rate": 0.000291865889212828,
+      "loss": 0.1235,
       "step": 1430
     },
     {
       "epoch": 2.07,
-      "learning_rate": 0.000286207065750736,
-      "loss": 0.9199,
+      "learning_rate": 0.0002917784256559767,
+      "loss": 0.1034,
       "step": 1440
     },
     {
       "epoch": 2.08,
-      "learning_rate": 0.00028605986261040234,
-      "loss": 0.9359,
+      "learning_rate": 0.00029169096209912535,
+      "loss": 0.1205,
       "step": 1450
     },
     {
       "epoch": 2.1,
-      "learning_rate": 0.00028591265947006865,
-      "loss": 0.9353,
+      "learning_rate": 0.000291603498542274,
+      "loss": 0.1044,
       "step": 1460
     },
     {
       "epoch": 2.11,
-      "learning_rate": 0.000285765456329735,
-      "loss": 0.8925,
+      "learning_rate": 0.0002915160349854227,
+      "loss": 0.0894,
       "step": 1470
     },
     {
       "epoch": 2.13,
-      "learning_rate": 0.00028561825318940137,
-      "loss": 0.9429,
+      "learning_rate": 0.00029143731778425656,
+      "loss": 0.1223,
       "step": 1480
     },
     {
       "epoch": 2.14,
-      "learning_rate": 0.0002854710500490677,
-      "loss": 0.8772,
+      "learning_rate": 0.0002913498542274052,
+      "loss": 0.0766,
       "step": 1490
     },
     {
       "epoch": 2.16,
-      "learning_rate": 0.00028532384690873404,
-      "loss": 0.9132,
+      "learning_rate": 0.00029126239067055394,
+      "loss": 0.1328,
       "step": 1500
     },
     {
       "epoch": 2.16,
-      "eval_loss": 0.802708089351654,
-      "eval_runtime": 39.7319,
-      "eval_samples_per_second": 11.502,
-      "eval_steps_per_second": 1.46,
-      "eval_wer": 0.6539677759373589,
+      "eval_loss": 0.1817779541015625,
+      "eval_runtime": 39.602,
+      "eval_samples_per_second": 11.54,
+      "eval_steps_per_second": 1.465,
+      "eval_wer": 0.18717060683631984,
       "step": 1500
     },
     {
       "epoch": 2.17,
-      "learning_rate": 0.00028517664376840035,
-      "loss": 0.8947,
+      "learning_rate": 0.0002911749271137026,
+      "loss": 0.0977,
       "step": 1510
     },
     {
       "epoch": 2.18,
-      "learning_rate": 0.0002850294406280667,
-      "loss": 0.8716,
+      "learning_rate": 0.00029108746355685126,
+      "loss": 0.0956,
       "step": 1520
     },
     {
       "epoch": 2.2,
-      "learning_rate": 0.000284882237487733,
-      "loss": 0.8453,
+      "learning_rate": 0.00029099999999999997,
+      "loss": 0.1013,
       "step": 1530
     },
     {
       "epoch": 2.21,
-      "learning_rate": 0.00028473503434739937,
-      "loss": 0.8724,
+      "learning_rate": 0.00029091253644314863,
+      "loss": 0.0918,
       "step": 1540
     },
     {
       "epoch": 2.23,
-      "learning_rate": 0.00028458783120706573,
-      "loss": 0.835,
+      "learning_rate": 0.00029082507288629735,
+      "loss": 0.1058,
       "step": 1550
     },
     {
       "epoch": 2.24,
-      "learning_rate": 0.0002844406280667321,
-      "loss": 0.8617,
+      "learning_rate": 0.00029073760932944606,
+      "loss": 0.0987,
       "step": 1560
     },
     {
       "epoch": 2.26,
-      "learning_rate": 0.0002842934249263984,
-      "loss": 0.8362,
+      "learning_rate": 0.0002906501457725947,
+      "loss": 0.0973,
       "step": 1570
     },
     {
       "epoch": 2.27,
-      "learning_rate": 0.00028414622178606476,
-      "loss": 0.7738,
+      "learning_rate": 0.00029056268221574344,
+      "loss": 0.0959,
       "step": 1580
     },
     {
       "epoch": 2.28,
-      "learning_rate": 0.00028399901864573107,
-      "loss": 0.8412,
+      "learning_rate": 0.0002904752186588921,
+      "loss": 0.0975,
       "step": 1590
     },
     {
       "epoch": 2.3,
-      "learning_rate": 0.0002838518155053974,
-      "loss": 0.8157,
+      "learning_rate": 0.00029038775510204076,
+      "loss": 0.1008,
       "step": 1600
     },
     {
       "epoch": 2.3,
-      "eval_loss": 0.7228513956069946,
-      "eval_runtime": 39.6806,
-      "eval_samples_per_second": 11.517,
+      "eval_loss": 0.172745481133461,
+      "eval_runtime": 39.676,
+      "eval_samples_per_second": 11.518,
       "eval_steps_per_second": 1.462,
-      "eval_wer": 0.6179792199969885,
+      "eval_wer": 0.18988104201174522,
       "step": 1600
     },
     {
       "epoch": 2.31,
-      "learning_rate": 0.00028370461236506373,
-      "loss": 0.7675,
+      "learning_rate": 0.0002903002915451895,
+      "loss": 0.0855,
       "step": 1610
     },
     {
       "epoch": 2.33,
-      "learning_rate": 0.0002835574092247301,
-      "loss": 0.8323,
+      "learning_rate": 0.00029021282798833814,
+      "loss": 0.098,
       "step": 1620
     },
     {
       "epoch": 2.34,
-      "learning_rate": 0.00028341020608439645,
-      "loss": 0.784,
+      "learning_rate": 0.00029012536443148685,
+      "loss": 0.0943,
       "step": 1630
     },
     {
       "epoch": 2.36,
-      "learning_rate": 0.00028326300294406276,
-      "loss": 0.782,
+      "learning_rate": 0.00029003790087463557,
+      "loss": 0.0767,
       "step": 1640
     },
     {
       "epoch": 2.37,
-      "learning_rate": 0.0002831157998037291,
-      "loss": 0.7919,
+      "learning_rate": 0.0002899504373177842,
+      "loss": 0.1074,
       "step": 1650
     },
     {
       "epoch": 2.39,
-      "learning_rate": 0.0002829685966633955,
-      "loss": 0.7782,
+      "learning_rate": 0.00028986297376093294,
+      "loss": 0.103,
       "step": 1660
     },
     {
       "epoch": 2.4,
-      "learning_rate": 0.0002828213935230618,
-      "loss": 0.8246,
+      "learning_rate": 0.0002897755102040816,
+      "loss": 0.1019,
       "step": 1670
     },
     {
       "epoch": 2.41,
-      "learning_rate": 0.00028267419038272815,
-      "loss": 0.7631,
+      "learning_rate": 0.0002896880466472303,
+      "loss": 0.1072,
       "step": 1680
     },
     {
       "epoch": 2.43,
-      "learning_rate": 0.00028252698724239445,
-      "loss": 0.7991,
+      "learning_rate": 0.000289600583090379,
+      "loss": 0.1117,
       "step": 1690
     },
     {
       "epoch": 2.44,
-      "learning_rate": 0.0002823797841020608,
-      "loss": 0.8154,
+      "learning_rate": 0.00028951311953352764,
+      "loss": 0.1097,
       "step": 1700
     },
     {
       "epoch": 2.44,
-      "eval_loss": 0.6884952783584595,
-      "eval_runtime": 39.9781,
-      "eval_samples_per_second": 11.431,
-      "eval_steps_per_second": 1.451,
-      "eval_wer": 0.6060834211715103,
+      "eval_loss": 0.17900413274765015,
+      "eval_runtime": 39.672,
+      "eval_samples_per_second": 11.519,
+      "eval_steps_per_second": 1.462,
+      "eval_wer": 0.19153741906339405,
       "step": 1700
     },
     {
       "epoch": 2.46,
-      "learning_rate": 0.0002822325809617272,
-      "loss": 0.7625,
+      "learning_rate": 0.00028942565597667635,
+      "loss": 0.086,
       "step": 1710
     },
     {
       "epoch": 2.47,
-      "learning_rate": 0.0002820853778213935,
-      "loss": 0.7888,
+      "learning_rate": 0.00028933819241982507,
+      "loss": 0.0891,
       "step": 1720
     },
     {
       "epoch": 2.49,
-      "learning_rate": 0.00028193817468105984,
-      "loss": 0.7091,
+      "learning_rate": 0.00028925072886297373,
+      "loss": 0.0871,
       "step": 1730
     },
     {
       "epoch": 2.5,
-      "learning_rate": 0.0002817909715407262,
-      "loss": 0.7846,
+      "learning_rate": 0.00028916326530612244,
+      "loss": 0.0927,
       "step": 1740
     },
     {
       "epoch": 2.51,
-      "learning_rate": 0.0002816437684003925,
-      "loss": 0.7651,
+      "learning_rate": 0.0002890758017492711,
+      "loss": 0.1066,
       "step": 1750
     },
     {
       "epoch": 2.53,
-      "learning_rate": 0.00028149656526005887,
-      "loss": 0.6977,
+      "learning_rate": 0.0002889883381924198,
+      "loss": 0.0864,
       "step": 1760
     },
     {
       "epoch": 2.54,
-      "learning_rate": 0.0002813493621197252,
-      "loss": 0.7321,
+      "learning_rate": 0.0002889008746355685,
+      "loss": 0.0953,
       "step": 1770
     },
     {
       "epoch": 2.56,
-      "learning_rate": 0.00028120215897939154,
-      "loss": 0.6753,
+      "learning_rate": 0.00028881341107871714,
+      "loss": 0.0818,
       "step": 1780
     },
     {
       "epoch": 2.57,
-      "learning_rate": 0.00028105495583905784,
-      "loss": 0.7269,
+      "learning_rate": 0.00028872594752186586,
+      "loss": 0.0872,
       "step": 1790
     },
     {
       "epoch": 2.59,
-      "learning_rate": 0.0002809077526987242,
-      "loss": 0.7424,
+      "learning_rate": 0.0002886384839650145,
+      "loss": 0.1117,
       "step": 1800
     },
     {
       "epoch": 2.59,
-      "eval_loss": 0.6070171594619751,
-      "eval_runtime": 39.745,
-      "eval_samples_per_second": 11.498,
-      "eval_steps_per_second": 1.459,
-      "eval_wer": 0.5691913868393315,
+      "eval_loss": 0.17877764999866486,
+      "eval_runtime": 39.5526,
+      "eval_samples_per_second": 11.554,
+      "eval_steps_per_second": 1.466,
+      "eval_wer": 0.1925914771871706,
       "step": 1800
     },
     {
       "epoch": 2.6,
-      "learning_rate": 0.00028076054955839056,
-      "loss": 0.6781,
+      "learning_rate": 0.00028855102040816323,
+      "loss": 0.0963,
       "step": 1810
     },
     {
       "epoch": 2.61,
-      "learning_rate": 0.0002806133464180569,
-      "loss": 0.7446,
+      "learning_rate": 0.00028846355685131195,
+      "loss": 0.1027,
       "step": 1820
     },
     {
       "epoch": 2.63,
-      "learning_rate": 0.00028046614327772323,
-      "loss": 0.7286,
+      "learning_rate": 0.0002883760932944606,
+      "loss": 0.1048,
       "step": 1830
     },
     {
       "epoch": 2.64,
-      "learning_rate": 0.0002803189401373896,
-      "loss": 0.707,
+      "learning_rate": 0.0002882886297376093,
+      "loss": 0.0812,
       "step": 1840
     },
     {
       "epoch": 2.66,
-      "learning_rate": 0.0002801717369970559,
-      "loss": 0.7032,
+      "learning_rate": 0.000288201166180758,
+      "loss": 0.1138,
       "step": 1850
     },
     {
       "epoch": 2.67,
-      "learning_rate": 0.00028002453385672226,
-      "loss": 0.7005,
+      "learning_rate": 0.0002881137026239067,
+      "loss": 0.0991,
       "step": 1860
     },
     {
       "epoch": 2.69,
-      "learning_rate": 0.00027987733071638856,
-      "loss": 0.6833,
+      "learning_rate": 0.00028802623906705536,
+      "loss": 0.11,
       "step": 1870
     },
     {
       "epoch": 2.7,
-      "learning_rate": 0.0002797301275760549,
-      "loss": 0.6812,
+      "learning_rate": 0.000287938775510204,
+      "loss": 0.1228,
       "step": 1880
     },
     {
       "epoch": 2.72,
-      "learning_rate": 0.0002795829244357213,
-      "loss": 0.6614,
+      "learning_rate": 0.00028785131195335274,
+      "loss": 0.0913,
       "step": 1890
     },
     {
       "epoch": 2.73,
-      "learning_rate": 0.00027943572129538765,
-      "loss": 0.6949,
+      "learning_rate": 0.00028776384839650145,
+      "loss": 0.1088,
       "step": 1900
     },
     {
       "epoch": 2.73,
-      "eval_loss": 0.5801286697387695,
-      "eval_runtime": 39.6689,
-      "eval_samples_per_second": 11.52,
-      "eval_steps_per_second": 1.462,
-      "eval_wer": 0.5460021081162475,
+      "eval_loss": 0.17957444489002228,
+      "eval_runtime": 39.6133,
+      "eval_samples_per_second": 11.537,
+      "eval_steps_per_second": 1.464,
+      "eval_wer": 0.17918988104201175,
       "step": 1900
     },
     {
       "epoch": 2.74,
-      "learning_rate": 0.00027928851815505395,
-      "loss": 0.676,
+      "learning_rate": 0.0002876763848396501,
+      "loss": 0.0987,
       "step": 1910
     },
     {
       "epoch": 2.76,
-      "learning_rate": 0.0002791413150147203,
-      "loss": 0.7185,
+      "learning_rate": 0.0002875889212827988,
+      "loss": 0.1139,
       "step": 1920
     },
     {
       "epoch": 2.77,
-      "learning_rate": 0.0002789941118743866,
-      "loss": 0.6175,
+      "learning_rate": 0.0002875014577259475,
+      "loss": 0.1016,
       "step": 1930
     },
     {
       "epoch": 2.79,
-      "learning_rate": 0.000278846908734053,
-      "loss": 0.6519,
+      "learning_rate": 0.0002874139941690962,
+      "loss": 0.0882,
       "step": 1940
     },
     {
       "epoch": 2.8,
-      "learning_rate": 0.0002786997055937193,
-      "loss": 0.662,
+      "learning_rate": 0.00028732653061224486,
+      "loss": 0.1073,
       "step": 1950
     },
     {
       "epoch": 2.82,
-      "learning_rate": 0.00027855250245338565,
-      "loss": 0.6317,
+      "learning_rate": 0.0002872390670553935,
+      "loss": 0.0956,
       "step": 1960
     },
     {
       "epoch": 2.83,
-      "learning_rate": 0.000278405299313052,
-      "loss": 0.6532,
+      "learning_rate": 0.00028715160349854224,
+      "loss": 0.0834,
       "step": 1970
     },
     {
       "epoch": 2.84,
-      "learning_rate": 0.0002782580961727183,
-      "loss": 0.6097,
+      "learning_rate": 0.00028706413994169095,
+      "loss": 0.1044,
       "step": 1980
     },
     {
       "epoch": 2.86,
-      "learning_rate": 0.0002781108930323847,
-      "loss": 0.6441,
+      "learning_rate": 0.0002869766763848396,
+      "loss": 0.0842,
       "step": 1990
     },
     {
       "epoch": 2.87,
-      "learning_rate": 0.00027796368989205103,
-      "loss": 0.6748,
+      "learning_rate": 0.00028688921282798833,
+      "loss": 0.1192,
       "step": 2000
     },
     {
       "epoch": 2.87,
-      "eval_loss": 0.5619427561759949,
-      "eval_runtime": 39.9278,
-      "eval_samples_per_second": 11.446,
-      "eval_steps_per_second": 1.453,
-      "eval_wer": 0.5387742809817798,
+      "eval_loss": 0.17964738607406616,
+      "eval_runtime": 39.5923,
+      "eval_samples_per_second": 11.543,
+      "eval_steps_per_second": 1.465,
+      "eval_wer": 0.1900316217437133,
       "step": 2000
     },
     {
       "epoch": 2.89,
-      "learning_rate": 0.00027781648675171734,
-      "loss": 0.6749,
+      "learning_rate": 0.000286801749271137,
+      "loss": 0.0993,
       "step": 2010
     },
     {
       "epoch": 2.9,
-      "learning_rate": 0.0002776692836113837,
-      "loss": 0.6225,
+      "learning_rate": 0.0002867142857142857,
+      "loss": 0.0792,
       "step": 2020
     },
     {
       "epoch": 2.92,
-      "learning_rate": 0.00027752208047105,
-      "loss": 0.592,
+      "learning_rate": 0.00028662682215743437,
+      "loss": 0.0904,
       "step": 2030
     },
     {
       "epoch": 2.93,
-      "learning_rate": 0.00027737487733071637,
-      "loss": 0.6007,
+      "learning_rate": 0.0002865393586005831,
+      "loss": 0.0849,
       "step": 2040
     },
     {
       "epoch": 2.95,
-      "learning_rate": 0.00027722767419038273,
-      "loss": 0.6488,
+      "learning_rate": 0.00028645189504373174,
+      "loss": 0.109,
       "step": 2050
     },
     {
       "epoch": 2.96,
-      "learning_rate": 0.00027708047105004903,
-      "loss": 0.6334,
+      "learning_rate": 0.0002863644314868804,
+      "loss": 0.0976,
       "step": 2060
     },
     {
       "epoch": 2.97,
-      "learning_rate": 0.0002769332679097154,
-      "loss": 0.6266,
+      "learning_rate": 0.0002862769679300291,
+      "loss": 0.1072,
       "step": 2070
     },
     {
       "epoch": 2.99,
-      "learning_rate": 0.00027678606476938175,
-      "loss": 0.5994,
+      "learning_rate": 0.00028618950437317783,
+      "loss": 0.0885,
       "step": 2080
     },
     {
       "epoch": 3.0,
-      "learning_rate": 0.00027663886162904806,
-      "loss": 0.6151,
+      "learning_rate": 0.0002861020408163265,
+      "loss": 0.0931,
       "step": 2090
     },
     {
       "epoch": 3.02,
-      "learning_rate": 0.0002764916584887144,
-      "loss": 0.5569,
+      "learning_rate": 0.0002860145772594752,
+      "loss": 0.1131,
       "step": 2100
     },
     {
       "epoch": 3.02,
-      "eval_loss": 0.5206429958343506,
-      "eval_runtime": 39.8569,
-      "eval_samples_per_second": 11.466,
-      "eval_steps_per_second": 1.455,
-      "eval_wer": 0.5130251468152387,
+      "eval_loss": 0.1814257949590683,
+      "eval_runtime": 39.7474,
+      "eval_samples_per_second": 11.498,
+      "eval_steps_per_second": 1.459,
+      "eval_wer": 0.18355669326908597,
       "step": 2100
     },
     {
       "epoch": 3.03,
-      "learning_rate": 0.00027634445534838073,
-      "loss": 0.5417,
+      "learning_rate": 0.00028592711370262387,
+      "loss": 0.0937,
       "step": 2110
     },
     {
       "epoch": 3.05,
-      "learning_rate": 0.0002761972522080471,
-      "loss": 0.6448,
+      "learning_rate": 0.0002858396501457726,
+      "loss": 0.126,
       "step": 2120
     },
     {
       "epoch": 3.06,
-      "learning_rate": 0.0002760500490677134,
-      "loss": 0.5474,
+      "learning_rate": 0.00028575218658892125,
+      "loss": 0.0868,
       "step": 2130
     },
     {
       "epoch": 3.07,
-      "learning_rate": 0.00027590284592737976,
-      "loss": 0.5912,
+      "learning_rate": 0.0002856647230320699,
+      "loss": 0.0902,
       "step": 2140
     },
     {
       "epoch": 3.09,
-      "learning_rate": 0.0002757556427870461,
-      "loss": 0.5482,
+      "learning_rate": 0.0002855772594752186,
+      "loss": 0.0938,
       "step": 2150
     },
     {
       "epoch": 3.1,
-      "learning_rate": 0.0002756084396467125,
-      "loss": 0.5619,
+      "learning_rate": 0.00028548979591836734,
+      "loss": 0.0834,
       "step": 2160
     },
     {
       "epoch": 3.12,
-      "learning_rate": 0.0002754612365063788,
-      "loss": 0.581,
+      "learning_rate": 0.000285402332361516,
+      "loss": 0.1019,
       "step": 2170
     },
     {
       "epoch": 3.13,
-      "learning_rate": 0.00027531403336604514,
-      "loss": 0.5275,
+      "learning_rate": 0.0002853148688046647,
+      "loss": 0.0826,
       "step": 2180
     },
     {
       "epoch": 3.15,
-      "learning_rate": 0.00027516683022571145,
-      "loss": 0.5497,
+      "learning_rate": 0.0002852274052478134,
+      "loss": 0.0969,
       "step": 2190
     },
     {
       "epoch": 3.16,
-      "learning_rate": 0.0002750196270853778,
-      "loss": 0.5584,
+      "learning_rate": 0.0002851399416909621,
+      "loss": 0.1054,
       "step": 2200
     },
     {
       "epoch": 3.16,
-      "eval_loss": 0.4875911772251129,
-      "eval_runtime": 39.9849,
-      "eval_samples_per_second": 11.429,
-      "eval_steps_per_second": 1.451,
-      "eval_wer": 0.49028760728805904,
+      "eval_loss": 0.1734437644481659,
+      "eval_runtime": 39.7519,
+      "eval_samples_per_second": 11.496,
+      "eval_steps_per_second": 1.459,
+      "eval_wer": 0.1794910405059479,
       "step": 2200
     },
     {
       "epoch": 3.18,
-      "learning_rate": 0.0002748724239450441,
-      "loss": 0.5276,
+      "learning_rate": 0.00028505247813411075,
+      "loss": 0.0836,
       "step": 2210
     },
     {
       "epoch": 3.19,
-      "learning_rate": 0.0002747252208047105,
-      "loss": 0.5582,
+      "learning_rate": 0.00028496501457725946,
+      "loss": 0.1054,
       "step": 2220
     },
     {
       "epoch": 3.2,
-      "learning_rate": 0.00027457801766437684,
-      "loss": 0.5294,
+      "learning_rate": 0.0002848775510204081,
+      "loss": 0.0864,
       "step": 2230
     },
     {
       "epoch": 3.22,
-      "learning_rate": 0.00027443081452404314,
-      "loss": 0.5978,
+      "learning_rate": 0.00028479008746355684,
+      "loss": 0.105,
       "step": 2240
     },
     {
       "epoch": 3.23,
-      "learning_rate": 0.0002742836113837095,
-      "loss": 0.5331,
+      "learning_rate": 0.0002847026239067055,
+      "loss": 0.104,
       "step": 2250
     },
     {
       "epoch": 3.25,
-      "learning_rate": 0.00027413640824337586,
-      "loss": 0.5714,
+      "learning_rate": 0.0002846151603498542,
+      "loss": 0.1042,
       "step": 2260
     },
     {
       "epoch": 3.26,
-      "learning_rate": 0.00027398920510304217,
-      "loss": 0.549,
+      "learning_rate": 0.0002845276967930029,
+      "loss": 0.1108,
       "step": 2270
     },
     {
       "epoch": 3.28,
-      "learning_rate": 0.0002738420019627085,
-      "loss": 0.5706,
+      "learning_rate": 0.0002844402332361516,
+      "loss": 0.1239,
       "step": 2280
     },
     {
       "epoch": 3.29,
-      "learning_rate": 0.00027369479882237484,
-      "loss": 0.5235,
+      "learning_rate": 0.00028435276967930025,
+      "loss": 0.0847,
       "step": 2290
     },
     {
       "epoch": 3.3,
-      "learning_rate": 0.0002735475956820412,
-      "loss": 0.5333,
+      "learning_rate": 0.00028426530612244897,
+      "loss": 0.1043,
       "step": 2300
     },
     {
       "epoch": 3.3,
-      "eval_loss": 0.46249067783355713,
-      "eval_runtime": 40.1021,
-      "eval_samples_per_second": 11.396,
-      "eval_steps_per_second": 1.446,
-      "eval_wer": 0.4661948501731667,
+      "eval_loss": 0.19418245553970337,
+      "eval_runtime": 39.5906,
+      "eval_samples_per_second": 11.543,
+      "eval_steps_per_second": 1.465,
+      "eval_wer": 0.18566480951663905,
       "step": 2300
     },
     {
       "epoch": 3.32,
-      "learning_rate": 0.00027340039254170756,
-      "loss": 0.5333,
+      "learning_rate": 0.0002841778425655977,
+      "loss": 0.1004,
       "step": 2310
     },
     {
       "epoch": 3.33,
-      "learning_rate": 0.00027325318940137386,
-      "loss": 0.578,
+      "learning_rate": 0.00028409037900874634,
+      "loss": 0.1123,
       "step": 2320
     },
     {
       "epoch": 3.35,
-      "learning_rate": 0.0002731059862610402,
-      "loss": 0.5243,
+      "learning_rate": 0.000284002915451895,
+      "loss": 0.0911,
       "step": 2330
     },
     {
       "epoch": 3.36,
-      "learning_rate": 0.0002729587831207066,
-      "loss": 0.5668,
+      "learning_rate": 0.0002839154518950437,
+      "loss": 0.0987,
       "step": 2340
     },
     {
       "epoch": 3.38,
-      "learning_rate": 0.0002728115799803729,
-      "loss": 0.5112,
+      "learning_rate": 0.0002838279883381924,
+      "loss": 0.0928,
       "step": 2350
     },
     {
       "epoch": 3.39,
-      "learning_rate": 0.0002726643768400392,
-      "loss": 0.5039,
+      "learning_rate": 0.0002837405247813411,
+      "loss": 0.0861,
       "step": 2360
     },
     {
       "epoch": 3.41,
-      "learning_rate": 0.00027251717369970556,
-      "loss": 0.5342,
+      "learning_rate": 0.00028365306122448976,
+      "loss": 0.1137,
       "step": 2370
     },
     {
       "epoch": 3.42,
-      "learning_rate": 0.0002723699705593719,
-      "loss": 0.4879,
+      "learning_rate": 0.00028356559766763847,
+      "loss": 0.0811,
       "step": 2380
     },
     {
       "epoch": 3.43,
-      "learning_rate": 0.0002722227674190382,
-      "loss": 0.5277,
+      "learning_rate": 0.0002834781341107872,
+      "loss": 0.0924,
       "step": 2390
     },
     {
       "epoch": 3.45,
-      "learning_rate": 0.0002720755642787046,
-      "loss": 0.5277,
+      "learning_rate": 0.00028339067055393585,
+      "loss": 0.1117,
       "step": 2400
     },
     {
       "epoch": 3.45,
-      "eval_loss": 0.4557473361492157,
-      "eval_runtime": 39.6153,
-      "eval_samples_per_second": 11.536,
-      "eval_steps_per_second": 1.464,
-      "eval_wer": 0.4631832555338051,
+      "eval_loss": 0.18514806032180786,
+      "eval_runtime": 39.8212,
+      "eval_samples_per_second": 11.476,
+      "eval_steps_per_second": 1.457,
+      "eval_wer": 0.19183857852733022,
       "step": 2400
     },
     {
       "epoch": 3.46,
-      "learning_rate": 0.00027192836113837095,
-      "loss": 0.5052,
+      "learning_rate": 0.0002833032069970845,
+      "loss": 0.0852,
       "step": 2410
     },
     {
       "epoch": 3.48,
-      "learning_rate": 0.0002717811579980373,
-      "loss": 0.5292,
+      "learning_rate": 0.0002832157434402332,
+      "loss": 0.1054,
       "step": 2420
     },
     {
       "epoch": 3.49,
-      "learning_rate": 0.0002716339548577036,
-      "loss": 0.5448,
+      "learning_rate": 0.0002831282798833819,
+      "loss": 0.1085,
       "step": 2430
     },
     {
       "epoch": 3.51,
-      "learning_rate": 0.0002714867517173699,
-      "loss": 0.5089,
+      "learning_rate": 0.0002830408163265306,
+      "loss": 0.0801,
       "step": 2440
     },
     {
       "epoch": 3.52,
-      "learning_rate": 0.0002713395485770363,
-      "loss": 0.5381,
+      "learning_rate": 0.00028295335276967926,
+      "loss": 0.0959,
       "step": 2450
     },
     {
       "epoch": 3.53,
-      "learning_rate": 0.00027119234543670264,
-      "loss": 0.5119,
+      "learning_rate": 0.000282865889212828,
+      "loss": 0.0874,
       "step": 2460
     },
     {
       "epoch": 3.55,
-      "learning_rate": 0.00027104514229636895,
-      "loss": 0.5093,
+      "learning_rate": 0.00028277842565597663,
+      "loss": 0.1038,
       "step": 2470
     },
     {
       "epoch": 3.56,
-      "learning_rate": 0.0002708979391560353,
-      "loss": 0.4931,
+      "learning_rate": 0.00028269096209912535,
+      "loss": 0.1012,
       "step": 2480
     },
     {
       "epoch": 3.58,
-      "learning_rate": 0.00027075073601570167,
-      "loss": 0.4823,
+      "learning_rate": 0.00028260349854227406,
+      "loss": 0.0862,
       "step": 2490
     },
     {
       "epoch": 3.59,
-      "learning_rate": 0.000270603532875368,
-      "loss": 0.479,
+      "learning_rate": 0.0002825160349854227,
+      "loss": 0.0922,
       "step": 2500
     },
     {
       "epoch": 3.59,
-      "eval_loss": 0.42780354619026184,
-      "eval_runtime": 39.8004,
-      "eval_samples_per_second": 11.482,
-      "eval_steps_per_second": 1.457,
-      "eval_wer": 0.44421020930582744,
+      "eval_loss": 0.18909135460853577,
+      "eval_runtime": 39.6765,
+      "eval_samples_per_second": 11.518,
+      "eval_steps_per_second": 1.462,
+      "eval_wer": 0.1846107513928625,
       "step": 2500
     },
     {
       "epoch": 3.61,
-      "learning_rate": 0.00027045632973503433,
-      "loss": 0.4703,
+      "learning_rate": 0.0002824285714285714,
+      "loss": 0.0741,
       "step": 2510
     },
     {
       "epoch": 3.62,
-      "learning_rate": 0.00027030912659470064,
-      "loss": 0.4812,
+      "learning_rate": 0.0002823411078717201,
+      "loss": 0.0942,
       "step": 2520
     },
     {
       "epoch": 3.64,
-      "learning_rate": 0.000270161923454367,
-      "loss": 0.5042,
+      "learning_rate": 0.00028225364431486876,
+      "loss": 0.089,
       "step": 2530
     },
     {
       "epoch": 3.65,
-      "learning_rate": 0.0002700147203140333,
-      "loss": 0.486,
+      "learning_rate": 0.0002821661807580175,
+      "loss": 0.0882,
       "step": 2540
     },
     {
       "epoch": 3.66,
-      "learning_rate": 0.00026986751717369967,
-      "loss": 0.4923,
+      "learning_rate": 0.00028207871720116614,
+      "loss": 0.092,
       "step": 2550
     },
     {
       "epoch": 3.68,
-      "learning_rate": 0.00026972031403336603,
-      "loss": 0.4595,
+      "learning_rate": 0.00028199125364431485,
+      "loss": 0.0834,
       "step": 2560
     },
     {
       "epoch": 3.69,
-      "learning_rate": 0.0002695731108930324,
-      "loss": 0.6039,
+      "learning_rate": 0.00028190379008746357,
+      "loss": 0.1412,
       "step": 2570
     },
     {
       "epoch": 3.71,
-      "learning_rate": 0.0002694259077526987,
-      "loss": 0.4577,
+      "learning_rate": 0.00028181632653061223,
+      "loss": 0.0847,
       "step": 2580
     },
     {
       "epoch": 3.72,
-      "learning_rate": 0.00026927870461236506,
-      "loss": 0.4911,
+      "learning_rate": 0.0002817288629737609,
+      "loss": 0.089,
       "step": 2590
     },
     {
       "epoch": 3.74,
-      "learning_rate": 0.00026913150147203136,
-      "loss": 0.5368,
+      "learning_rate": 0.0002816413994169096,
+      "loss": 0.1084,
       "step": 2600
     },
     {
       "epoch": 3.74,
-      "eval_loss": 0.43260544538497925,
-      "eval_runtime": 39.7779,
-      "eval_samples_per_second": 11.489,
-      "eval_steps_per_second": 1.458,
-      "eval_wer": 0.44677006474928477,
+      "eval_loss": 0.18555234372615814,
+      "eval_runtime": 39.7139,
+      "eval_samples_per_second": 11.507,
+      "eval_steps_per_second": 1.46,
+      "eval_wer": 0.18536365005270292,
       "step": 2600
     },
     {
       "epoch": 3.75,
-      "learning_rate": 0.0002689842983316977,
-      "loss": 0.4625,
+      "learning_rate": 0.00028155393586005826,
+      "loss": 0.0743,
       "step": 2610
     },
     {
       "epoch": 3.76,
-      "learning_rate": 0.00026883709519136403,
-      "loss": 0.5655,
+      "learning_rate": 0.000281466472303207,
+      "loss": 0.1098,
       "step": 2620
     },
     {
       "epoch": 3.78,
-      "learning_rate": 0.0002686898920510304,
-      "loss": 0.4575,
+      "learning_rate": 0.00028137900874635564,
+      "loss": 0.0825,
       "step": 2630
     },
     {
       "epoch": 3.79,
-      "learning_rate": 0.00026854268891069675,
-      "loss": 0.5072,
+      "learning_rate": 0.00028129154518950436,
+      "loss": 0.1,
       "step": 2640
     },
     {
       "epoch": 3.81,
-      "learning_rate": 0.00026839548577036306,
-      "loss": 0.4799,
+      "learning_rate": 0.00028120408163265307,
+      "loss": 0.0886,
       "step": 2650
     },
     {
       "epoch": 3.82,
-      "learning_rate": 0.0002682482826300294,
-      "loss": 0.4849,
+      "learning_rate": 0.00028111661807580173,
+      "loss": 0.0783,
       "step": 2660
     },
     {
       "epoch": 3.84,
-      "learning_rate": 0.0002681010794896958,
-      "loss": 0.5101,
+      "learning_rate": 0.00028102915451895045,
+      "loss": 0.1155,
       "step": 2670
     },
     {
       "epoch": 3.85,
-      "learning_rate": 0.0002679538763493621,
-      "loss": 0.4985,
+      "learning_rate": 0.0002809416909620991,
+      "loss": 0.0977,
       "step": 2680
     },
     {
       "epoch": 3.86,
-      "learning_rate": 0.00026780667320902844,
-      "loss": 0.4664,
+      "learning_rate": 0.00028085422740524777,
+      "loss": 0.0882,
       "step": 2690
     },
     {
       "epoch": 3.88,
-      "learning_rate": 0.00026765947006869475,
-      "loss": 0.5113,
+      "learning_rate": 0.0002807667638483965,
+      "loss": 0.1132,
       "step": 2700
     },
     {
       "epoch": 3.88,
-      "eval_loss": 0.40216904878616333,
-      "eval_runtime": 40.3356,
-      "eval_samples_per_second": 11.33,
-      "eval_steps_per_second": 1.438,
-      "eval_wer": 0.4338202078000301,
+      "eval_loss": 0.1814679652452469,
+      "eval_runtime": 39.8894,
+      "eval_samples_per_second": 11.457,
+      "eval_steps_per_second": 1.454,
+      "eval_wer": 0.19018220147568138,
       "step": 2700
     },
     {
       "epoch": 3.89,
-      "learning_rate": 0.0002675122669283611,
-      "loss": 0.4684,
+      "learning_rate": 0.00028067930029154514,
+      "loss": 0.0943,
       "step": 2710
     },
     {
       "epoch": 3.91,
-      "learning_rate": 0.00026736506378802747,
-      "loss": 0.4687,
+      "learning_rate": 0.00028059183673469386,
+      "loss": 0.1055,
       "step": 2720
     },
     {
       "epoch": 3.92,
-      "learning_rate": 0.0002672178606476938,
-      "loss": 0.4609,
+      "learning_rate": 0.0002805043731778425,
+      "loss": 0.0958,
       "step": 2730
     },
     {
       "epoch": 3.94,
-      "learning_rate": 0.00026707065750736014,
-      "loss": 0.4715,
+      "learning_rate": 0.00028041690962099123,
+      "loss": 0.1021,
       "step": 2740
     },
     {
       "epoch": 3.95,
-      "learning_rate": 0.0002669234543670265,
-      "loss": 0.5112,
+      "learning_rate": 0.00028032944606413995,
+      "loss": 0.1098,
       "step": 2750
     },
     {
       "epoch": 3.97,
-      "learning_rate": 0.0002667762512266928,
-      "loss": 0.4614,
+      "learning_rate": 0.0002802419825072886,
+      "loss": 0.0958,
       "step": 2760
     },
     {
       "epoch": 3.98,
-      "learning_rate": 0.00026662904808635917,
-      "loss": 0.4613,
+      "learning_rate": 0.00028015451895043727,
+      "loss": 0.0834,
       "step": 2770
     },
     {
       "epoch": 3.99,
-      "learning_rate": 0.00026648184494602547,
-      "loss": 0.4496,
+      "learning_rate": 0.000280067055393586,
+      "loss": 0.0845,
       "step": 2780
     },
     {
       "epoch": 4.01,
-      "learning_rate": 0.00026633464180569183,
-      "loss": 0.4614,
+      "learning_rate": 0.00027997959183673465,
+      "loss": 0.0859,
       "step": 2790
     },
     {
       "epoch": 4.02,
-      "learning_rate": 0.00026618743866535814,
-      "loss": 0.4377,
+      "learning_rate": 0.00027989212827988336,
+      "loss": 0.0979,
       "step": 2800
     },
     {
       "epoch": 4.02,
-      "eval_loss": 0.3904813528060913,
-      "eval_runtime": 40.1396,
-      "eval_samples_per_second": 11.385,
-      "eval_steps_per_second": 1.445,
-      "eval_wer": 0.42824875771721127,
+      "eval_loss": 0.17596615850925446,
+      "eval_runtime": 39.7684,
+      "eval_samples_per_second": 11.492,
+      "eval_steps_per_second": 1.458,
+      "eval_wer": 0.18039451889775637,
       "step": 2800
     },
     {
       "epoch": 4.04,
-      "learning_rate": 0.0002660402355250245,
-      "loss": 0.4191,
+      "learning_rate": 0.000279804664723032,
+      "loss": 0.0867,
       "step": 2810
     },
     {
       "epoch": 4.05,
-      "learning_rate": 0.00026589303238469086,
-      "loss": 0.3959,
+      "learning_rate": 0.00027971720116618074,
+      "loss": 0.0853,
       "step": 2820
     },
     {
       "epoch": 4.07,
-      "learning_rate": 0.0002657458292443572,
-      "loss": 0.4222,
+      "learning_rate": 0.00027962973760932945,
+      "loss": 0.0766,
       "step": 2830
     },
     {
       "epoch": 4.08,
-      "learning_rate": 0.0002655986261040235,
-      "loss": 0.3959,
+      "learning_rate": 0.0002795422740524781,
+      "loss": 0.0863,
       "step": 2840
     },
     {
       "epoch": 4.09,
-      "learning_rate": 0.0002654514229636899,
-      "loss": 0.3709,
+      "learning_rate": 0.00027945481049562683,
+      "loss": 0.0715,
       "step": 2850
     },
     {
       "epoch": 4.11,
-      "learning_rate": 0.0002653042198233562,
-      "loss": 0.4196,
+      "learning_rate": 0.0002793673469387755,
+      "loss": 0.0864,
       "step": 2860
     },
     {
       "epoch": 4.12,
-      "learning_rate": 0.00026515701668302255,
-      "loss": 0.4266,
+      "learning_rate": 0.00027927988338192415,
+      "loss": 0.0931,
       "step": 2870
     },
     {
       "epoch": 4.14,
-      "learning_rate": 0.00026500981354268886,
-      "loss": 0.4069,
+      "learning_rate": 0.00027919241982507286,
+      "loss": 0.0739,
       "step": 2880
     },
     {
       "epoch": 4.15,
-      "learning_rate": 0.0002648626104023552,
-      "loss": 0.4123,
+      "learning_rate": 0.0002791049562682215,
+      "loss": 0.0761,
       "step": 2890
     },
     {
       "epoch": 4.17,
-      "learning_rate": 0.0002647154072620216,
-      "loss": 0.3715,
+      "learning_rate": 0.00027901749271137024,
+      "loss": 0.0698,
       "step": 2900
     },
     {
       "epoch": 4.17,
-      "eval_loss": 0.3695952892303467,
-      "eval_runtime": 39.8629,
-      "eval_samples_per_second": 11.464,
-      "eval_steps_per_second": 1.455,
-      "eval_wer": 0.40581237765396777,
+      "eval_loss": 0.1889517903327942,
+      "eval_runtime": 39.7733,
+      "eval_samples_per_second": 11.49,
+      "eval_steps_per_second": 1.458,
+      "eval_wer": 0.1826532148772775,
       "step": 2900
     },
     {
       "epoch": 4.18,
-      "learning_rate": 0.00026456820412168794,
-      "loss": 0.4446,
+      "learning_rate": 0.00027893002915451896,
+      "loss": 0.1061,
       "step": 2910
     },
     {
       "epoch": 4.2,
-      "learning_rate": 0.00026442100098135425,
-      "loss": 0.4482,
+      "learning_rate": 0.0002788425655976676,
+      "loss": 0.1011,
       "step": 2920
     },
     {
       "epoch": 4.21,
-      "learning_rate": 0.0002642737978410206,
-      "loss": 0.4032,
+      "learning_rate": 0.00027875510204081633,
+      "loss": 0.0772,
       "step": 2930
     },
     {
       "epoch": 4.22,
-      "learning_rate": 0.0002641265947006869,
-      "loss": 0.425,
+      "learning_rate": 0.000278667638483965,
+      "loss": 0.0964,
       "step": 2940
     },
     {
       "epoch": 4.24,
-      "learning_rate": 0.0002639793915603533,
-      "loss": 0.3987,
+      "learning_rate": 0.00027858017492711365,
+      "loss": 0.0824,
       "step": 2950
     },
     {
       "epoch": 4.25,
-      "learning_rate": 0.0002638321884200196,
-      "loss": 0.3933,
+      "learning_rate": 0.00027849271137026237,
+      "loss": 0.0788,
       "step": 2960
     },
     {
       "epoch": 4.27,
-      "learning_rate": 0.00026368498527968594,
-      "loss": 0.4381,
+      "learning_rate": 0.00027840524781341103,
+      "loss": 0.1011,
       "step": 2970
     },
     {
       "epoch": 4.28,
-      "learning_rate": 0.0002635377821393523,
-      "loss": 0.3662,
+      "learning_rate": 0.00027831778425655974,
+      "loss": 0.0775,
       "step": 2980
     },
     {
       "epoch": 4.3,
-      "learning_rate": 0.0002633905789990186,
-      "loss": 0.4008,
+      "learning_rate": 0.0002782303206997084,
+      "loss": 0.0912,
       "step": 2990
     },
     {
       "epoch": 4.31,
-      "learning_rate": 0.00026324337585868497,
-      "loss": 0.3764,
+      "learning_rate": 0.0002781428571428571,
+      "loss": 0.0862,
       "step": 3000
     },
     {
       "epoch": 4.31,
-      "eval_loss": 0.36124685406684875,
-      "eval_runtime": 39.8744,
-      "eval_samples_per_second": 11.461,
-      "eval_steps_per_second": 1.455,
-      "eval_wer": 0.3918084625809366,
+      "eval_loss": 0.17421171069145203,
+      "eval_runtime": 40.0805,
+      "eval_samples_per_second": 11.402,
+      "eval_steps_per_second": 1.447,
+      "eval_wer": 0.1715103147116398,
       "step": 3000
     },
     {
       "epoch": 4.32,
-      "learning_rate": 0.00026309617271835133,
-      "loss": 0.4272,
+      "learning_rate": 0.00027805539358600583,
+      "loss": 0.1038,
       "step": 3010
     },
     {
       "epoch": 4.34,
-      "learning_rate": 0.00026294896957801764,
-      "loss": 0.4066,
+      "learning_rate": 0.0002779679300291545,
+      "loss": 0.0934,
       "step": 3020
     },
     {
       "epoch": 4.35,
-      "learning_rate": 0.000262801766437684,
-      "loss": 0.3517,
+      "learning_rate": 0.0002778804664723032,
+      "loss": 0.0597,
       "step": 3030
     },
     {
       "epoch": 4.37,
-      "learning_rate": 0.0002626545632973503,
-      "loss": 0.4108,
+      "learning_rate": 0.00027779300291545187,
+      "loss": 0.1031,
       "step": 3040
     },
     {
       "epoch": 4.38,
-      "learning_rate": 0.00026250736015701666,
-      "loss": 0.3775,
+      "learning_rate": 0.00027770553935860053,
+      "loss": 0.073,
       "step": 3050
     },
     {
       "epoch": 4.4,
-      "learning_rate": 0.000262360157016683,
-      "loss": 0.3643,
+      "learning_rate": 0.00027761807580174925,
+      "loss": 0.0767,
       "step": 3060
     },
     {
       "epoch": 4.41,
-      "learning_rate": 0.00026221295387634933,
-      "loss": 0.395,
+      "learning_rate": 0.0002775306122448979,
+      "loss": 0.0834,
       "step": 3070
     },
     {
       "epoch": 4.43,
-      "learning_rate": 0.0002620657507360157,
-      "loss": 0.3699,
+      "learning_rate": 0.0002774431486880466,
+      "loss": 0.0829,
       "step": 3080
     },
     {
       "epoch": 4.44,
-      "learning_rate": 0.00026191854759568205,
-      "loss": 0.3657,
+      "learning_rate": 0.00027735568513119534,
+      "loss": 0.0804,
       "step": 3090
     },
     {
       "epoch": 4.45,
-      "learning_rate": 0.00026177134445534836,
-      "loss": 0.3915,
+      "learning_rate": 0.000277268221574344,
+      "loss": 0.0833,
       "step": 3100
     },
     {
       "epoch": 4.45,
-      "eval_loss": 0.3556678593158722,
-      "eval_runtime": 39.9374,
-      "eval_samples_per_second": 11.443,
-      "eval_steps_per_second": 1.452,
-      "eval_wer": 0.384731215178437,
+      "eval_loss": 0.17197643220424652,
+      "eval_runtime": 39.7627,
+      "eval_samples_per_second": 11.493,
+      "eval_steps_per_second": 1.459,
+      "eval_wer": 0.1794910405059479,
       "step": 3100
     },
     {
       "epoch": 4.47,
-      "learning_rate": 0.00026162414131501466,
-      "loss": 0.3968,
+      "learning_rate": 0.0002771807580174927,
+      "loss": 0.0851,
       "step": 3110
     },
     {
       "epoch": 4.48,
-      "learning_rate": 0.000261476938174681,
-      "loss": 0.4099,
+      "learning_rate": 0.0002770932944606414,
+      "loss": 0.0938,
       "step": 3120
     },
     {
       "epoch": 4.5,
-      "learning_rate": 0.0002613297350343474,
-      "loss": 0.422,
+      "learning_rate": 0.00027700583090379004,
+      "loss": 0.093,
       "step": 3130
     },
     {
       "epoch": 4.51,
-      "learning_rate": 0.0002611825318940137,
-      "loss": 0.4122,
+      "learning_rate": 0.00027691836734693875,
+      "loss": 0.0975,
       "step": 3140
     },
     {
       "epoch": 4.53,
-      "learning_rate": 0.00026103532875368005,
-      "loss": 0.3962,
+      "learning_rate": 0.0002768309037900874,
+      "loss": 0.0825,
       "step": 3150
     },
     {
       "epoch": 4.54,
-      "learning_rate": 0.0002608881256133464,
-      "loss": 0.3721,
+      "learning_rate": 0.0002767434402332361,
+      "loss": 0.075,
       "step": 3160
     },
     {
       "epoch": 4.55,
-      "learning_rate": 0.00026074092247301277,
-      "loss": 0.468,
+      "learning_rate": 0.00027665597667638484,
+      "loss": 0.119,
       "step": 3170
     },
     {
       "epoch": 4.57,
-      "learning_rate": 0.0002605937193326791,
-      "loss": 0.3766,
+      "learning_rate": 0.0002765685131195335,
+      "loss": 0.0823,
       "step": 3180
     },
     {
       "epoch": 4.58,
-      "learning_rate": 0.0002604465161923454,
-      "loss": 0.3768,
+      "learning_rate": 0.0002764810495626822,
+      "loss": 0.0881,
       "step": 3190
     },
     {
       "epoch": 4.6,
-      "learning_rate": 0.00026029931305201175,
-      "loss": 0.369,
+      "learning_rate": 0.0002763935860058309,
+      "loss": 0.0781,
       "step": 3200
     },
     {
       "epoch": 4.6,
-      "eval_loss": 0.36044731736183167,
-      "eval_runtime": 39.912,
-      "eval_samples_per_second": 11.45,
-      "eval_steps_per_second": 1.453,
-      "eval_wer": 0.39286252070471317,
+      "eval_loss": 0.18404971063137054,
+      "eval_runtime": 39.7696,
+      "eval_samples_per_second": 11.491,
+      "eval_steps_per_second": 1.458,
+      "eval_wer": 0.18551422978467097,
       "step": 3200
     },
     {
       "epoch": 4.61,
-      "learning_rate": 0.0002601521099116781,
-      "loss": 0.4042,
+      "learning_rate": 0.0002763061224489796,
+      "loss": 0.0898,
       "step": 3210
     },
     {
       "epoch": 4.63,
-      "learning_rate": 0.0002600049067713444,
-      "loss": 0.3628,
+      "learning_rate": 0.00027621865889212825,
+      "loss": 0.0843,
       "step": 3220
     },
     {
       "epoch": 4.64,
-      "learning_rate": 0.00025985770363101077,
-      "loss": 0.3977,
+      "learning_rate": 0.0002761311953352769,
+      "loss": 0.0868,
       "step": 3230
     },
     {
       "epoch": 4.66,
-      "learning_rate": 0.00025971050049067713,
-      "loss": 0.4106,
+      "learning_rate": 0.00027604373177842563,
+      "loss": 0.1109,
       "step": 3240
     },
     {
       "epoch": 4.67,
-      "learning_rate": 0.00025956329735034344,
-      "loss": 0.3944,
+      "learning_rate": 0.0002759562682215743,
+      "loss": 0.1084,
       "step": 3250
     },
     {
       "epoch": 4.68,
-      "learning_rate": 0.0002594160942100098,
-      "loss": 0.3964,
+      "learning_rate": 0.000275868804664723,
+      "loss": 0.0948,
       "step": 3260
     },
     {
       "epoch": 4.7,
-      "learning_rate": 0.0002592688910696761,
-      "loss": 0.3937,
+      "learning_rate": 0.0002757813411078717,
+      "loss": 0.106,
       "step": 3270
     },
     {
       "epoch": 4.71,
-      "learning_rate": 0.00025912168792934247,
-      "loss": 0.33,
+      "learning_rate": 0.0002756938775510204,
+      "loss": 0.0703,
       "step": 3280
     },
     {
       "epoch": 4.73,
-      "learning_rate": 0.00025897448478900877,
-      "loss": 0.4144,
+      "learning_rate": 0.0002756064139941691,
+      "loss": 0.1003,
       "step": 3290
     },
     {
       "epoch": 4.74,
-      "learning_rate": 0.00025882728164867513,
-      "loss": 0.3644,
+      "learning_rate": 0.00027551895043731776,
+      "loss": 0.0907,
       "step": 3300
     },
     {
       "epoch": 4.74,
-      "eval_loss": 0.3271957039833069,
-      "eval_runtime": 39.9356,
-      "eval_samples_per_second": 11.443,
-      "eval_steps_per_second": 1.452,
-      "eval_wer": 0.3604878783315766,
+      "eval_loss": 0.16755123436450958,
+      "eval_runtime": 40.1876,
+      "eval_samples_per_second": 11.372,
+      "eval_steps_per_second": 1.443,
+      "eval_wer": 0.17903930131004367,
       "step": 3300
     },
     {
       "epoch": 4.76,
-      "learning_rate": 0.0002586800785083415,
-      "loss": 0.3605,
+      "learning_rate": 0.0002754314868804664,
+      "loss": 0.0845,
       "step": 3310
     },
     {
       "epoch": 4.77,
-      "learning_rate": 0.00025853287536800785,
-      "loss": 0.3926,
+      "learning_rate": 0.00027534402332361513,
+      "loss": 0.0957,
       "step": 3320
     },
     {
       "epoch": 4.78,
-      "learning_rate": 0.00025838567222767416,
-      "loss": 0.3565,
+      "learning_rate": 0.0002752565597667638,
+      "loss": 0.0825,
       "step": 3330
     },
     {
       "epoch": 4.8,
-      "learning_rate": 0.0002582384690873405,
-      "loss": 0.3856,
+      "learning_rate": 0.0002751690962099125,
+      "loss": 0.1048,
       "step": 3340
     },
     {
       "epoch": 4.81,
-      "learning_rate": 0.00025809126594700683,
-      "loss": 0.3295,
+      "learning_rate": 0.0002750816326530612,
+      "loss": 0.0857,
       "step": 3350
     },
     {
       "epoch": 4.83,
-      "learning_rate": 0.0002579440628066732,
-      "loss": 0.3327,
+      "learning_rate": 0.0002749941690962099,
+      "loss": 0.0704,
       "step": 3360
     },
     {
       "epoch": 4.84,
-      "learning_rate": 0.0002577968596663395,
-      "loss": 0.374,
+      "learning_rate": 0.0002749067055393586,
+      "loss": 0.0895,
       "step": 3370
     },
     {
       "epoch": 4.86,
-      "learning_rate": 0.00025764965652600585,
-      "loss": 0.3291,
+      "learning_rate": 0.00027481924198250726,
+      "loss": 0.0637,
       "step": 3380
     },
     {
       "epoch": 4.87,
-      "learning_rate": 0.0002575024533856722,
-      "loss": 0.3902,
+      "learning_rate": 0.000274731778425656,
+      "loss": 0.0972,
       "step": 3390
     },
     {
       "epoch": 4.89,
-      "learning_rate": 0.0002573552502453385,
-      "loss": 0.3959,
+      "learning_rate": 0.00027464431486880464,
+      "loss": 0.0998,
       "step": 3400
     },
     {
       "epoch": 4.89,
-      "eval_loss": 0.3158166706562042,
-      "eval_runtime": 40.0972,
-      "eval_samples_per_second": 11.397,
-      "eval_steps_per_second": 1.446,
-      "eval_wer": 0.36681222707423583,
+      "eval_loss": 0.18358713388442993,
+      "eval_runtime": 39.8198,
+      "eval_samples_per_second": 11.477,
+      "eval_steps_per_second": 1.457,
+      "eval_wer": 0.1847613311248306,
       "step": 3400
     },
     {
       "epoch": 4.9,
-      "learning_rate": 0.0002572080471050049,
-      "loss": 0.3636,
+      "learning_rate": 0.0002745568513119533,
+      "loss": 0.0758,
       "step": 3410
     },
     {
       "epoch": 4.91,
-      "learning_rate": 0.00025706084396467124,
-      "loss": 0.3581,
+      "learning_rate": 0.000274469387755102,
+      "loss": 0.0953,
       "step": 3420
     },
     {
       "epoch": 4.93,
-      "learning_rate": 0.00025691364082433755,
-      "loss": 0.3505,
+      "learning_rate": 0.0002743819241982507,
+      "loss": 0.0848,
       "step": 3430
     },
     {
       "epoch": 4.94,
-      "learning_rate": 0.0002567664376840039,
-      "loss": 0.3967,
+      "learning_rate": 0.0002742944606413994,
+      "loss": 0.1245,
       "step": 3440
     },
     {
       "epoch": 4.96,
-      "learning_rate": 0.0002566192345436702,
-      "loss": 0.3328,
+      "learning_rate": 0.0002742069970845481,
+      "loss": 0.0751,
       "step": 3450
     },
     {
       "epoch": 4.97,
-      "learning_rate": 0.0002564720314033366,
-      "loss": 0.3625,
+      "learning_rate": 0.00027411953352769676,
+      "loss": 0.0844,
       "step": 3460
     },
     {
       "epoch": 4.99,
-      "learning_rate": 0.00025632482826300294,
-      "loss": 0.386,
+      "learning_rate": 0.0002740320699708455,
+      "loss": 0.1096,
       "step": 3470
     },
     {
       "epoch": 5.0,
-      "learning_rate": 0.00025617762512266924,
-      "loss": 0.3102,
+      "learning_rate": 0.00027394460641399414,
+      "loss": 0.0643,
       "step": 3480
     },
     {
       "epoch": 5.01,
-      "learning_rate": 0.0002560304219823356,
-      "loss": 0.3945,
+      "learning_rate": 0.0002738571428571428,
+      "loss": 0.1243,
       "step": 3490
     },
     {
       "epoch": 5.03,
-      "learning_rate": 0.00025588321884200196,
-      "loss": 0.3307,
+      "learning_rate": 0.0002737696793002915,
+      "loss": 0.0886,
       "step": 3500
     },
     {
       "epoch": 5.03,
-      "eval_loss": 0.32223132252693176,
-      "eval_runtime": 40.0169,
-      "eval_samples_per_second": 11.42,
-      "eval_steps_per_second": 1.449,
-      "eval_wer": 0.35702454449631077,
+      "eval_loss": 0.17481039464473724,
+      "eval_runtime": 39.889,
+      "eval_samples_per_second": 11.457,
+      "eval_steps_per_second": 1.454,
+      "eval_wer": 0.18039451889775637,
       "step": 3500
     },
     {
       "epoch": 5.04,
-      "learning_rate": 0.00025573601570166827,
-      "loss": 0.3412,
+      "learning_rate": 0.00027368221574344023,
+      "loss": 0.1015,
       "step": 3510
     },
     {
       "epoch": 5.06,
-      "learning_rate": 0.00025558881256133463,
-      "loss": 0.3243,
+      "learning_rate": 0.0002735947521865889,
+      "loss": 0.0802,
       "step": 3520
     },
     {
       "epoch": 5.07,
-      "learning_rate": 0.00025544160942100094,
-      "loss": 0.3513,
+      "learning_rate": 0.0002735072886297376,
+      "loss": 0.0888,
       "step": 3530
     },
     {
       "epoch": 5.09,
-      "learning_rate": 0.0002552944062806673,
-      "loss": 0.3336,
+      "learning_rate": 0.00027341982507288627,
+      "loss": 0.0844,
       "step": 3540
     },
     {
       "epoch": 5.1,
-      "learning_rate": 0.0002551472031403336,
-      "loss": 0.318,
+      "learning_rate": 0.000273332361516035,
+      "loss": 0.0687,
       "step": 3550
     },
     {
       "epoch": 5.11,
-      "learning_rate": 0.00025499999999999996,
-      "loss": 0.3662,
+      "learning_rate": 0.00027324489795918364,
+      "loss": 0.0928,
       "step": 3560
     },
     {
       "epoch": 5.13,
-      "learning_rate": 0.0002548527968596663,
-      "loss": 0.3265,
+      "learning_rate": 0.00027315743440233236,
+      "loss": 0.0816,
       "step": 3570
     },
     {
       "epoch": 5.14,
-      "learning_rate": 0.0002547055937193327,
-      "loss": 0.2784,
+      "learning_rate": 0.000273069970845481,
+      "loss": 0.0643,
       "step": 3580
     },
     {
       "epoch": 5.16,
-      "learning_rate": 0.000254558390578999,
-      "loss": 0.3218,
+      "learning_rate": 0.0002729825072886297,
+      "loss": 0.0854,
       "step": 3590
     },
     {
       "epoch": 5.17,
-      "learning_rate": 0.00025441118743866535,
-      "loss": 0.3283,
+      "learning_rate": 0.0002728950437317784,
+      "loss": 0.0798,
       "step": 3600
     },
     {
       "epoch": 5.17,
-      "eval_loss": 0.3063613176345825,
-      "eval_runtime": 40.0819,
-      "eval_samples_per_second": 11.402,
-      "eval_steps_per_second": 1.447,
-      "eval_wer": 0.3576268634241831,
+      "eval_loss": 0.18081925809383392,
+      "eval_runtime": 39.9753,
+      "eval_samples_per_second": 11.432,
+      "eval_steps_per_second": 1.451,
+      "eval_wer": 0.19153741906339405,
       "step": 3600
     },
     {
       "epoch": 5.19,
-      "learning_rate": 0.00025426398429833166,
-      "loss": 0.3153,
+      "learning_rate": 0.0002728075801749271,
+      "loss": 0.0843,
       "step": 3610
     },
     {
       "epoch": 5.2,
-      "learning_rate": 0.000254116781157998,
-      "loss": 0.2926,
+      "learning_rate": 0.00027272011661807577,
+      "loss": 0.0814,
       "step": 3620
     },
     {
       "epoch": 5.22,
-      "learning_rate": 0.0002539695780176643,
-      "loss": 0.3065,
+      "learning_rate": 0.0002726326530612245,
+      "loss": 0.0706,
       "step": 3630
     },
     {
       "epoch": 5.23,
-      "learning_rate": 0.0002538223748773307,
-      "loss": 0.3656,
+      "learning_rate": 0.00027254518950437315,
+      "loss": 0.099,
       "step": 3640
     },
     {
       "epoch": 5.24,
-      "learning_rate": 0.00025367517173699705,
-      "loss": 0.2764,
+      "learning_rate": 0.00027245772594752186,
+      "loss": 0.0597,
       "step": 3650
     },
     {
       "epoch": 5.26,
-      "learning_rate": 0.0002535279685966634,
-      "loss": 0.3197,
+      "learning_rate": 0.0002723702623906705,
+      "loss": 0.085,
       "step": 3660
     },
     {
       "epoch": 5.27,
-      "learning_rate": 0.0002533807654563297,
-      "loss": 0.3113,
+      "learning_rate": 0.0002722827988338192,
+      "loss": 0.0802,
       "step": 3670
     },
     {
       "epoch": 5.29,
-      "learning_rate": 0.0002532335623159961,
-      "loss": 0.3154,
+      "learning_rate": 0.0002721953352769679,
+      "loss": 0.0664,
       "step": 3680
     },
     {
       "epoch": 5.3,
-      "learning_rate": 0.0002530863591756624,
-      "loss": 0.3771,
+      "learning_rate": 0.0002721078717201166,
+      "loss": 0.0971,
       "step": 3690
     },
     {
       "epoch": 5.32,
-      "learning_rate": 0.00025293915603532874,
-      "loss": 0.2934,
+      "learning_rate": 0.00027202040816326527,
+      "loss": 0.065,
       "step": 3700
     },
     {
       "epoch": 5.32,
-      "eval_loss": 0.30499157309532166,
-      "eval_runtime": 39.856,
-      "eval_samples_per_second": 11.466,
-      "eval_steps_per_second": 1.455,
-      "eval_wer": 0.34829092004216233,
+      "eval_loss": 0.1817695051431656,
+      "eval_runtime": 39.9937,
+      "eval_samples_per_second": 11.427,
+      "eval_steps_per_second": 1.45,
+      "eval_wer": 0.1841590121969583,
       "step": 3700
     },
     {
       "epoch": 5.33,
-      "learning_rate": 0.00025279195289499505,
-      "loss": 0.3419,
+      "learning_rate": 0.000271932944606414,
+      "loss": 0.0862,
       "step": 3710
     },
     {
       "epoch": 5.34,
-      "learning_rate": 0.0002526447497546614,
-      "loss": 0.3312,
+      "learning_rate": 0.00027184548104956265,
+      "loss": 0.0886,
       "step": 3720
     },
     {
       "epoch": 5.36,
-      "learning_rate": 0.00025249754661432777,
-      "loss": 0.2881,
+      "learning_rate": 0.00027175801749271136,
+      "loss": 0.0663,
       "step": 3730
     },
     {
       "epoch": 5.37,
-      "learning_rate": 0.0002523503434739941,
-      "loss": 0.3223,
+      "learning_rate": 0.00027167055393586,
+      "loss": 0.0857,
       "step": 3740
     },
     {
       "epoch": 5.39,
-      "learning_rate": 0.00025220314033366043,
-      "loss": 0.3118,
+      "learning_rate": 0.00027158309037900874,
+      "loss": 0.0826,
       "step": 3750
     },
     {
       "epoch": 5.4,
-      "learning_rate": 0.0002520559371933268,
-      "loss": 0.3139,
+      "learning_rate": 0.0002714956268221574,
+      "loss": 0.0862,
       "step": 3760
     },
     {
       "epoch": 5.42,
-      "learning_rate": 0.0002519087340529931,
-      "loss": 0.3198,
+      "learning_rate": 0.0002714081632653061,
+      "loss": 0.0865,
       "step": 3770
     },
     {
       "epoch": 5.43,
-      "learning_rate": 0.00025176153091265946,
-      "loss": 0.2908,
+      "learning_rate": 0.0002713206997084548,
+      "loss": 0.0731,
       "step": 3780
     },
     {
       "epoch": 5.45,
-      "learning_rate": 0.00025161432777232577,
-      "loss": 0.3446,
+      "learning_rate": 0.0002712332361516035,
+      "loss": 0.0986,
       "step": 3790
     },
     {
       "epoch": 5.46,
-      "learning_rate": 0.00025146712463199213,
-      "loss": 0.3015,
+      "learning_rate": 0.00027114577259475215,
+      "loss": 0.0854,
       "step": 3800
     },
     {
       "epoch": 5.46,
-      "eval_loss": 0.28972870111465454,
-      "eval_runtime": 40.1556,
-      "eval_samples_per_second": 11.381,
-      "eval_steps_per_second": 1.444,
-      "eval_wer": 0.3416654118355669,
+      "eval_loss": 0.16976070404052734,
+      "eval_runtime": 39.7742,
+      "eval_samples_per_second": 11.49,
+      "eval_steps_per_second": 1.458,
+      "eval_wer": 0.1840084324649902,
       "step": 3800
     },
     {
       "epoch": 5.47,
-      "learning_rate": 0.0002513199214916585,
-      "loss": 0.3488,
+      "learning_rate": 0.00027105830903790087,
+      "loss": 0.0984,
       "step": 3810
     },
     {
       "epoch": 5.49,
-      "learning_rate": 0.0002511727183513248,
-      "loss": 0.2873,
+      "learning_rate": 0.00027097084548104953,
+      "loss": 0.0782,
       "step": 3820
     },
     {
       "epoch": 5.5,
-      "learning_rate": 0.00025102551521099116,
-      "loss": 0.2951,
+      "learning_rate": 0.00027088338192419824,
+      "loss": 0.0754,
       "step": 3830
     },
     {
       "epoch": 5.52,
-      "learning_rate": 0.0002508783120706575,
-      "loss": 0.3305,
+      "learning_rate": 0.00027079591836734696,
+      "loss": 0.0915,
       "step": 3840
     },
     {
       "epoch": 5.53,
-      "learning_rate": 0.0002507311089303238,
-      "loss": 0.3062,
+      "learning_rate": 0.00027070845481049556,
+      "loss": 0.0784,
       "step": 3850
     },
     {
       "epoch": 5.55,
-      "learning_rate": 0.0002505839057899902,
-      "loss": 0.3912,
+      "learning_rate": 0.0002706209912536443,
+      "loss": 0.1126,
       "step": 3860
     },
     {
       "epoch": 5.56,
-      "learning_rate": 0.0002504367026496565,
-      "loss": 0.2884,
+      "learning_rate": 0.000270533527696793,
+      "loss": 0.0727,
       "step": 3870
     },
     {
       "epoch": 5.57,
-      "learning_rate": 0.00025028949950932285,
-      "loss": 0.3128,
+      "learning_rate": 0.00027044606413994165,
+      "loss": 0.0842,
       "step": 3880
     },
     {
       "epoch": 5.59,
-      "learning_rate": 0.00025014229636898916,
-      "loss": 0.3343,
+      "learning_rate": 0.00027035860058309037,
+      "loss": 0.0961,
       "step": 3890
     },
     {
       "epoch": 5.6,
-      "learning_rate": 0.0002499950932286555,
-      "loss": 0.2929,
+      "learning_rate": 0.00027027113702623903,
+      "loss": 0.0745,
       "step": 3900
     },
     {
       "epoch": 5.6,
-      "eval_loss": 0.28876641392707825,
-      "eval_runtime": 40.338,
-      "eval_samples_per_second": 11.329,
-      "eval_steps_per_second": 1.438,
-      "eval_wer": 0.31862671284445115,
+      "eval_loss": 0.17608921229839325,
+      "eval_runtime": 39.9521,
+      "eval_samples_per_second": 11.439,
+      "eval_steps_per_second": 1.452,
+      "eval_wer": 0.18114741755759675,
       "step": 3900
     },
     {
       "epoch": 5.62,
-      "learning_rate": 0.0002498478900883219,
-      "loss": 0.3341,
+      "learning_rate": 0.00027018367346938775,
+      "loss": 0.0905,
       "step": 3910
     },
     {
       "epoch": 5.63,
-      "learning_rate": 0.00024970068694798824,
-      "loss": 0.3196,
+      "learning_rate": 0.0002700962099125364,
+      "loss": 0.0842,
       "step": 3920
     },
     {
       "epoch": 5.65,
-      "learning_rate": 0.00024955348380765454,
-      "loss": 0.3025,
+      "learning_rate": 0.0002700087463556851,
+      "loss": 0.0843,
       "step": 3930
     },
     {
       "epoch": 5.66,
-      "learning_rate": 0.0002494062806673209,
-      "loss": 0.3231,
+      "learning_rate": 0.0002699212827988338,
+      "loss": 0.0955,
       "step": 3940
     },
     {
       "epoch": 5.68,
-      "learning_rate": 0.0002492590775269872,
-      "loss": 0.2989,
+      "learning_rate": 0.0002698338192419825,
+      "loss": 0.0784,
       "step": 3950
     },
     {
       "epoch": 5.69,
-      "learning_rate": 0.00024911187438665357,
-      "loss": 0.304,
+      "learning_rate": 0.00026974635568513116,
+      "loss": 0.0801,
       "step": 3960
     },
     {
       "epoch": 5.7,
-      "learning_rate": 0.0002489646712463199,
-      "loss": 0.2816,
+      "learning_rate": 0.00026965889212827987,
+      "loss": 0.0815,
       "step": 3970
     },
     {
       "epoch": 5.72,
-      "learning_rate": 0.00024881746810598624,
-      "loss": 0.3096,
+      "learning_rate": 0.00026957142857142853,
+      "loss": 0.0825,
       "step": 3980
     },
     {
       "epoch": 5.73,
-      "learning_rate": 0.0002486702649656526,
-      "loss": 0.3,
+      "learning_rate": 0.00026948396501457725,
+      "loss": 0.0836,
       "step": 3990
     },
     {
       "epoch": 5.75,
-      "learning_rate": 0.0002485230618253189,
-      "loss": 0.2815,
+      "learning_rate": 0.0002693965014577259,
+      "loss": 0.0789,
       "step": 4000
     },
     {
       "epoch": 5.75,
-      "eval_loss": 0.28788086771965027,
-      "eval_runtime": 40.0114,
-      "eval_samples_per_second": 11.422,
-      "eval_steps_per_second": 1.45,
-      "eval_wer": 0.31666917632886615,
+      "eval_loss": 0.17331229150295258,
+      "eval_runtime": 40.1008,
+      "eval_samples_per_second": 11.396,
+      "eval_steps_per_second": 1.446,
+      "eval_wer": 0.1840084324649902,
       "step": 4000
     },
     {
       "epoch": 5.76,
-      "learning_rate": 0.00024837585868498526,
-      "loss": 0.3083,
+      "learning_rate": 0.0002693090379008746,
+      "loss": 0.0866,
       "step": 4010
     },
     {
       "epoch": 5.78,
-      "learning_rate": 0.0002482286555446516,
-      "loss": 0.2857,
+      "learning_rate": 0.00026922157434402334,
+      "loss": 0.0779,
       "step": 4020
     },
     {
       "epoch": 5.79,
-      "learning_rate": 0.00024808145240431793,
-      "loss": 0.3066,
+      "learning_rate": 0.000269134110787172,
+      "loss": 0.0789,
       "step": 4030
     },
     {
       "epoch": 5.8,
-      "learning_rate": 0.00024793424926398424,
-      "loss": 0.2842,
+      "learning_rate": 0.00026904664723032066,
+      "loss": 0.0841,
       "step": 4040
     },
     {
       "epoch": 5.82,
-      "learning_rate": 0.0002477870461236506,
-      "loss": 0.2664,
+      "learning_rate": 0.0002689591836734694,
+      "loss": 0.0677,
       "step": 4050
     },
     {
       "epoch": 5.83,
-      "learning_rate": 0.00024763984298331696,
-      "loss": 0.3455,
+      "learning_rate": 0.00026887172011661804,
+      "loss": 0.096,
       "step": 4060
     },
     {
       "epoch": 5.85,
-      "learning_rate": 0.0002474926398429833,
-      "loss": 0.2824,
+      "learning_rate": 0.00026878425655976675,
+      "loss": 0.0854,
       "step": 4070
     },
     {
       "epoch": 5.86,
-      "learning_rate": 0.0002473454367026496,
-      "loss": 0.2934,
+      "learning_rate": 0.0002686967930029154,
+      "loss": 0.0847,
       "step": 4080
     },
     {
       "epoch": 5.88,
-      "learning_rate": 0.000247198233562316,
-      "loss": 0.3343,
+      "learning_rate": 0.00026860932944606413,
+      "loss": 0.0986,
       "step": 4090
     },
     {
       "epoch": 5.89,
-      "learning_rate": 0.00024705103042198235,
-      "loss": 0.2778,
+      "learning_rate": 0.00026852186588921284,
+      "loss": 0.0903,
       "step": 4100
     },
     {
       "epoch": 5.89,
-      "eval_loss": 0.2641572952270508,
-      "eval_runtime": 39.8509,
-      "eval_samples_per_second": 11.468,
-      "eval_steps_per_second": 1.455,
-      "eval_wer": 0.3136575816895046,
+      "eval_loss": 0.17269666492938995,
+      "eval_runtime": 39.8943,
+      "eval_samples_per_second": 11.455,
+      "eval_steps_per_second": 1.454,
+      "eval_wer": 0.18370727300105405,
       "step": 4100
     },
     {
       "epoch": 5.91,
-      "learning_rate": 0.00024690382728164865,
-      "loss": 0.2685,
+      "learning_rate": 0.0002684344023323615,
+      "loss": 0.079,
       "step": 4110
     },
     {
       "epoch": 5.92,
-      "learning_rate": 0.00024675662414131496,
-      "loss": 0.3127,
+      "learning_rate": 0.00026834693877551016,
+      "loss": 0.1023,
       "step": 4120
     },
     {
       "epoch": 5.93,
-      "learning_rate": 0.0002466094210009813,
-      "loss": 0.2618,
+      "learning_rate": 0.0002682594752186589,
+      "loss": 0.0685,
       "step": 4130
     },
     {
       "epoch": 5.95,
-      "learning_rate": 0.0002464622178606477,
-      "loss": 0.3303,
+      "learning_rate": 0.00026817201166180754,
+      "loss": 0.1132,
       "step": 4140
     },
     {
       "epoch": 5.96,
-      "learning_rate": 0.000246315014720314,
-      "loss": 0.2635,
+      "learning_rate": 0.00026808454810495625,
+      "loss": 0.0695,
       "step": 4150
     },
     {
       "epoch": 5.98,
-      "learning_rate": 0.00024616781157998035,
-      "loss": 0.3033,
+      "learning_rate": 0.0002679970845481049,
+      "loss": 0.0909,
       "step": 4160
     },
     {
       "epoch": 5.99,
-      "learning_rate": 0.0002460206084396467,
-      "loss": 0.2614,
+      "learning_rate": 0.00026790962099125363,
+      "loss": 0.0826,
       "step": 4170
     },
     {
       "epoch": 6.01,
-      "learning_rate": 0.00024587340529931307,
-      "loss": 0.2824,
+      "learning_rate": 0.0002678221574344023,
+      "loss": 0.0844,
       "step": 4180
     },
     {
       "epoch": 6.02,
-      "learning_rate": 0.0002457262021589794,
-      "loss": 0.247,
+      "learning_rate": 0.000267734693877551,
+      "loss": 0.0805,
       "step": 4190
     },
     {
       "epoch": 6.03,
-      "learning_rate": 0.0002455789990186457,
-      "loss": 0.2822,
+      "learning_rate": 0.0002676472303206997,
+      "loss": 0.0774,
       "step": 4200
     },
     {
       "epoch": 6.03,
-      "eval_loss": 0.2591544985771179,
-      "eval_runtime": 40.0116,
-      "eval_samples_per_second": 11.422,
-      "eval_steps_per_second": 1.45,
-      "eval_wer": 0.29920192742056917,
+      "eval_loss": 0.19473043084144592,
+      "eval_runtime": 39.8592,
+      "eval_samples_per_second": 11.465,
+      "eval_steps_per_second": 1.455,
+      "eval_wer": 0.18686944737238367,
       "step": 4200
     },
     {
       "epoch": 6.05,
-      "learning_rate": 0.00024543179587831204,
-      "loss": 0.2866,
+      "learning_rate": 0.0002675597667638484,
+      "loss": 0.0907,
       "step": 4210
     },
     {
       "epoch": 6.06,
-      "learning_rate": 0.0002452845927379784,
-      "loss": 0.24,
+      "learning_rate": 0.00026747230320699704,
+      "loss": 0.0737,
       "step": 4220
     },
     {
       "epoch": 6.08,
-      "learning_rate": 0.0002451373895976447,
-      "loss": 0.249,
+      "learning_rate": 0.00026738483965014576,
+      "loss": 0.0701,
       "step": 4230
     },
     {
       "epoch": 6.09,
-      "learning_rate": 0.00024499018645731107,
-      "loss": 0.2542,
+      "learning_rate": 0.0002672973760932944,
+      "loss": 0.0774,
       "step": 4240
     },
     {
       "epoch": 6.11,
-      "learning_rate": 0.00024484298331697743,
-      "loss": 0.2672,
+      "learning_rate": 0.00026720991253644313,
+      "loss": 0.075,
       "step": 4250
     },
     {
       "epoch": 6.12,
-      "learning_rate": 0.0002446957801766438,
-      "loss": 0.3306,
+      "learning_rate": 0.0002671224489795918,
+      "loss": 0.102,
       "step": 4260
     },
     {
       "epoch": 6.14,
-      "learning_rate": 0.0002445485770363101,
-      "loss": 0.2413,
+      "learning_rate": 0.0002670349854227405,
+      "loss": 0.0744,
       "step": 4270
     },
     {
       "epoch": 6.15,
-      "learning_rate": 0.0002444013738959764,
-      "loss": 0.2675,
+      "learning_rate": 0.0002669475218658892,
+      "loss": 0.085,
       "step": 4280
     },
     {
       "epoch": 6.16,
-      "learning_rate": 0.00024425417075564276,
-      "loss": 0.2492,
+      "learning_rate": 0.0002668600583090379,
+      "loss": 0.0779,
       "step": 4290
     },
     {
       "epoch": 6.18,
-      "learning_rate": 0.0002441069676153091,
-      "loss": 0.2343,
+      "learning_rate": 0.00026677259475218655,
+      "loss": 0.0697,
       "step": 4300
     },
     {
       "epoch": 6.18,
-      "eval_loss": 0.2688565254211426,
-      "eval_runtime": 39.6355,
-      "eval_samples_per_second": 11.53,
-      "eval_steps_per_second": 1.463,
-      "eval_wer": 0.294835115193495,
+      "eval_loss": 0.18682928383350372,
+      "eval_runtime": 39.9127,
+      "eval_samples_per_second": 11.45,
+      "eval_steps_per_second": 1.453,
+      "eval_wer": 0.18129799728956483,
       "step": 4300
     },
     {
       "epoch": 6.19,
-      "learning_rate": 0.00024395976447497543,
-      "loss": 0.3075,
+      "learning_rate": 0.00026668513119533526,
+      "loss": 0.1041,
       "step": 4310
     },
     {
       "epoch": 6.21,
-      "learning_rate": 0.0002438125613346418,
-      "loss": 0.2398,
+      "learning_rate": 0.0002665976676384839,
+      "loss": 0.0732,
       "step": 4320
     },
     {
       "epoch": 6.22,
-      "learning_rate": 0.00024366535819430812,
-      "loss": 0.2782,
+      "learning_rate": 0.00026651020408163264,
+      "loss": 0.0875,
       "step": 4330
     },
     {
       "epoch": 6.24,
-      "learning_rate": 0.00024351815505397446,
-      "loss": 0.2511,
+      "learning_rate": 0.0002664227405247813,
+      "loss": 0.0816,
       "step": 4340
     },
     {
       "epoch": 6.25,
-      "learning_rate": 0.0002433709519136408,
-      "loss": 0.234,
+      "learning_rate": 0.00026633527696793,
+      "loss": 0.0722,
       "step": 4350
     },
     {
       "epoch": 6.26,
-      "learning_rate": 0.00024322374877330715,
-      "loss": 0.2972,
+      "learning_rate": 0.00026624781341107873,
+      "loss": 0.0973,
       "step": 4360
     },
     {
       "epoch": 6.28,
-      "learning_rate": 0.00024307654563297348,
-      "loss": 0.2553,
+      "learning_rate": 0.0002661603498542274,
+      "loss": 0.0724,
       "step": 4370
     },
     {
       "epoch": 6.29,
-      "learning_rate": 0.00024292934249263982,
-      "loss": 0.2697,
+      "learning_rate": 0.0002660816326530612,
+      "loss": 0.0795,
       "step": 4380
     },
     {
       "epoch": 6.31,
-      "learning_rate": 0.00024278213935230615,
-      "loss": 0.2303,
+      "learning_rate": 0.0002659941690962099,
+      "loss": 0.0762,
       "step": 4390
     },
     {
       "epoch": 6.32,
-      "learning_rate": 0.0002426349362119725,
-      "loss": 0.2556,
+      "learning_rate": 0.0002659067055393586,
+      "loss": 0.0778,
       "step": 4400
     },
     {
       "epoch": 6.32,
-      "eval_loss": 0.2687687575817108,
-      "eval_runtime": 40.0355,
-      "eval_samples_per_second": 11.415,
-      "eval_steps_per_second": 1.449,
-      "eval_wer": 0.28896250564674,
+      "eval_loss": 0.1720825582742691,
+      "eval_runtime": 39.8776,
+      "eval_samples_per_second": 11.46,
+      "eval_steps_per_second": 1.454,
+      "eval_wer": 0.18114741755759675,
       "step": 4400
     },
     {
       "epoch": 6.34,
-      "learning_rate": 0.00024248773307163884,
-      "loss": 0.2581,
+      "learning_rate": 0.00026581924198250726,
+      "loss": 0.0782,
       "step": 4410
     },
     {
       "epoch": 6.35,
-      "learning_rate": 0.00024234052993130518,
-      "loss": 0.2807,
+      "learning_rate": 0.00026573177842565597,
+      "loss": 0.0822,
       "step": 4420
     },
     {
       "epoch": 6.36,
-      "learning_rate": 0.0002421933267909715,
-      "loss": 0.2925,
+      "learning_rate": 0.00026564431486880463,
+      "loss": 0.0935,
       "step": 4430
     },
     {
       "epoch": 6.38,
-      "learning_rate": 0.00024204612365063787,
-      "loss": 0.2417,
+      "learning_rate": 0.00026555685131195335,
+      "loss": 0.0818,
       "step": 4440
     },
     {
       "epoch": 6.39,
-      "learning_rate": 0.00024189892051030418,
-      "loss": 0.2539,
+      "learning_rate": 0.000265469387755102,
+      "loss": 0.0758,
       "step": 4450
     },
     {
       "epoch": 6.41,
-      "learning_rate": 0.00024175171736997054,
-      "loss": 0.2756,
+      "learning_rate": 0.00026538192419825067,
+      "loss": 0.0805,
       "step": 4460
     },
     {
       "epoch": 6.42,
-      "learning_rate": 0.00024160451422963687,
-      "loss": 0.223,
+      "learning_rate": 0.0002652944606413994,
+      "loss": 0.0667,
       "step": 4470
     },
     {
       "epoch": 6.44,
-      "learning_rate": 0.00024145731108930323,
-      "loss": 0.2717,
+      "learning_rate": 0.0002652069970845481,
+      "loss": 0.0728,
       "step": 4480
     },
     {
       "epoch": 6.45,
-      "learning_rate": 0.00024131010794896954,
-      "loss": 0.2731,
+      "learning_rate": 0.00026511953352769676,
+      "loss": 0.0813,
       "step": 4490
     },
     {
       "epoch": 6.47,
-      "learning_rate": 0.0002411629048086359,
-      "loss": 0.2437,
+      "learning_rate": 0.0002650320699708455,
+      "loss": 0.0771,
       "step": 4500
     },
     {
       "epoch": 6.47,
-      "eval_loss": 0.2555336654186249,
-      "eval_runtime": 40.1329,
-      "eval_samples_per_second": 11.387,
-      "eval_steps_per_second": 1.445,
-      "eval_wer": 0.28760728805902724,
+      "eval_loss": 0.1848333477973938,
+      "eval_runtime": 39.9888,
+      "eval_samples_per_second": 11.428,
+      "eval_steps_per_second": 1.45,
+      "eval_wer": 0.199216985393766,
       "step": 4500
     },
     {
       "epoch": 6.48,
-      "learning_rate": 0.00024101570166830223,
-      "loss": 0.2746,
+      "learning_rate": 0.00026494460641399414,
+      "loss": 0.0893,
       "step": 4510
     },
     {
       "epoch": 6.49,
-      "learning_rate": 0.0002408684985279686,
-      "loss": 0.2506,
+      "learning_rate": 0.00026485714285714285,
+      "loss": 0.083,
       "step": 4520
     },
     {
       "epoch": 6.51,
-      "learning_rate": 0.0002407212953876349,
-      "loss": 0.2604,
+      "learning_rate": 0.0002647696793002915,
+      "loss": 0.0837,
       "step": 4530
     },
     {
       "epoch": 6.52,
-      "learning_rate": 0.00024057409224730126,
-      "loss": 0.2634,
+      "learning_rate": 0.0002646822157434402,
+      "loss": 0.0827,
       "step": 4540
     },
     {
       "epoch": 6.54,
-      "learning_rate": 0.0002404268891069676,
-      "loss": 0.2561,
+      "learning_rate": 0.0002645947521865889,
+      "loss": 0.0708,
       "step": 4550
     },
     {
       "epoch": 6.55,
-      "learning_rate": 0.00024027968596663395,
-      "loss": 0.2648,
+      "learning_rate": 0.0002645072886297376,
+      "loss": 0.0849,
       "step": 4560
     },
     {
       "epoch": 6.57,
-      "learning_rate": 0.00024013248282630026,
-      "loss": 0.2446,
+      "learning_rate": 0.00026441982507288626,
+      "loss": 0.0738,
       "step": 4570
     },
     {
       "epoch": 6.58,
-      "learning_rate": 0.00023998527968596662,
-      "loss": 0.2685,
+      "learning_rate": 0.000264332361516035,
+      "loss": 0.0838,
       "step": 4580
     },
     {
       "epoch": 6.59,
-      "learning_rate": 0.00023983807654563295,
-      "loss": 0.2434,
+      "learning_rate": 0.00026424489795918364,
+      "loss": 0.0758,
       "step": 4590
     },
     {
       "epoch": 6.61,
-      "learning_rate": 0.0002396908734052993,
-      "loss": 0.237,
+      "learning_rate": 0.00026415743440233235,
+      "loss": 0.0717,
       "step": 4600
     },
     {
       "epoch": 6.61,
-      "eval_loss": 0.2497735321521759,
-      "eval_runtime": 40.0633,
-      "eval_samples_per_second": 11.407,
-      "eval_steps_per_second": 1.448,
-      "eval_wer": 0.28233699744014457,
+      "eval_loss": 0.17897100746631622,
+      "eval_runtime": 39.9583,
+      "eval_samples_per_second": 11.437,
+      "eval_steps_per_second": 1.452,
+      "eval_wer": 0.1919891582592983,
       "step": 4600
     },
     {
       "epoch": 6.62,
-      "learning_rate": 0.00023954367026496562,
-      "loss": 0.2712,
+      "learning_rate": 0.000264069970845481,
+      "loss": 0.0982,
       "step": 4610
     },
     {
       "epoch": 6.64,
-      "learning_rate": 0.00023939646712463198,
-      "loss": 0.2464,
+      "learning_rate": 0.00026398250728862973,
+      "loss": 0.09,
       "step": 4620
     },
     {
       "epoch": 6.65,
-      "learning_rate": 0.00023924926398429831,
-      "loss": 0.2903,
+      "learning_rate": 0.0002638950437317784,
+      "loss": 0.0911,
       "step": 4630
     },
     {
       "epoch": 6.67,
-      "learning_rate": 0.00023910206084396465,
-      "loss": 0.2472,
+      "learning_rate": 0.0002638075801749271,
+      "loss": 0.0833,
       "step": 4640
     },
     {
       "epoch": 6.68,
-      "learning_rate": 0.00023895485770363098,
-      "loss": 0.2602,
+      "learning_rate": 0.00026372011661807577,
+      "loss": 0.0892,
       "step": 4650
     },
     {
       "epoch": 6.7,
-      "learning_rate": 0.00023880765456329734,
-      "loss": 0.2728,
+      "learning_rate": 0.0002636326530612245,
+      "loss": 0.0936,
       "step": 4660
     },
     {
       "epoch": 6.71,
-      "learning_rate": 0.00023866045142296368,
-      "loss": 0.2636,
+      "learning_rate": 0.00026354518950437314,
+      "loss": 0.0812,
       "step": 4670
     },
     {
       "epoch": 6.72,
-      "learning_rate": 0.00023851324828263,
-      "loss": 0.2515,
+      "learning_rate": 0.00026345772594752186,
+      "loss": 0.0823,
       "step": 4680
     },
     {
       "epoch": 6.74,
-      "learning_rate": 0.00023836604514229634,
-      "loss": 0.2602,
+      "learning_rate": 0.0002633702623906705,
+      "loss": 0.0896,
       "step": 4690
     },
     {
       "epoch": 6.75,
-      "learning_rate": 0.0002382188420019627,
-      "loss": 0.2375,
+      "learning_rate": 0.00026328279883381923,
+      "loss": 0.0772,
       "step": 4700
     },
     {
       "epoch": 6.75,
-      "eval_loss": 0.2609308660030365,
-      "eval_runtime": 39.7988,
-      "eval_samples_per_second": 11.483,
-      "eval_steps_per_second": 1.457,
-      "eval_wer": 0.29363047733775033,
+      "eval_loss": 0.1973699927330017,
+      "eval_runtime": 39.8929,
+      "eval_samples_per_second": 11.456,
+      "eval_steps_per_second": 1.454,
+      "eval_wer": 0.19274205691913868,
       "step": 4700
     },
     {
       "epoch": 6.77,
-      "learning_rate": 0.00023807163886162904,
-      "loss": 0.2602,
+      "learning_rate": 0.00026319533527696795,
+      "loss": 0.0983,
       "step": 4710
     },
     {
       "epoch": 6.78,
-      "learning_rate": 0.00023792443572129537,
-      "loss": 0.2384,
+      "learning_rate": 0.00026310787172011656,
+      "loss": 0.0822,
       "step": 4720
     },
     {
       "epoch": 6.8,
-      "learning_rate": 0.0002377772325809617,
-      "loss": 0.2466,
+      "learning_rate": 0.00026302040816326527,
+      "loss": 0.0826,
       "step": 4730
     },
     {
       "epoch": 6.81,
-      "learning_rate": 0.00023763002944062806,
-      "loss": 0.2674,
+      "learning_rate": 0.000262932944606414,
+      "loss": 0.0938,
       "step": 4740
     },
     {
       "epoch": 6.82,
-      "learning_rate": 0.00023748282630029437,
-      "loss": 0.2497,
+      "learning_rate": 0.00026284548104956265,
+      "loss": 0.0796,
       "step": 4750
     },
     {
       "epoch": 6.84,
-      "learning_rate": 0.00023733562315996073,
-      "loss": 0.2771,
+      "learning_rate": 0.00026275801749271136,
+      "loss": 0.1045,
       "step": 4760
     },
     {
       "epoch": 6.85,
-      "learning_rate": 0.00023718842001962706,
-      "loss": 0.2276,
+      "learning_rate": 0.00026267055393586,
+      "loss": 0.0803,
       "step": 4770
     },
     {
       "epoch": 6.87,
-      "learning_rate": 0.00023704121687929342,
-      "loss": 0.2459,
+      "learning_rate": 0.00026258309037900874,
+      "loss": 0.0886,
       "step": 4780
     },
     {
       "epoch": 6.88,
-      "learning_rate": 0.00023689401373895973,
-      "loss": 0.2425,
+      "learning_rate": 0.0002624956268221574,
+      "loss": 0.0872,
       "step": 4790
     },
     {
       "epoch": 6.9,
-      "learning_rate": 0.0002367468105986261,
-      "loss": 0.2361,
+      "learning_rate": 0.0002624081632653061,
+      "loss": 0.0721,
       "step": 4800
     },
     {
       "epoch": 6.9,
-      "eval_loss": 0.2400684505701065,
-      "eval_runtime": 39.986,
-      "eval_samples_per_second": 11.429,
-      "eval_steps_per_second": 1.451,
-      "eval_wer": 0.28037946092455956,
+      "eval_loss": 0.19075419008731842,
+      "eval_runtime": 39.9033,
+      "eval_samples_per_second": 11.453,
+      "eval_steps_per_second": 1.454,
+      "eval_wer": 0.19364553531094714,
       "step": 4800
     },
     {
       "epoch": 6.91,
-      "learning_rate": 0.00023659960745829242,
-      "loss": 0.2473,
+      "learning_rate": 0.0002623206997084548,
+      "loss": 0.0853,
       "step": 4810
     },
     {
       "epoch": 6.93,
-      "learning_rate": 0.00023645240431795878,
-      "loss": 0.2144,
+      "learning_rate": 0.0002622332361516035,
+      "loss": 0.0657,
       "step": 4820
     },
     {
       "epoch": 6.94,
-      "learning_rate": 0.0002363052011776251,
-      "loss": 0.2641,
+      "learning_rate": 0.00026214577259475215,
+      "loss": 0.0864,
       "step": 4830
     },
     {
       "epoch": 6.95,
-      "learning_rate": 0.00023615799803729145,
-      "loss": 0.2559,
+      "learning_rate": 0.00026205830903790086,
+      "loss": 0.0914,
       "step": 4840
     },
     {
       "epoch": 6.97,
-      "learning_rate": 0.00023601079489695778,
-      "loss": 0.2404,
+      "learning_rate": 0.0002619708454810495,
+      "loss": 0.0767,
       "step": 4850
     },
     {
       "epoch": 6.98,
-      "learning_rate": 0.0002358635917566241,
-      "loss": 0.2533,
+      "learning_rate": 0.00026188338192419824,
+      "loss": 0.0872,
       "step": 4860
     },
     {
       "epoch": 7.0,
-      "learning_rate": 0.00023571638861629045,
-      "loss": 0.2301,
+      "learning_rate": 0.0002617959183673469,
+      "loss": 0.0734,
       "step": 4870
     },
     {
       "epoch": 7.01,
-      "learning_rate": 0.0002355691854759568,
-      "loss": 0.2553,
+      "learning_rate": 0.0002617084548104956,
+      "loss": 0.1009,
       "step": 4880
     },
     {
       "epoch": 7.03,
-      "learning_rate": 0.00023542198233562315,
-      "loss": 0.2102,
+      "learning_rate": 0.00026162099125364433,
+      "loss": 0.0684,
       "step": 4890
     },
     {
       "epoch": 7.04,
-      "learning_rate": 0.00023527477919528945,
-      "loss": 0.2266,
+      "learning_rate": 0.000261533527696793,
+      "loss": 0.0736,
       "step": 4900
     },
     {
       "epoch": 7.04,
-      "eval_loss": 0.2387649267911911,
-      "eval_runtime": 40.0231,
-      "eval_samples_per_second": 11.418,
-      "eval_steps_per_second": 1.449,
-      "eval_wer": 0.2708929378105707,
+      "eval_loss": 0.1875203400850296,
+      "eval_runtime": 39.8411,
+      "eval_samples_per_second": 11.471,
+      "eval_steps_per_second": 1.456,
+      "eval_wer": 0.18506249058876675,
       "step": 4900
     },
     {
       "epoch": 7.05,
-      "learning_rate": 0.0002351275760549558,
-      "loss": 0.2141,
+      "learning_rate": 0.00026144606413994165,
+      "loss": 0.0692,
       "step": 4910
     },
     {
       "epoch": 7.07,
-      "learning_rate": 0.00023498037291462217,
-      "loss": 0.2192,
+      "learning_rate": 0.00026135860058309037,
+      "loss": 0.0701,
       "step": 4920
     },
     {
       "epoch": 7.08,
-      "learning_rate": 0.0002348331697742885,
-      "loss": 0.2375,
+      "learning_rate": 0.00026127113702623903,
+      "loss": 0.0786,
       "step": 4930
     },
     {
       "epoch": 7.1,
-      "learning_rate": 0.0002346859666339548,
-      "loss": 0.2104,
+      "learning_rate": 0.00026118367346938774,
+      "loss": 0.0796,
       "step": 4940
     },
     {
       "epoch": 7.11,
-      "learning_rate": 0.00023453876349362117,
-      "loss": 0.2054,
+      "learning_rate": 0.0002610962099125364,
+      "loss": 0.0654,
       "step": 4950
     },
     {
       "epoch": 7.13,
-      "learning_rate": 0.00023439156035328753,
-      "loss": 0.2081,
+      "learning_rate": 0.0002610087463556851,
+      "loss": 0.0765,
       "step": 4960
     },
     {
       "epoch": 7.14,
-      "learning_rate": 0.00023424435721295387,
-      "loss": 0.2141,
+      "learning_rate": 0.00026092128279883383,
+      "loss": 0.0696,
       "step": 4970
     },
     {
       "epoch": 7.16,
-      "learning_rate": 0.00023409715407262017,
-      "loss": 0.2106,
+      "learning_rate": 0.0002608338192419825,
+      "loss": 0.0723,
       "step": 4980
     },
     {
       "epoch": 7.17,
-      "learning_rate": 0.00023394995093228653,
-      "loss": 0.1952,
+      "learning_rate": 0.00026074635568513116,
+      "loss": 0.0664,
       "step": 4990
     },
     {
       "epoch": 7.18,
-      "learning_rate": 0.0002338027477919529,
-      "loss": 0.2299,
+      "learning_rate": 0.00026065889212827987,
+      "loss": 0.0779,
       "step": 5000
     },
     {
       "epoch": 7.18,
-      "eval_loss": 0.24284440279006958,
-      "eval_runtime": 39.8933,
-      "eval_samples_per_second": 11.456,
-      "eval_steps_per_second": 1.454,
-      "eval_wer": 0.2772172865532299,
+      "eval_loss": 0.182932049036026,
+      "eval_runtime": 39.8597,
+      "eval_samples_per_second": 11.465,
+      "eval_steps_per_second": 1.455,
+      "eval_wer": 0.18325553380514983,
       "step": 5000
     },
     {
       "epoch": 7.2,
-      "learning_rate": 0.0002336555446516192,
-      "loss": 0.2312,
+      "learning_rate": 0.00026057142857142853,
+      "loss": 0.0822,
       "step": 5010
     },
     {
       "epoch": 7.21,
-      "learning_rate": 0.00023350834151128553,
-      "loss": 0.2173,
+      "learning_rate": 0.00026048396501457725,
+      "loss": 0.0723,
       "step": 5020
     },
     {
       "epoch": 7.23,
-      "learning_rate": 0.0002333611383709519,
-      "loss": 0.2383,
+      "learning_rate": 0.0002603965014577259,
+      "loss": 0.0802,
       "step": 5030
     },
     {
       "epoch": 7.24,
-      "learning_rate": 0.00023321393523061825,
-      "loss": 0.2071,
+      "learning_rate": 0.0002603090379008746,
+      "loss": 0.0725,
       "step": 5040
     },
     {
       "epoch": 7.26,
-      "learning_rate": 0.00023306673209028456,
-      "loss": 0.2402,
+      "learning_rate": 0.0002602215743440233,
+      "loss": 0.083,
       "step": 5050
     },
     {
       "epoch": 7.27,
-      "learning_rate": 0.0002329195289499509,
-      "loss": 0.2559,
+      "learning_rate": 0.000260134110787172,
+      "loss": 0.0941,
       "step": 5060
     },
     {
       "epoch": 7.28,
-      "learning_rate": 0.00023277232580961725,
-      "loss": 0.2094,
+      "learning_rate": 0.0002600466472303207,
+      "loss": 0.0617,
       "step": 5070
     },
     {
       "epoch": 7.3,
-      "learning_rate": 0.00023262512266928362,
-      "loss": 0.2216,
+      "learning_rate": 0.0002599591836734694,
+      "loss": 0.07,
       "step": 5080
     },
     {
       "epoch": 7.31,
-      "learning_rate": 0.00023247791952894992,
-      "loss": 0.2184,
+      "learning_rate": 0.00025987172011661803,
+      "loss": 0.0673,
       "step": 5090
     },
     {
       "epoch": 7.33,
-      "learning_rate": 0.00023233071638861625,
-      "loss": 0.2322,
+      "learning_rate": 0.00025978425655976675,
+      "loss": 0.0738,
       "step": 5100
     },
     {
       "epoch": 7.33,
-      "eval_loss": 0.24409331381320953,
-      "eval_runtime": 40.158,
-      "eval_samples_per_second": 11.38,
-      "eval_steps_per_second": 1.444,
-      "eval_wer": 0.27450685137780456,
+      "eval_loss": 0.1895502805709839,
+      "eval_runtime": 39.9633,
+      "eval_samples_per_second": 11.435,
+      "eval_steps_per_second": 1.451,
+      "eval_wer": 0.18521307032073484,
       "step": 5100
     },
     {
       "epoch": 7.34,
-      "learning_rate": 0.00023218351324828262,
-      "loss": 0.2169,
+      "learning_rate": 0.0002596967930029154,
+      "loss": 0.0739,
       "step": 5110
     },
     {
       "epoch": 7.36,
-      "learning_rate": 0.00023203631010794898,
-      "loss": 0.243,
+      "learning_rate": 0.0002596093294460641,
+      "loss": 0.0779,
       "step": 5120
     },
     {
       "epoch": 7.37,
-      "learning_rate": 0.00023188910696761528,
-      "loss": 0.2499,
+      "learning_rate": 0.0002595218658892128,
+      "loss": 0.0878,
       "step": 5130
     },
     {
       "epoch": 7.39,
-      "learning_rate": 0.00023174190382728162,
-      "loss": 0.2151,
+      "learning_rate": 0.0002594344023323615,
+      "loss": 0.0739,
       "step": 5140
     },
     {
       "epoch": 7.4,
-      "learning_rate": 0.00023159470068694798,
-      "loss": 0.2384,
+      "learning_rate": 0.0002593469387755102,
+      "loss": 0.0836,
       "step": 5150
     },
     {
       "epoch": 7.41,
-      "learning_rate": 0.00023144749754661428,
-      "loss": 0.2193,
+      "learning_rate": 0.0002592594752186589,
+      "loss": 0.073,
       "step": 5160
     },
     {
       "epoch": 7.43,
-      "learning_rate": 0.00023130029440628064,
-      "loss": 0.214,
+      "learning_rate": 0.00025917201166180754,
+      "loss": 0.064,
       "step": 5170
     },
     {
       "epoch": 7.44,
-      "learning_rate": 0.00023115309126594698,
-      "loss": 0.2394,
+      "learning_rate": 0.00025908454810495625,
+      "loss": 0.0852,
       "step": 5180
     },
     {
       "epoch": 7.46,
-      "learning_rate": 0.00023100588812561334,
-      "loss": 0.1871,
+      "learning_rate": 0.0002589970845481049,
+      "loss": 0.0673,
       "step": 5190
     },
     {
       "epoch": 7.47,
-      "learning_rate": 0.00023085868498527964,
-      "loss": 0.214,
+      "learning_rate": 0.00025890962099125363,
+      "loss": 0.0799,
       "step": 5200
     },
     {
       "epoch": 7.47,
-      "eval_loss": 0.24094292521476746,
-      "eval_runtime": 40.0199,
-      "eval_samples_per_second": 11.419,
-      "eval_steps_per_second": 1.449,
-      "eval_wer": 0.2722481553982834,
+      "eval_loss": 0.1826663762331009,
+      "eval_runtime": 39.9859,
+      "eval_samples_per_second": 11.429,
+      "eval_steps_per_second": 1.451,
+      "eval_wer": 0.19244089745520254,
       "step": 5200
     },
     {
       "epoch": 7.49,
-      "learning_rate": 0.000230711481844946,
-      "loss": 0.2096,
+      "learning_rate": 0.0002588221574344023,
+      "loss": 0.0698,
       "step": 5210
     },
     {
       "epoch": 7.5,
-      "learning_rate": 0.00023056427870461234,
-      "loss": 0.2079,
+      "learning_rate": 0.000258734693877551,
+      "loss": 0.0658,
       "step": 5220
     },
     {
       "epoch": 7.51,
-      "learning_rate": 0.0002304170755642787,
-      "loss": 0.233,
+      "learning_rate": 0.0002586472303206997,
+      "loss": 0.081,
       "step": 5230
     },
     {
       "epoch": 7.53,
-      "learning_rate": 0.000230269872423945,
-      "loss": 0.244,
+      "learning_rate": 0.0002585597667638484,
+      "loss": 0.0755,
       "step": 5240
     },
     {
       "epoch": 7.54,
-      "learning_rate": 0.00023012266928361136,
-      "loss": 0.2502,
+      "learning_rate": 0.0002584723032069971,
+      "loss": 0.0824,
       "step": 5250
     },
     {
       "epoch": 7.56,
-      "learning_rate": 0.0002299754661432777,
-      "loss": 0.2079,
+      "learning_rate": 0.00025838483965014576,
+      "loss": 0.0666,
       "step": 5260
     },
     {
       "epoch": 7.57,
-      "learning_rate": 0.00022982826300294406,
-      "loss": 0.2067,
+      "learning_rate": 0.0002582973760932944,
+      "loss": 0.0679,
       "step": 5270
     },
     {
       "epoch": 7.59,
-      "learning_rate": 0.00022968105986261036,
-      "loss": 0.2274,
+      "learning_rate": 0.00025820991253644313,
+      "loss": 0.0774,
       "step": 5280
     },
     {
       "epoch": 7.6,
-      "learning_rate": 0.00022953385672227672,
-      "loss": 0.2019,
+      "learning_rate": 0.0002581224489795918,
+      "loss": 0.0699,
       "step": 5290
     },
     {
       "epoch": 7.61,
-      "learning_rate": 0.00022938665358194306,
-      "loss": 0.2223,
+      "learning_rate": 0.0002580349854227405,
+      "loss": 0.0682,
       "step": 5300
     },
     {
       "epoch": 7.61,
-      "eval_loss": 0.24156498908996582,
-      "eval_runtime": 39.8819,
-      "eval_samples_per_second": 11.459,
-      "eval_steps_per_second": 1.454,
-      "eval_wer": 0.26923656075892183,
+      "eval_loss": 0.1933245211839676,
+      "eval_runtime": 40.0463,
+      "eval_samples_per_second": 11.412,
+      "eval_steps_per_second": 1.448,
+      "eval_wer": 0.19575365155850022,
       "step": 5300
     },
     {
       "epoch": 7.63,
-      "learning_rate": 0.0002292394504416094,
-      "loss": 0.2297,
+      "learning_rate": 0.00025794752186588917,
+      "loss": 0.0878,
       "step": 5310
     },
     {
       "epoch": 7.64,
-      "learning_rate": 0.00022909224730127573,
-      "loss": 0.1978,
+      "learning_rate": 0.0002578600583090379,
+      "loss": 0.0668,
       "step": 5320
     },
     {
       "epoch": 7.66,
-      "learning_rate": 0.00022894504416094209,
-      "loss": 0.2437,
+      "learning_rate": 0.0002577725947521866,
+      "loss": 0.0886,
       "step": 5330
     },
     {
       "epoch": 7.67,
-      "learning_rate": 0.00022879784102060842,
-      "loss": 0.2047,
+      "learning_rate": 0.00025768513119533526,
+      "loss": 0.0625,
       "step": 5340
     },
     {
       "epoch": 7.69,
-      "learning_rate": 0.00022865063788027475,
-      "loss": 0.2329,
+      "learning_rate": 0.0002575976676384839,
+      "loss": 0.0748,
       "step": 5350
     },
     {
       "epoch": 7.7,
-      "learning_rate": 0.00022850343473994109,
-      "loss": 0.2225,
+      "learning_rate": 0.00025751020408163264,
+      "loss": 0.0811,
       "step": 5360
     },
     {
       "epoch": 7.72,
-      "learning_rate": 0.00022835623159960745,
-      "loss": 0.2198,
+      "learning_rate": 0.0002574227405247813,
+      "loss": 0.0765,
       "step": 5370
     },
     {
       "epoch": 7.73,
-      "learning_rate": 0.00022820902845927378,
-      "loss": 0.2289,
+      "learning_rate": 0.00025733527696793,
+      "loss": 0.0797,
       "step": 5380
     },
     {
       "epoch": 7.74,
-      "learning_rate": 0.0002280618253189401,
-      "loss": 0.2167,
+      "learning_rate": 0.00025724781341107867,
+      "loss": 0.0728,
       "step": 5390
     },
     {
       "epoch": 7.76,
-      "learning_rate": 0.00022791462217860645,
-      "loss": 0.2201,
+      "learning_rate": 0.0002571603498542274,
+      "loss": 0.0702,
       "step": 5400
     },
     {
       "epoch": 7.76,
-      "eval_loss": 0.22141778469085693,
-      "eval_runtime": 39.8574,
-      "eval_samples_per_second": 11.466,
-      "eval_steps_per_second": 1.455,
-      "eval_wer": 0.2546303267580184,
+      "eval_loss": 0.16960883140563965,
+      "eval_runtime": 40.0131,
+      "eval_samples_per_second": 11.421,
+      "eval_steps_per_second": 1.45,
+      "eval_wer": 0.17723234452642675,
       "step": 5400
     },
     {
       "epoch": 7.77,
-      "learning_rate": 0.0002277674190382728,
-      "loss": 0.2066,
+      "learning_rate": 0.0002570728862973761,
+      "loss": 0.0755,
       "step": 5410
     },
     {
       "epoch": 7.79,
-      "learning_rate": 0.00022762021589793914,
-      "loss": 0.2034,
+      "learning_rate": 0.00025698542274052476,
+      "loss": 0.0628,
       "step": 5420
     },
     {
       "epoch": 7.8,
-      "learning_rate": 0.00022747301275760547,
-      "loss": 0.2038,
+      "learning_rate": 0.0002568979591836735,
+      "loss": 0.0693,
       "step": 5430
     },
     {
       "epoch": 7.82,
-      "learning_rate": 0.0002273258096172718,
-      "loss": 0.2034,
+      "learning_rate": 0.00025681049562682214,
+      "loss": 0.0673,
       "step": 5440
     },
     {
       "epoch": 7.83,
-      "learning_rate": 0.00022717860647693817,
-      "loss": 0.2328,
+      "learning_rate": 0.0002567230320699708,
+      "loss": 0.0816,
       "step": 5450
     },
     {
       "epoch": 7.84,
-      "learning_rate": 0.00022703140333660447,
-      "loss": 0.1861,
+      "learning_rate": 0.0002566355685131195,
+      "loss": 0.0656,
       "step": 5460
     },
     {
       "epoch": 7.86,
-      "learning_rate": 0.00022688420019627083,
-      "loss": 0.2146,
+      "learning_rate": 0.0002565481049562682,
+      "loss": 0.0686,
       "step": 5470
     },
     {
       "epoch": 7.87,
-      "learning_rate": 0.00022673699705593717,
-      "loss": 0.2463,
+      "learning_rate": 0.0002564606413994169,
+      "loss": 0.0899,
       "step": 5480
     },
     {
       "epoch": 7.89,
-      "learning_rate": 0.00022658979391560353,
-      "loss": 0.1953,
+      "learning_rate": 0.0002563731778425656,
+      "loss": 0.0673,
       "step": 5490
     },
     {
       "epoch": 7.9,
-      "learning_rate": 0.00022644259077526983,
-      "loss": 0.2225,
+      "learning_rate": 0.00025628571428571427,
+      "loss": 0.0784,
       "step": 5500
     },
     {
       "epoch": 7.9,
-      "eval_loss": 0.22256095707416534,
-      "eval_runtime": 39.882,
-      "eval_samples_per_second": 11.459,
-      "eval_steps_per_second": 1.454,
-      "eval_wer": 0.2653214877277518,
+      "eval_loss": 0.17280295491218567,
+      "eval_runtime": 39.9805,
+      "eval_samples_per_second": 11.431,
+      "eval_steps_per_second": 1.451,
+      "eval_wer": 0.1846107513928625,
       "step": 5500
     },
     {
       "epoch": 7.92,
-      "learning_rate": 0.0002262953876349362,
-      "loss": 0.2189,
+      "learning_rate": 0.000256198250728863,
+      "loss": 0.0795,
       "step": 5510
     },
     {
       "epoch": 7.93,
-      "learning_rate": 0.00022614818449460253,
-      "loss": 0.1998,
+      "learning_rate": 0.00025611078717201164,
+      "loss": 0.0675,
       "step": 5520
     },
     {
       "epoch": 7.95,
-      "learning_rate": 0.0002260009813542689,
-      "loss": 0.2587,
+      "learning_rate": 0.0002560233236151603,
+      "loss": 0.0959,
       "step": 5530
     },
     {
       "epoch": 7.96,
-      "learning_rate": 0.0002258537782139352,
-      "loss": 0.2029,
+      "learning_rate": 0.000255935860058309,
+      "loss": 0.0652,
       "step": 5540
     },
     {
       "epoch": 7.97,
-      "learning_rate": 0.00022570657507360156,
-      "loss": 0.1974,
+      "learning_rate": 0.0002558483965014577,
+      "loss": 0.0707,
       "step": 5550
     },
     {
       "epoch": 7.99,
-      "learning_rate": 0.0002255593719332679,
-      "loss": 0.2219,
+      "learning_rate": 0.0002557609329446064,
+      "loss": 0.0832,
       "step": 5560
     },
     {
       "epoch": 8.0,
-      "learning_rate": 0.00022541216879293425,
-      "loss": 0.2126,
+      "learning_rate": 0.0002556734693877551,
+      "loss": 0.0724,
       "step": 5570
     },
     {
       "epoch": 8.02,
-      "learning_rate": 0.00022526496565260056,
-      "loss": 0.2219,
+      "learning_rate": 0.00025558600583090377,
+      "loss": 0.0819,
       "step": 5580
     },
     {
       "epoch": 8.03,
-      "learning_rate": 0.00022511776251226692,
-      "loss": 0.1896,
+      "learning_rate": 0.0002554985422740525,
+      "loss": 0.0677,
       "step": 5590
     },
     {
       "epoch": 8.05,
-      "learning_rate": 0.00022497055937193325,
-      "loss": 0.2159,
+      "learning_rate": 0.00025541107871720114,
+      "loss": 0.0778,
       "step": 5600
     },
     {
       "epoch": 8.05,
-      "eval_loss": 0.21107521653175354,
-      "eval_runtime": 40.0872,
-      "eval_samples_per_second": 11.4,
-      "eval_steps_per_second": 1.447,
-      "eval_wer": 0.24318626712844452,
+      "eval_loss": 0.1783214509487152,
+      "eval_runtime": 39.8325,
+      "eval_samples_per_second": 11.473,
+      "eval_steps_per_second": 1.456,
+      "eval_wer": 0.1808462580936606,
       "step": 5600
     },
     {
       "epoch": 8.06,
-      "learning_rate": 0.00022482335623159958,
-      "loss": 0.1805,
+      "learning_rate": 0.00025532361516034986,
+      "loss": 0.0661,
       "step": 5610
     },
     {
       "epoch": 8.07,
-      "learning_rate": 0.00022467615309126592,
-      "loss": 0.1778,
+      "learning_rate": 0.0002552361516034985,
+      "loss": 0.0588,
       "step": 5620
     },
     {
       "epoch": 8.09,
-      "learning_rate": 0.00022452894995093228,
-      "loss": 0.1797,
+      "learning_rate": 0.0002551486880466472,
+      "loss": 0.0762,
       "step": 5630
     },
     {
       "epoch": 8.1,
-      "learning_rate": 0.0002243817468105986,
-      "loss": 0.1686,
+      "learning_rate": 0.0002550612244897959,
+      "loss": 0.0714,
       "step": 5640
     },
     {
       "epoch": 8.12,
-      "learning_rate": 0.00022423454367026494,
-      "loss": 0.1895,
+      "learning_rate": 0.00025497376093294456,
+      "loss": 0.0757,
       "step": 5650
     },
     {
       "epoch": 8.13,
-      "learning_rate": 0.00022408734052993128,
-      "loss": 0.1648,
+      "learning_rate": 0.00025488629737609327,
+      "loss": 0.0614,
       "step": 5660
     },
     {
       "epoch": 8.15,
-      "learning_rate": 0.00022394013738959764,
-      "loss": 0.2194,
+      "learning_rate": 0.000254798833819242,
+      "loss": 0.0774,
       "step": 5670
     },
     {
       "epoch": 8.16,
-      "learning_rate": 0.00022379293424926397,
-      "loss": 0.1901,
+      "learning_rate": 0.00025471137026239065,
+      "loss": 0.0766,
       "step": 5680
     },
     {
       "epoch": 8.18,
-      "learning_rate": 0.0002236457311089303,
-      "loss": 0.1913,
+      "learning_rate": 0.00025462390670553936,
+      "loss": 0.0722,
       "step": 5690
     },
     {
       "epoch": 8.19,
-      "learning_rate": 0.00022349852796859664,
-      "loss": 0.2291,
+      "learning_rate": 0.000254536443148688,
+      "loss": 0.0906,
       "step": 5700
     },
     {
       "epoch": 8.19,
-      "eval_loss": 0.20670226216316223,
-      "eval_runtime": 40.0404,
-      "eval_samples_per_second": 11.413,
-      "eval_steps_per_second": 1.449,
-      "eval_wer": 0.24680018069567836,
+      "eval_loss": 0.1898173987865448,
+      "eval_runtime": 40.2295,
+      "eval_samples_per_second": 11.36,
+      "eval_steps_per_second": 1.442,
+      "eval_wer": 0.197259448878181,
       "step": 5700
     },
     {
       "epoch": 8.2,
-      "learning_rate": 0.000223351324828263,
-      "loss": 0.1716,
+      "learning_rate": 0.0002544489795918367,
+      "loss": 0.0685,
       "step": 5710
     },
     {
       "epoch": 8.22,
-      "learning_rate": 0.00022320412168792933,
-      "loss": 0.1894,
+      "learning_rate": 0.0002543615160349854,
+      "loss": 0.0812,
       "step": 5720
     },
     {
       "epoch": 8.23,
-      "learning_rate": 0.00022305691854759567,
-      "loss": 0.2024,
+      "learning_rate": 0.00025427405247813406,
+      "loss": 0.0798,
       "step": 5730
     },
     {
       "epoch": 8.25,
-      "learning_rate": 0.000222909715407262,
-      "loss": 0.1778,
+      "learning_rate": 0.0002541865889212828,
+      "loss": 0.0655,
       "step": 5740
     },
     {
       "epoch": 8.26,
-      "learning_rate": 0.00022276251226692836,
-      "loss": 0.2244,
+      "learning_rate": 0.0002540991253644315,
+      "loss": 0.0838,
       "step": 5750
     },
     {
       "epoch": 8.28,
-      "learning_rate": 0.00022261530912659467,
-      "loss": 0.2153,
+      "learning_rate": 0.00025401166180758015,
+      "loss": 0.0831,
       "step": 5760
     },
     {
       "epoch": 8.29,
-      "learning_rate": 0.00022246810598626103,
-      "loss": 0.2085,
+      "learning_rate": 0.00025392419825072887,
+      "loss": 0.0761,
       "step": 5770
     },
     {
       "epoch": 8.3,
-      "learning_rate": 0.00022232090284592736,
-      "loss": 0.1727,
+      "learning_rate": 0.0002538367346938775,
+      "loss": 0.0653,
       "step": 5780
     },
     {
       "epoch": 8.32,
-      "learning_rate": 0.00022217369970559372,
-      "loss": 0.1769,
+      "learning_rate": 0.00025374927113702624,
+      "loss": 0.0701,
       "step": 5790
     },
     {
       "epoch": 8.33,
-      "learning_rate": 0.00022202649656526003,
-      "loss": 0.2067,
+      "learning_rate": 0.0002536618075801749,
+      "loss": 0.0842,
       "step": 5800
     },
     {
       "epoch": 8.33,
-      "eval_loss": 0.2133297473192215,
-      "eval_runtime": 40.2177,
-      "eval_samples_per_second": 11.363,
-      "eval_steps_per_second": 1.442,
-      "eval_wer": 0.23701249811775335,
+      "eval_loss": 0.17102932929992676,
+      "eval_runtime": 40.0617,
+      "eval_samples_per_second": 11.407,
+      "eval_steps_per_second": 1.448,
+      "eval_wer": 0.17934046077397983,
       "step": 5800
     },
     {
       "epoch": 8.35,
-      "learning_rate": 0.00022187929342492639,
-      "loss": 0.1915,
+      "learning_rate": 0.00025357434402332356,
+      "loss": 0.0707,
       "step": 5810
     },
     {
       "epoch": 8.36,
-      "learning_rate": 0.00022173209028459272,
-      "loss": 0.1666,
+      "learning_rate": 0.0002534868804664723,
+      "loss": 0.0625,
       "step": 5820
     },
     {
       "epoch": 8.38,
-      "learning_rate": 0.00022158488714425908,
-      "loss": 0.2038,
+      "learning_rate": 0.000253399416909621,
+      "loss": 0.0818,
       "step": 5830
     },
     {
       "epoch": 8.39,
-      "learning_rate": 0.0002214376840039254,
-      "loss": 0.1724,
+      "learning_rate": 0.00025331195335276965,
+      "loss": 0.0644,
       "step": 5840
     },
     {
       "epoch": 8.41,
-      "learning_rate": 0.00022129048086359175,
-      "loss": 0.2457,
+      "learning_rate": 0.00025322448979591837,
+      "loss": 0.1082,
       "step": 5850
     },
     {
       "epoch": 8.42,
-      "learning_rate": 0.00022114327772325808,
-      "loss": 0.1776,
+      "learning_rate": 0.00025313702623906703,
+      "loss": 0.0661,
       "step": 5860
     },
     {
       "epoch": 8.43,
-      "learning_rate": 0.00022099607458292444,
-      "loss": 0.1952,
+      "learning_rate": 0.00025304956268221574,
+      "loss": 0.0726,
       "step": 5870
     },
     {
       "epoch": 8.45,
-      "learning_rate": 0.00022084887144259075,
-      "loss": 0.2253,
+      "learning_rate": 0.0002529620991253644,
+      "loss": 0.0922,
       "step": 5880
     },
     {
       "epoch": 8.46,
-      "learning_rate": 0.0002207016683022571,
-      "loss": 0.1772,
+      "learning_rate": 0.00025287463556851307,
+      "loss": 0.0689,
       "step": 5890
     },
     {
       "epoch": 8.48,
-      "learning_rate": 0.00022055446516192344,
-      "loss": 0.1882,
+      "learning_rate": 0.0002527871720116618,
+      "loss": 0.0701,
       "step": 5900
     },
     {
       "epoch": 8.48,
-      "eval_loss": 0.21073894202709198,
-      "eval_runtime": 39.9763,
-      "eval_samples_per_second": 11.432,
-      "eval_steps_per_second": 1.451,
-      "eval_wer": 0.23520554133413643,
+      "eval_loss": 0.18022069334983826,
+      "eval_runtime": 39.8801,
+      "eval_samples_per_second": 11.459,
+      "eval_steps_per_second": 1.454,
+      "eval_wer": 0.17918988104201175,
       "step": 5900
     },
     {
       "epoch": 8.49,
-      "learning_rate": 0.00022040726202158975,
-      "loss": 0.1848,
+      "learning_rate": 0.00025269970845481044,
+      "loss": 0.0709,
       "step": 5910
     },
     {
       "epoch": 8.51,
-      "learning_rate": 0.0002202600588812561,
-      "loss": 0.1788,
+      "learning_rate": 0.00025261224489795916,
+      "loss": 0.0679,
       "step": 5920
     },
     {
       "epoch": 8.52,
-      "learning_rate": 0.00022011285574092247,
-      "loss": 0.2083,
+      "learning_rate": 0.00025252478134110787,
+      "loss": 0.0898,
       "step": 5930
     },
     {
       "epoch": 8.53,
-      "learning_rate": 0.0002199656526005888,
-      "loss": 0.206,
+      "learning_rate": 0.00025243731778425653,
+      "loss": 0.0816,
       "step": 5940
     },
     {
       "epoch": 8.55,
-      "learning_rate": 0.0002198184494602551,
-      "loss": 0.2169,
+      "learning_rate": 0.00025234985422740525,
+      "loss": 0.0832,
       "step": 5950
     },
     {
       "epoch": 8.56,
-      "learning_rate": 0.00021967124631992147,
-      "loss": 0.1949,
+      "learning_rate": 0.0002522623906705539,
+      "loss": 0.0701,
       "step": 5960
     },
     {
       "epoch": 8.58,
-      "learning_rate": 0.00021952404317958783,
-      "loss": 0.1824,
+      "learning_rate": 0.0002521749271137026,
+      "loss": 0.0605,
       "step": 5970
     },
     {
       "epoch": 8.59,
-      "learning_rate": 0.00021937684003925416,
-      "loss": 0.2049,
+      "learning_rate": 0.0002520874635568513,
+      "loss": 0.0796,
       "step": 5980
     },
     {
       "epoch": 8.61,
-      "learning_rate": 0.00021922963689892047,
-      "loss": 0.166,
+      "learning_rate": 0.00025199999999999995,
+      "loss": 0.0542,
       "step": 5990
     },
     {
       "epoch": 8.62,
-      "learning_rate": 0.00021908243375858683,
-      "loss": 0.2055,
+      "learning_rate": 0.00025191253644314866,
+      "loss": 0.0758,
       "step": 6000
     },
     {
       "epoch": 8.62,
-      "eval_loss": 0.21278780698776245,
-      "eval_runtime": 39.9574,
-      "eval_samples_per_second": 11.437,
-      "eval_steps_per_second": 1.452,
-      "eval_wer": 0.2550820659539226,
+      "eval_loss": 0.18393321335315704,
+      "eval_runtime": 39.9261,
+      "eval_samples_per_second": 11.446,
+      "eval_steps_per_second": 1.453,
+      "eval_wer": 0.18099683782562867,
       "step": 6000
     },
     {
       "epoch": 8.64,
-      "learning_rate": 0.0002189352306182532,
-      "loss": 0.1874,
+      "learning_rate": 0.0002518250728862974,
+      "loss": 0.075,
       "step": 6010
     },
     {
       "epoch": 8.65,
-      "learning_rate": 0.00021878802747791952,
-      "loss": 0.1799,
+      "learning_rate": 0.00025173760932944604,
+      "loss": 0.0671,
       "step": 6020
     },
     {
       "epoch": 8.66,
-      "learning_rate": 0.00021864082433758583,
-      "loss": 0.2084,
+      "learning_rate": 0.00025165014577259475,
+      "loss": 0.0867,
       "step": 6030
     },
     {
       "epoch": 8.68,
-      "learning_rate": 0.0002184936211972522,
-      "loss": 0.207,
+      "learning_rate": 0.0002515626822157434,
+      "loss": 0.0804,
       "step": 6040
     },
     {
       "epoch": 8.69,
-      "learning_rate": 0.00021834641805691855,
-      "loss": 0.215,
+      "learning_rate": 0.0002514752186588921,
+      "loss": 0.0806,
       "step": 6050
     },
     {
       "epoch": 8.71,
-      "learning_rate": 0.00021819921491658486,
-      "loss": 0.1749,
+      "learning_rate": 0.0002513877551020408,
+      "loss": 0.0708,
       "step": 6060
     },
     {
       "epoch": 8.72,
-      "learning_rate": 0.0002180520117762512,
-      "loss": 0.1796,
+      "learning_rate": 0.00025130029154518945,
+      "loss": 0.0759,
       "step": 6070
     },
     {
       "epoch": 8.74,
-      "learning_rate": 0.00021790480863591755,
-      "loss": 0.1734,
+      "learning_rate": 0.00025121282798833816,
+      "loss": 0.0626,
       "step": 6080
     },
     {
       "epoch": 8.75,
-      "learning_rate": 0.0002177576054955839,
-      "loss": 0.1894,
+      "learning_rate": 0.0002511253644314869,
+      "loss": 0.0645,
       "step": 6090
     },
     {
       "epoch": 8.76,
-      "learning_rate": 0.00021761040235525022,
-      "loss": 0.2144,
+      "learning_rate": 0.00025103790087463554,
+      "loss": 0.0751,
       "step": 6100
     },
     {
       "epoch": 8.76,
-      "eval_loss": 0.20413599908351898,
-      "eval_runtime": 40.0035,
-      "eval_samples_per_second": 11.424,
-      "eval_steps_per_second": 1.45,
-      "eval_wer": 0.2302364101791899,
+      "eval_loss": 0.18380357325077057,
+      "eval_runtime": 39.9599,
+      "eval_samples_per_second": 11.436,
+      "eval_steps_per_second": 1.451,
+      "eval_wer": 0.1840084324649902,
       "step": 6100
     },
     {
       "epoch": 8.78,
-      "learning_rate": 0.00021746319921491655,
-      "loss": 0.2061,
+      "learning_rate": 0.00025095043731778425,
+      "loss": 0.0868,
       "step": 6110
     },
     {
       "epoch": 8.79,
-      "learning_rate": 0.0002173159960745829,
-      "loss": 0.1862,
+      "learning_rate": 0.0002508629737609329,
+      "loss": 0.07,
       "step": 6120
     },
     {
       "epoch": 8.81,
-      "learning_rate": 0.00021716879293424927,
-      "loss": 0.2183,
+      "learning_rate": 0.00025077551020408163,
+      "loss": 0.0913,
       "step": 6130
     },
     {
       "epoch": 8.82,
-      "learning_rate": 0.00021702158979391558,
-      "loss": 0.1819,
+      "learning_rate": 0.0002506880466472303,
+      "loss": 0.0712,
       "step": 6140
     },
     {
       "epoch": 8.84,
-      "learning_rate": 0.0002168743866535819,
-      "loss": 0.2072,
+      "learning_rate": 0.00025060058309037895,
+      "loss": 0.0836,
       "step": 6150
     },
     {
       "epoch": 8.85,
-      "learning_rate": 0.00021672718351324827,
-      "loss": 0.1876,
+      "learning_rate": 0.00025051311953352767,
+      "loss": 0.074,
       "step": 6160
     },
     {
       "epoch": 8.86,
-      "learning_rate": 0.0002165799803729146,
-      "loss": 0.2077,
+      "learning_rate": 0.00025042565597667633,
+      "loss": 0.0745,
       "step": 6170
     },
     {
       "epoch": 8.88,
-      "learning_rate": 0.00021643277723258094,
-      "loss": 0.2064,
+      "learning_rate": 0.00025033819241982504,
+      "loss": 0.0895,
       "step": 6180
     },
     {
       "epoch": 8.89,
-      "learning_rate": 0.00021628557409224727,
-      "loss": 0.1796,
+      "learning_rate": 0.00025025072886297376,
+      "loss": 0.0674,
       "step": 6190
     },
     {
       "epoch": 8.91,
-      "learning_rate": 0.00021613837095191363,
-      "loss": 0.1948,
+      "learning_rate": 0.0002501632653061224,
+      "loss": 0.0773,
       "step": 6200
     },
     {
       "epoch": 8.91,
-      "eval_loss": 0.207631453871727,
-      "eval_runtime": 39.8141,
-      "eval_samples_per_second": 11.478,
-      "eval_steps_per_second": 1.457,
-      "eval_wer": 0.2347538021382322,
+      "eval_loss": 0.187347874045372,
+      "eval_runtime": 40.007,
+      "eval_samples_per_second": 11.423,
+      "eval_steps_per_second": 1.45,
+      "eval_wer": 0.1886764041560006,
       "step": 6200
     },
     {
       "epoch": 8.92,
-      "learning_rate": 0.00021599116781157994,
-      "loss": 0.1759,
+      "learning_rate": 0.00025007580174927113,
+      "loss": 0.0648,
       "step": 6210
     },
     {
       "epoch": 8.94,
-      "learning_rate": 0.0002158439646712463,
-      "loss": 0.2199,
+      "learning_rate": 0.0002499883381924198,
+      "loss": 0.0906,
       "step": 6220
     },
     {
       "epoch": 8.95,
-      "learning_rate": 0.00021569676153091263,
-      "loss": 0.1782,
+      "learning_rate": 0.0002499008746355685,
+      "loss": 0.0673,
       "step": 6230
     },
     {
       "epoch": 8.97,
-      "learning_rate": 0.000215549558390579,
-      "loss": 0.1646,
+      "learning_rate": 0.00024981341107871717,
+      "loss": 0.0498,
       "step": 6240
     },
     {
       "epoch": 8.98,
-      "learning_rate": 0.0002154023552502453,
-      "loss": 0.203,
+      "learning_rate": 0.00024972594752186583,
+      "loss": 0.0808,
       "step": 6250
     },
     {
       "epoch": 8.99,
-      "learning_rate": 0.00021525515210991166,
-      "loss": 0.185,
+      "learning_rate": 0.00024963848396501455,
+      "loss": 0.0661,
       "step": 6260
     },
     {
       "epoch": 9.01,
-      "learning_rate": 0.000215107948969578,
-      "loss": 0.1783,
+      "learning_rate": 0.00024955102040816326,
+      "loss": 0.0707,
       "step": 6270
     },
     {
       "epoch": 9.02,
-      "learning_rate": 0.00021496074582924435,
-      "loss": 0.163,
+      "learning_rate": 0.0002494635568513119,
+      "loss": 0.0672,
       "step": 6280
     },
     {
       "epoch": 9.04,
-      "learning_rate": 0.00021481354268891066,
-      "loss": 0.1734,
+      "learning_rate": 0.00024937609329446064,
+      "loss": 0.0678,
       "step": 6290
     },
     {
       "epoch": 9.05,
-      "learning_rate": 0.00021466633954857702,
-      "loss": 0.176,
+      "learning_rate": 0.0002492886297376093,
+      "loss": 0.0735,
       "step": 6300
     },
     {
       "epoch": 9.05,
-      "eval_loss": 0.20396728813648224,
-      "eval_runtime": 39.8795,
-      "eval_samples_per_second": 11.46,
-      "eval_steps_per_second": 1.454,
-      "eval_wer": 0.22933293178738143,
+      "eval_loss": 0.18213719129562378,
+      "eval_runtime": 40.0397,
+      "eval_samples_per_second": 11.414,
+      "eval_steps_per_second": 1.449,
+      "eval_wer": 0.16789640114440596,
       "step": 6300
     },
     {
       "epoch": 9.07,
-      "learning_rate": 0.00021451913640824335,
-      "loss": 0.1499,
+      "learning_rate": 0.000249201166180758,
+      "loss": 0.0521,
       "step": 6310
     },
     {
       "epoch": 9.08,
-      "learning_rate": 0.00021437193326790971,
-      "loss": 0.1693,
+      "learning_rate": 0.0002491137026239067,
+      "loss": 0.0594,
       "step": 6320
     },
     {
       "epoch": 9.09,
-      "learning_rate": 0.00021422473012757602,
-      "loss": 0.1554,
+      "learning_rate": 0.00024902623906705533,
+      "loss": 0.068,
       "step": 6330
     },
     {
       "epoch": 9.11,
-      "learning_rate": 0.00021407752698724238,
-      "loss": 0.158,
+      "learning_rate": 0.00024893877551020405,
+      "loss": 0.0625,
       "step": 6340
     },
     {
       "epoch": 9.12,
-      "learning_rate": 0.00021393032384690871,
-      "loss": 0.1957,
+      "learning_rate": 0.00024885131195335276,
+      "loss": 0.088,
       "step": 6350
     },
     {
       "epoch": 9.14,
-      "learning_rate": 0.00021378312070657505,
-      "loss": 0.1862,
+      "learning_rate": 0.0002487638483965014,
+      "loss": 0.0838,
       "step": 6360
     },
     {
       "epoch": 9.15,
-      "learning_rate": 0.00021363591756624138,
-      "loss": 0.1581,
+      "learning_rate": 0.00024867638483965014,
+      "loss": 0.062,
       "step": 6370
     },
     {
       "epoch": 9.17,
-      "learning_rate": 0.00021348871442590774,
-      "loss": 0.133,
+      "learning_rate": 0.0002485889212827988,
+      "loss": 0.0583,
       "step": 6380
     },
     {
       "epoch": 9.18,
-      "learning_rate": 0.00021334151128557408,
-      "loss": 0.1534,
+      "learning_rate": 0.0002485014577259475,
+      "loss": 0.057,
       "step": 6390
     },
     {
       "epoch": 9.2,
-      "learning_rate": 0.0002131943081452404,
-      "loss": 0.1892,
+      "learning_rate": 0.0002484139941690962,
+      "loss": 0.0779,
       "step": 6400
     },
     {
       "epoch": 9.2,
-      "eval_loss": 0.1975768804550171,
-      "eval_runtime": 39.8911,
-      "eval_samples_per_second": 11.456,
-      "eval_steps_per_second": 1.454,
-      "eval_wer": 0.22466496009637102,
+      "eval_loss": 0.165808767080307,
+      "eval_runtime": 39.9557,
+      "eval_samples_per_second": 11.438,
+      "eval_steps_per_second": 1.452,
+      "eval_wer": 0.17060683631983134,
       "step": 6400
     },
     {
       "epoch": 9.21,
-      "learning_rate": 0.00021304710500490674,
-      "loss": 0.146,
+      "learning_rate": 0.0002483265306122449,
+      "loss": 0.0522,
       "step": 6410
     },
     {
       "epoch": 9.22,
-      "learning_rate": 0.0002128999018645731,
-      "loss": 0.1676,
+      "learning_rate": 0.00024823906705539355,
+      "loss": 0.0686,
       "step": 6420
     },
     {
       "epoch": 9.24,
-      "learning_rate": 0.00021275269872423944,
-      "loss": 0.135,
+      "learning_rate": 0.0002481516034985422,
+      "loss": 0.0545,
       "step": 6430
     },
     {
       "epoch": 9.25,
-      "learning_rate": 0.00021260549558390577,
-      "loss": 0.1776,
+      "learning_rate": 0.00024806413994169093,
+      "loss": 0.0709,
       "step": 6440
     },
     {
       "epoch": 9.27,
-      "learning_rate": 0.0002124582924435721,
-      "loss": 0.155,
+      "learning_rate": 0.00024797667638483964,
+      "loss": 0.0623,
       "step": 6450
     },
     {
       "epoch": 9.28,
-      "learning_rate": 0.00021231108930323846,
-      "loss": 0.1667,
+      "learning_rate": 0.0002478892128279883,
+      "loss": 0.0681,
       "step": 6460
     },
     {
       "epoch": 9.3,
-      "learning_rate": 0.0002121638861629048,
-      "loss": 0.1514,
+      "learning_rate": 0.000247801749271137,
+      "loss": 0.0617,
       "step": 6470
     },
     {
       "epoch": 9.31,
-      "learning_rate": 0.00021201668302257113,
-      "loss": 0.1718,
+      "learning_rate": 0.0002477142857142857,
+      "loss": 0.0659,
       "step": 6480
     },
     {
       "epoch": 9.32,
-      "learning_rate": 0.00021186947988223746,
-      "loss": 0.1912,
+      "learning_rate": 0.0002476268221574344,
+      "loss": 0.0674,
       "step": 6490
     },
     {
       "epoch": 9.34,
-      "learning_rate": 0.00021172227674190382,
-      "loss": 0.1645,
+      "learning_rate": 0.00024753935860058306,
+      "loss": 0.0655,
       "step": 6500
     },
     {
       "epoch": 9.34,
-      "eval_loss": 0.19538117945194244,
-      "eval_runtime": 40.169,
-      "eval_samples_per_second": 11.377,
-      "eval_steps_per_second": 1.444,
-      "eval_wer": 0.21547959644631834,
+      "eval_loss": 0.16366757452487946,
+      "eval_runtime": 39.9847,
+      "eval_samples_per_second": 11.429,
+      "eval_steps_per_second": 1.451,
+      "eval_wer": 0.1683481403403102,
       "step": 6500
     },
     {
       "epoch": 9.35,
-      "learning_rate": 0.00021157507360157013,
-      "loss": 0.1581,
+      "learning_rate": 0.0002474518950437317,
+      "loss": 0.0694,
       "step": 6510
     },
     {
       "epoch": 9.37,
-      "learning_rate": 0.0002114278704612365,
-      "loss": 0.1781,
+      "learning_rate": 0.00024736443148688043,
+      "loss": 0.073,
       "step": 6520
     },
     {
       "epoch": 9.38,
-      "learning_rate": 0.00021128066732090282,
-      "loss": 0.1437,
+      "learning_rate": 0.00024727696793002915,
+      "loss": 0.0614,
       "step": 6530
     },
     {
       "epoch": 9.4,
-      "learning_rate": 0.00021113346418056918,
-      "loss": 0.1538,
+      "learning_rate": 0.0002471895043731778,
+      "loss": 0.0538,
       "step": 6540
     },
     {
       "epoch": 9.41,
-      "learning_rate": 0.0002109862610402355,
-      "loss": 0.1945,
+      "learning_rate": 0.0002471020408163265,
+      "loss": 0.0779,
       "step": 6550
     },
     {
       "epoch": 9.43,
-      "learning_rate": 0.00021083905789990185,
-      "loss": 0.1629,
+      "learning_rate": 0.0002470145772594752,
+      "loss": 0.0556,
       "step": 6560
     },
     {
       "epoch": 9.44,
-      "learning_rate": 0.00021069185475956818,
-      "loss": 0.1905,
+      "learning_rate": 0.0002469271137026239,
+      "loss": 0.0848,
       "step": 6570
     },
     {
       "epoch": 9.45,
-      "learning_rate": 0.00021054465161923455,
-      "loss": 0.1688,
+      "learning_rate": 0.00024683965014577256,
+      "loss": 0.0705,
       "step": 6580
     },
     {
       "epoch": 9.47,
-      "learning_rate": 0.00021039744847890085,
-      "loss": 0.1678,
+      "learning_rate": 0.0002467521865889213,
+      "loss": 0.0638,
       "step": 6590
     },
     {
       "epoch": 9.48,
-      "learning_rate": 0.0002102502453385672,
-      "loss": 0.1723,
+      "learning_rate": 0.00024666472303206993,
+      "loss": 0.0651,
       "step": 6600
     },
     {
       "epoch": 9.48,
-      "eval_loss": 0.19737230241298676,
-      "eval_runtime": 40.207,
-      "eval_samples_per_second": 11.366,
-      "eval_steps_per_second": 1.443,
-      "eval_wer": 0.2215027857250414,
+      "eval_loss": 0.16613377630710602,
+      "eval_runtime": 40.0138,
+      "eval_samples_per_second": 11.421,
+      "eval_steps_per_second": 1.449,
+      "eval_wer": 0.16940219846408675,
       "step": 6600
     },
     {
       "epoch": 9.5,
-      "learning_rate": 0.00021010304219823355,
-      "loss": 0.1664,
+      "learning_rate": 0.00024657725947521865,
+      "loss": 0.055,
       "step": 6610
     },
     {
       "epoch": 9.51,
-      "learning_rate": 0.0002099558390578999,
-      "loss": 0.1783,
+      "learning_rate": 0.0002464897959183673,
+      "loss": 0.0659,
       "step": 6620
     },
     {
       "epoch": 9.53,
-      "learning_rate": 0.0002098086359175662,
-      "loss": 0.1776,
+      "learning_rate": 0.000246402332361516,
+      "loss": 0.0704,
       "step": 6630
     },
     {
       "epoch": 9.54,
-      "learning_rate": 0.00020966143277723257,
-      "loss": 0.1511,
+      "learning_rate": 0.0002463148688046647,
+      "loss": 0.0488,
       "step": 6640
     },
     {
       "epoch": 9.55,
-      "learning_rate": 0.0002095142296368989,
-      "loss": 0.1905,
+      "learning_rate": 0.0002462274052478134,
+      "loss": 0.0831,
       "step": 6650
     },
     {
       "epoch": 9.57,
-      "learning_rate": 0.00020936702649656524,
-      "loss": 0.1478,
+      "learning_rate": 0.00024613994169096206,
+      "loss": 0.0603,
       "step": 6660
     },
     {
       "epoch": 9.58,
-      "learning_rate": 0.00020921982335623157,
-      "loss": 0.1802,
+      "learning_rate": 0.0002460524781341108,
+      "loss": 0.0739,
       "step": 6670
     },
     {
       "epoch": 9.6,
-      "learning_rate": 0.00020907262021589793,
-      "loss": 0.1668,
+      "learning_rate": 0.0002459650145772595,
+      "loss": 0.0708,
       "step": 6680
     },
     {
       "epoch": 9.61,
-      "learning_rate": 0.00020892541707556427,
-      "loss": 0.1529,
+      "learning_rate": 0.00024587755102040815,
+      "loss": 0.0579,
       "step": 6690
     },
     {
       "epoch": 9.63,
-      "learning_rate": 0.0002087782139352306,
-      "loss": 0.1685,
+      "learning_rate": 0.0002457900874635568,
+      "loss": 0.0806,
       "step": 6700
     },
     {
       "epoch": 9.63,
-      "eval_loss": 0.20146189630031586,
-      "eval_runtime": 39.9776,
-      "eval_samples_per_second": 11.431,
-      "eval_steps_per_second": 1.451,
-      "eval_wer": 0.21773829242583947,
+      "eval_loss": 0.16738936305046082,
+      "eval_runtime": 39.7838,
+      "eval_samples_per_second": 11.487,
+      "eval_steps_per_second": 1.458,
+      "eval_wer": 0.17467248908296942,
       "step": 6700
     },
     {
       "epoch": 9.64,
-      "learning_rate": 0.00020863101079489693,
-      "loss": 0.1339,
+      "learning_rate": 0.00024570262390670553,
+      "loss": 0.0572,
       "step": 6710
     },
     {
       "epoch": 9.66,
-      "learning_rate": 0.0002084838076545633,
-      "loss": 0.175,
+      "learning_rate": 0.0002456151603498542,
+      "loss": 0.0713,
       "step": 6720
     },
     {
       "epoch": 9.67,
-      "learning_rate": 0.00020833660451422963,
-      "loss": 0.1607,
+      "learning_rate": 0.0002455276967930029,
+      "loss": 0.0787,
       "step": 6730
     },
     {
       "epoch": 9.68,
-      "learning_rate": 0.00020818940137389596,
-      "loss": 0.1571,
+      "learning_rate": 0.00024544023323615156,
+      "loss": 0.0651,
       "step": 6740
     },
     {
       "epoch": 9.7,
-      "learning_rate": 0.0002080421982335623,
-      "loss": 0.1678,
+      "learning_rate": 0.0002453527696793003,
+      "loss": 0.0677,
       "step": 6750
     },
     {
       "epoch": 9.71,
-      "learning_rate": 0.00020789499509322865,
-      "loss": 0.1585,
+      "learning_rate": 0.000245265306122449,
+      "loss": 0.0641,
       "step": 6760
     },
     {
       "epoch": 9.73,
-      "learning_rate": 0.000207747791952895,
-      "loss": 0.2009,
+      "learning_rate": 0.00024517784256559766,
+      "loss": 0.0768,
       "step": 6770
     },
     {
       "epoch": 9.74,
-      "learning_rate": 0.00020760058881256132,
-      "loss": 0.1597,
+      "learning_rate": 0.0002450903790087463,
+      "loss": 0.0651,
       "step": 6780
     },
     {
       "epoch": 9.76,
-      "learning_rate": 0.00020745338567222765,
-      "loss": 0.1763,
+      "learning_rate": 0.00024500291545189503,
+      "loss": 0.0676,
       "step": 6790
     },
     {
       "epoch": 9.77,
-      "learning_rate": 0.00020730618253189402,
-      "loss": 0.1617,
+      "learning_rate": 0.0002449154518950437,
+      "loss": 0.065,
       "step": 6800
     },
     {
       "epoch": 9.77,
-      "eval_loss": 0.20486493408679962,
-      "eval_runtime": 39.9706,
+      "eval_loss": 0.17342041432857513,
+      "eval_runtime": 39.9722,
       "eval_samples_per_second": 11.433,
       "eval_steps_per_second": 1.451,
-      "eval_wer": 0.22195452492094564,
+      "eval_wer": 0.17271495256738442,
       "step": 6800
     },
     {
       "epoch": 9.78,
-      "learning_rate": 0.00020715897939156032,
-      "loss": 0.1613,
+      "learning_rate": 0.0002448279883381924,
+      "loss": 0.0582,
       "step": 6810
     },
     {
       "epoch": 9.8,
-      "learning_rate": 0.00020701177625122666,
-      "loss": 0.1911,
+      "learning_rate": 0.00024474052478134107,
+      "loss": 0.0798,
       "step": 6820
     },
     {
       "epoch": 9.81,
-      "learning_rate": 0.00020686457311089302,
-      "loss": 0.1849,
+      "learning_rate": 0.0002446530612244898,
+      "loss": 0.0757,
       "step": 6830
     },
     {
       "epoch": 9.83,
-      "learning_rate": 0.00020671736997055938,
-      "loss": 0.1719,
+      "learning_rate": 0.00024456559766763844,
+      "loss": 0.0672,
       "step": 6840
     },
     {
       "epoch": 9.84,
-      "learning_rate": 0.00020657016683022568,
-      "loss": 0.1817,
+      "learning_rate": 0.00024447813411078716,
+      "loss": 0.0694,
       "step": 6850
     },
     {
       "epoch": 9.86,
-      "learning_rate": 0.00020642296368989202,
-      "loss": 0.1575,
+      "learning_rate": 0.0002443906705539359,
+      "loss": 0.0722,
       "step": 6860
     },
     {
       "epoch": 9.87,
-      "learning_rate": 0.00020627576054955838,
-      "loss": 0.1655,
+      "learning_rate": 0.00024430320699708453,
+      "loss": 0.0676,
       "step": 6870
     },
     {
       "epoch": 9.89,
-      "learning_rate": 0.00020612855740922474,
-      "loss": 0.1773,
+      "learning_rate": 0.0002442157434402332,
+      "loss": 0.071,
       "step": 6880
     },
     {
       "epoch": 9.9,
-      "learning_rate": 0.00020598135426889104,
-      "loss": 0.1749,
+      "learning_rate": 0.00024412827988338188,
+      "loss": 0.0635,
       "step": 6890
     },
     {
       "epoch": 9.91,
-      "learning_rate": 0.00020583415112855738,
-      "loss": 0.1669,
+      "learning_rate": 0.0002440408163265306,
+      "loss": 0.0719,
       "step": 6900
     },
     {
       "epoch": 9.91,
-      "eval_loss": 0.19124703109264374,
-      "eval_runtime": 39.8375,
-      "eval_samples_per_second": 11.472,
-      "eval_steps_per_second": 1.456,
-      "eval_wer": 0.21547959644631834,
+      "eval_loss": 0.1549205482006073,
+      "eval_runtime": 39.9005,
+      "eval_samples_per_second": 11.453,
+      "eval_steps_per_second": 1.454,
+      "eval_wer": 0.1615720524017467,
       "step": 6900
     },
     {
       "epoch": 9.93,
-      "learning_rate": 0.00020568694798822374,
-      "loss": 0.1504,
+      "learning_rate": 0.00024395335276967929,
+      "loss": 0.0605,
       "step": 6910
     },
     {
       "epoch": 9.94,
-      "learning_rate": 0.0002055397448478901,
-      "loss": 0.1713,
+      "learning_rate": 0.00024386588921282797,
+      "loss": 0.0661,
       "step": 6920
     },
     {
       "epoch": 9.96,
-      "learning_rate": 0.0002053925417075564,
-      "loss": 0.1773,
+      "learning_rate": 0.00024377842565597666,
+      "loss": 0.0729,
       "step": 6930
     },
     {
       "epoch": 9.97,
-      "learning_rate": 0.00020524533856722274,
-      "loss": 0.158,
+      "learning_rate": 0.00024369096209912535,
+      "loss": 0.0594,
       "step": 6940
     },
     {
       "epoch": 9.99,
-      "learning_rate": 0.0002050981354268891,
-      "loss": 0.1632,
+      "learning_rate": 0.00024360349854227404,
+      "loss": 0.0677,
       "step": 6950
     },
     {
       "epoch": 10.0,
-      "learning_rate": 0.0002049509322865554,
-      "loss": 0.1622,
+      "learning_rate": 0.0002435160349854227,
+      "loss": 0.0626,
       "step": 6960
     },
     {
       "epoch": 10.01,
-      "learning_rate": 0.00020480372914622176,
-      "loss": 0.1693,
+      "learning_rate": 0.00024342857142857139,
+      "loss": 0.0754,
       "step": 6970
     },
     {
       "epoch": 10.03,
-      "learning_rate": 0.0002046565260058881,
-      "loss": 0.1505,
+      "learning_rate": 0.00024334110787172007,
+      "loss": 0.0585,
       "step": 6980
     },
     {
       "epoch": 10.04,
-      "learning_rate": 0.00020450932286555446,
-      "loss": 0.1477,
+      "learning_rate": 0.0002432536443148688,
+      "loss": 0.0591,
       "step": 6990
     },
     {
       "epoch": 10.06,
-      "learning_rate": 0.00020436211972522076,
-      "loss": 0.1504,
+      "learning_rate": 0.00024316618075801748,
+      "loss": 0.0672,
       "step": 7000
     },
     {
       "epoch": 10.06,
-      "eval_loss": 0.1917961686849594,
-      "eval_runtime": 40.0723,
-      "eval_samples_per_second": 11.404,
-      "eval_steps_per_second": 1.447,
-      "eval_wer": 0.21547959644631834,
+      "eval_loss": 0.1598873883485794,
+      "eval_runtime": 39.9955,
+      "eval_samples_per_second": 11.426,
+      "eval_steps_per_second": 1.45,
+      "eval_wer": 0.16292726998945942,
       "step": 7000
     },
     {
       "epoch": 10.07,
-      "learning_rate": 0.00020421491658488713,
-      "loss": 0.121,
+      "learning_rate": 0.00024307871720116617,
+      "loss": 0.0445,
       "step": 7010
     },
     {
       "epoch": 10.09,
-      "learning_rate": 0.00020406771344455346,
-      "loss": 0.1654,
+      "learning_rate": 0.00024299125364431485,
+      "loss": 0.075,
       "step": 7020
     },
     {
       "epoch": 10.1,
-      "learning_rate": 0.00020392051030421982,
-      "loss": 0.1567,
+      "learning_rate": 0.00024290379008746354,
+      "loss": 0.067,
       "step": 7030
     },
     {
       "epoch": 10.11,
-      "learning_rate": 0.00020377330716388613,
-      "loss": 0.1748,
+      "learning_rate": 0.00024281632653061223,
+      "loss": 0.0765,
       "step": 7040
     },
     {
       "epoch": 10.13,
-      "learning_rate": 0.00020362610402355249,
-      "loss": 0.1355,
+      "learning_rate": 0.0002427288629737609,
+      "loss": 0.0614,
       "step": 7050
     },
     {
       "epoch": 10.14,
-      "learning_rate": 0.00020347890088321882,
-      "loss": 0.1484,
+      "learning_rate": 0.00024264139941690958,
+      "loss": 0.0541,
       "step": 7060
     },
     {
       "epoch": 10.16,
-      "learning_rate": 0.00020333169774288518,
-      "loss": 0.1646,
+      "learning_rate": 0.0002425539358600583,
+      "loss": 0.076,
       "step": 7070
     },
     {
       "epoch": 10.17,
-      "learning_rate": 0.00020318449460255149,
-      "loss": 0.1906,
+      "learning_rate": 0.00024246647230320698,
+      "loss": 0.0833,
       "step": 7080
     },
     {
       "epoch": 10.19,
-      "learning_rate": 0.00020303729146221785,
-      "loss": 0.1629,
+      "learning_rate": 0.00024237900874635567,
+      "loss": 0.069,
       "step": 7090
     },
     {
       "epoch": 10.2,
-      "learning_rate": 0.00020289008832188418,
-      "loss": 0.1715,
+      "learning_rate": 0.00024229154518950436,
+      "loss": 0.0712,
       "step": 7100
     },
     {
       "epoch": 10.2,
-      "eval_loss": 0.20631931722164154,
-      "eval_runtime": 40.224,
-      "eval_samples_per_second": 11.361,
-      "eval_steps_per_second": 1.442,
-      "eval_wer": 0.2124680018069568,
+      "eval_loss": 0.1684901863336563,
+      "eval_runtime": 40.0595,
+      "eval_samples_per_second": 11.408,
+      "eval_steps_per_second": 1.448,
+      "eval_wer": 0.16654118355669326,
       "step": 7100
     },
     {
       "epoch": 10.22,
-      "learning_rate": 0.0002027428851815505,
-      "loss": 0.1555,
+      "learning_rate": 0.00024220408163265304,
+      "loss": 0.0585,
       "step": 7110
     },
     {
       "epoch": 10.23,
-      "learning_rate": 0.00020259568204121685,
-      "loss": 0.1834,
+      "learning_rate": 0.00024211661807580173,
+      "loss": 0.0802,
       "step": 7120
     },
     {
       "epoch": 10.24,
-      "learning_rate": 0.0002024484789008832,
-      "loss": 0.1437,
+      "learning_rate": 0.00024202915451895042,
+      "loss": 0.0573,
       "step": 7130
     },
     {
       "epoch": 10.26,
-      "learning_rate": 0.00020230127576054954,
-      "loss": 0.1608,
+      "learning_rate": 0.00024194169096209908,
+      "loss": 0.0692,
       "step": 7140
     },
     {
       "epoch": 10.27,
-      "learning_rate": 0.00020215407262021587,
-      "loss": 0.159,
+      "learning_rate": 0.00024185422740524777,
+      "loss": 0.0664,
       "step": 7150
     },
     {
       "epoch": 10.29,
-      "learning_rate": 0.0002020068694798822,
-      "loss": 0.1512,
+      "learning_rate": 0.00024176676384839648,
+      "loss": 0.0585,
       "step": 7160
     },
     {
       "epoch": 10.3,
-      "learning_rate": 0.00020185966633954857,
-      "loss": 0.1701,
+      "learning_rate": 0.00024167930029154517,
+      "loss": 0.0776,
       "step": 7170
     },
     {
       "epoch": 10.32,
-      "learning_rate": 0.0002017124631992149,
-      "loss": 0.1324,
+      "learning_rate": 0.00024159183673469386,
+      "loss": 0.0606,
       "step": 7180
     },
     {
       "epoch": 10.33,
-      "learning_rate": 0.00020156526005888123,
-      "loss": 0.1629,
+      "learning_rate": 0.00024150437317784255,
+      "loss": 0.073,
       "step": 7190
     },
     {
       "epoch": 10.34,
-      "learning_rate": 0.00020141805691854757,
-      "loss": 0.1604,
+      "learning_rate": 0.00024141690962099124,
+      "loss": 0.0687,
       "step": 7200
     },
     {
       "epoch": 10.34,
-      "eval_loss": 0.1930931955575943,
-      "eval_runtime": 40.1705,
-      "eval_samples_per_second": 11.377,
-      "eval_steps_per_second": 1.444,
-      "eval_wer": 0.21261858153892485,
+      "eval_loss": 0.16169100999832153,
+      "eval_runtime": 39.9721,
+      "eval_samples_per_second": 11.433,
+      "eval_steps_per_second": 1.451,
+      "eval_wer": 0.1644330673091402,
       "step": 7200
     },
     {
       "epoch": 10.36,
-      "learning_rate": 0.00020127085377821393,
-      "loss": 0.1234,
+      "learning_rate": 0.00024132944606413992,
+      "loss": 0.0499,
       "step": 7210
     },
     {
       "epoch": 10.37,
-      "learning_rate": 0.00020112365063788026,
-      "loss": 0.1739,
+      "learning_rate": 0.0002412419825072886,
+      "loss": 0.0765,
       "step": 7220
     },
     {
       "epoch": 10.39,
-      "learning_rate": 0.0002009764474975466,
-      "loss": 0.1482,
+      "learning_rate": 0.00024115451895043727,
+      "loss": 0.0591,
       "step": 7230
     },
     {
       "epoch": 10.4,
-      "learning_rate": 0.00020082924435721293,
-      "loss": 0.1717,
+      "learning_rate": 0.00024106705539358596,
+      "loss": 0.074,
       "step": 7240
     },
     {
       "epoch": 10.42,
-      "learning_rate": 0.0002006820412168793,
-      "loss": 0.1697,
+      "learning_rate": 0.00024097959183673467,
+      "loss": 0.0747,
       "step": 7250
     },
     {
       "epoch": 10.43,
-      "learning_rate": 0.0002005348380765456,
-      "loss": 0.1609,
+      "learning_rate": 0.00024089212827988336,
+      "loss": 0.0677,
       "step": 7260
     },
     {
       "epoch": 10.45,
-      "learning_rate": 0.00020038763493621196,
-      "loss": 0.1583,
+      "learning_rate": 0.00024080466472303205,
+      "loss": 0.075,
       "step": 7270
     },
     {
       "epoch": 10.46,
-      "learning_rate": 0.0002002404317958783,
-      "loss": 0.1319,
+      "learning_rate": 0.00024071720116618074,
+      "loss": 0.0523,
       "step": 7280
     },
     {
       "epoch": 10.47,
-      "learning_rate": 0.00020009322865554465,
-      "loss": 0.1548,
+      "learning_rate": 0.00024062973760932943,
+      "loss": 0.0693,
       "step": 7290
     },
     {
       "epoch": 10.49,
-      "learning_rate": 0.00019994602551521096,
-      "loss": 0.1296,
+      "learning_rate": 0.00024054227405247811,
+      "loss": 0.0586,
       "step": 7300
     },
     {
       "epoch": 10.49,
-      "eval_loss": 0.1908556967973709,
-      "eval_runtime": 39.9017,
-      "eval_samples_per_second": 11.453,
+      "eval_loss": 0.1741245537996292,
+      "eval_runtime": 39.8869,
+      "eval_samples_per_second": 11.457,
       "eval_steps_per_second": 1.454,
-      "eval_wer": 0.21261858153892485,
+      "eval_wer": 0.17000451739195904,
       "step": 7300
     },
     {
       "epoch": 10.5,
-      "learning_rate": 0.00019979882237487732,
-      "loss": 0.1352,
+      "learning_rate": 0.0002404548104956268,
+      "loss": 0.059,
       "step": 7310
     },
     {
       "epoch": 10.52,
-      "learning_rate": 0.00019965161923454365,
-      "loss": 0.1664,
+      "learning_rate": 0.00024036734693877546,
+      "loss": 0.0691,
       "step": 7320
     },
     {
       "epoch": 10.53,
-      "learning_rate": 0.00019950441609421,
-      "loss": 0.1202,
+      "learning_rate": 0.00024027988338192418,
+      "loss": 0.0508,
       "step": 7330
     },
     {
       "epoch": 10.55,
-      "learning_rate": 0.00019935721295387632,
-      "loss": 0.1642,
+      "learning_rate": 0.00024019241982507287,
+      "loss": 0.0682,
       "step": 7340
     },
     {
       "epoch": 10.56,
-      "learning_rate": 0.00019921000981354268,
-      "loss": 0.1413,
+      "learning_rate": 0.00024010495626822155,
+      "loss": 0.0682,
       "step": 7350
     },
     {
       "epoch": 10.57,
-      "learning_rate": 0.000199062806673209,
-      "loss": 0.1233,
+      "learning_rate": 0.00024001749271137024,
+      "loss": 0.0532,
       "step": 7360
     },
     {
       "epoch": 10.59,
-      "learning_rate": 0.00019891560353287537,
-      "loss": 0.1356,
+      "learning_rate": 0.00023993002915451893,
+      "loss": 0.0629,
       "step": 7370
     },
     {
       "epoch": 10.6,
-      "learning_rate": 0.00019876840039254168,
-      "loss": 0.1423,
+      "learning_rate": 0.00023984256559766762,
+      "loss": 0.064,
       "step": 7380
     },
     {
       "epoch": 10.62,
-      "learning_rate": 0.00019862119725220804,
-      "loss": 0.1568,
+      "learning_rate": 0.0002397551020408163,
+      "loss": 0.0692,
       "step": 7390
     },
     {
       "epoch": 10.63,
-      "learning_rate": 0.00019847399411187437,
-      "loss": 0.1405,
+      "learning_rate": 0.00023966763848396502,
+      "loss": 0.0628,
       "step": 7400
     },
     {
       "epoch": 10.63,
-      "eval_loss": 0.19520747661590576,
-      "eval_runtime": 39.9029,
-      "eval_samples_per_second": 11.453,
-      "eval_steps_per_second": 1.454,
-      "eval_wer": 0.21066104502333985,
+      "eval_loss": 0.175329327583313,
+      "eval_runtime": 40.1747,
+      "eval_samples_per_second": 11.375,
+      "eval_steps_per_second": 1.444,
+      "eval_wer": 0.1675952416804698,
       "step": 7400
     },
     {
       "epoch": 10.65,
-      "learning_rate": 0.0001983267909715407,
-      "loss": 0.1478,
+      "learning_rate": 0.00023958017492711365,
+      "loss": 0.0557,
       "step": 7410
     },
     {
       "epoch": 10.66,
-      "learning_rate": 0.00019817958783120704,
-      "loss": 0.1633,
+      "learning_rate": 0.00023949271137026237,
+      "loss": 0.0771,
       "step": 7420
     },
     {
       "epoch": 10.68,
-      "learning_rate": 0.0001980323846908734,
-      "loss": 0.1438,
+      "learning_rate": 0.00023940524781341106,
+      "loss": 0.0599,
       "step": 7430
     },
     {
       "epoch": 10.69,
-      "learning_rate": 0.00019788518155053973,
-      "loss": 0.1668,
+      "learning_rate": 0.00023931778425655974,
+      "loss": 0.0782,
       "step": 7440
     },
     {
       "epoch": 10.7,
-      "learning_rate": 0.00019773797841020607,
-      "loss": 0.1472,
+      "learning_rate": 0.00023923032069970843,
+      "loss": 0.0611,
       "step": 7450
     },
     {
       "epoch": 10.72,
-      "learning_rate": 0.0001975907752698724,
-      "loss": 0.1323,
+      "learning_rate": 0.00023914285714285712,
+      "loss": 0.0488,
       "step": 7460
     },
     {
       "epoch": 10.73,
-      "learning_rate": 0.00019744357212953876,
-      "loss": 0.1607,
+      "learning_rate": 0.0002390553935860058,
+      "loss": 0.072,
       "step": 7470
     },
     {
       "epoch": 10.75,
-      "learning_rate": 0.0001972963689892051,
-      "loss": 0.1649,
+      "learning_rate": 0.0002389679300291545,
+      "loss": 0.0644,
       "step": 7480
     },
     {
       "epoch": 10.76,
-      "learning_rate": 0.00019714916584887143,
-      "loss": 0.1469,
+      "learning_rate": 0.0002388804664723032,
+      "loss": 0.0577,
       "step": 7490
     },
     {
       "epoch": 10.78,
-      "learning_rate": 0.00019700196270853776,
-      "loss": 0.1423,
+      "learning_rate": 0.00023879300291545187,
+      "loss": 0.0644,
       "step": 7500
     },
     {
       "epoch": 10.78,
-      "eval_loss": 0.20175614953041077,
-      "eval_runtime": 39.9105,
-      "eval_samples_per_second": 11.451,
-      "eval_steps_per_second": 1.453,
-      "eval_wer": 0.2096069868995633,
+      "eval_loss": 0.18111708760261536,
+      "eval_runtime": 40.0144,
+      "eval_samples_per_second": 11.421,
+      "eval_steps_per_second": 1.449,
+      "eval_wer": 0.17542538774280983,
       "step": 7500
     },
     {
       "epoch": 10.79,
-      "learning_rate": 0.00019685475956820412,
-      "loss": 0.1369,
+      "learning_rate": 0.00023870553935860056,
+      "loss": 0.0564,
       "step": 7510
     },
     {
       "epoch": 10.8,
-      "learning_rate": 0.00019670755642787043,
-      "loss": 0.1513,
+      "learning_rate": 0.00023861807580174925,
+      "loss": 0.0675,
       "step": 7520
     },
     {
       "epoch": 10.82,
-      "learning_rate": 0.0001965603532875368,
-      "loss": 0.1426,
+      "learning_rate": 0.00023853061224489794,
+      "loss": 0.0585,
       "step": 7530
     },
     {
       "epoch": 10.83,
-      "learning_rate": 0.00019641315014720312,
-      "loss": 0.182,
+      "learning_rate": 0.00023844314868804662,
+      "loss": 0.0811,
       "step": 7540
     },
     {
       "epoch": 10.85,
-      "learning_rate": 0.00019626594700686948,
-      "loss": 0.1548,
+      "learning_rate": 0.0002383556851311953,
+      "loss": 0.0647,
       "step": 7550
     },
     {
       "epoch": 10.86,
-      "learning_rate": 0.0001961187438665358,
-      "loss": 0.1208,
+      "learning_rate": 0.000238268221574344,
+      "loss": 0.0447,
       "step": 7560
     },
     {
       "epoch": 10.88,
-      "learning_rate": 0.00019597154072620215,
-      "loss": 0.1596,
+      "learning_rate": 0.00023818075801749271,
+      "loss": 0.0748,
       "step": 7570
     },
     {
       "epoch": 10.89,
-      "learning_rate": 0.00019582433758586848,
-      "loss": 0.1369,
+      "learning_rate": 0.0002380932944606414,
+      "loss": 0.0535,
       "step": 7580
     },
     {
       "epoch": 10.91,
-      "learning_rate": 0.00019567713444553484,
-      "loss": 0.1583,
+      "learning_rate": 0.00023800583090379006,
+      "loss": 0.0638,
       "step": 7590
     },
     {
       "epoch": 10.92,
-      "learning_rate": 0.00019552993130520115,
-      "loss": 0.1434,
+      "learning_rate": 0.00023791836734693875,
+      "loss": 0.0583,
       "step": 7600
     },
     {
       "epoch": 10.92,
-      "eval_loss": 0.18903155624866486,
-      "eval_runtime": 40.197,
-      "eval_samples_per_second": 11.369,
-      "eval_steps_per_second": 1.443,
-      "eval_wer": 0.2064448125282337,
+      "eval_loss": 0.16907170414924622,
+      "eval_runtime": 39.9078,
+      "eval_samples_per_second": 11.451,
+      "eval_steps_per_second": 1.453,
+      "eval_wer": 0.17647944586658637,
       "step": 7600
     },
     {
       "epoch": 10.93,
-      "learning_rate": 0.0001953827281648675,
-      "loss": 0.1513,
+      "learning_rate": 0.00023783090379008744,
+      "loss": 0.053,
       "step": 7610
     },
     {
       "epoch": 10.95,
-      "learning_rate": 0.00019523552502453384,
-      "loss": 0.1791,
+      "learning_rate": 0.00023774344023323613,
+      "loss": 0.0822,
       "step": 7620
     },
     {
       "epoch": 10.96,
-      "learning_rate": 0.0001950883218842002,
-      "loss": 0.1272,
+      "learning_rate": 0.00023765597667638481,
+      "loss": 0.0535,
       "step": 7630
     },
     {
       "epoch": 10.98,
-      "learning_rate": 0.0001949411187438665,
-      "loss": 0.1441,
+      "learning_rate": 0.0002375685131195335,
+      "loss": 0.0611,
       "step": 7640
     },
     {
       "epoch": 10.99,
-      "learning_rate": 0.00019479391560353287,
-      "loss": 0.1205,
+      "learning_rate": 0.0002374810495626822,
+      "loss": 0.0452,
       "step": 7650
     },
     {
       "epoch": 11.01,
-      "learning_rate": 0.0001946467124631992,
-      "loss": 0.151,
+      "learning_rate": 0.0002373935860058309,
+      "loss": 0.0676,
       "step": 7660
     },
     {
       "epoch": 11.02,
-      "learning_rate": 0.0001944995093228655,
-      "loss": 0.136,
+      "learning_rate": 0.0002373061224489796,
+      "loss": 0.0563,
       "step": 7670
     },
     {
       "epoch": 11.03,
-      "learning_rate": 0.00019435230618253187,
-      "loss": 0.1342,
+      "learning_rate": 0.00023721865889212825,
+      "loss": 0.0537,
       "step": 7680
     },
     {
       "epoch": 11.05,
-      "learning_rate": 0.00019420510304219823,
-      "loss": 0.1474,
+      "learning_rate": 0.00023713119533527694,
+      "loss": 0.0702,
       "step": 7690
     },
     {
       "epoch": 11.06,
-      "learning_rate": 0.00019405789990186456,
-      "loss": 0.1439,
+      "learning_rate": 0.00023704373177842563,
+      "loss": 0.0645,
       "step": 7700
     },
     {
       "epoch": 11.06,
-      "eval_loss": 0.1940617859363556,
-      "eval_runtime": 39.8635,
-      "eval_samples_per_second": 11.464,
+      "eval_loss": 0.16939429938793182,
+      "eval_runtime": 39.8761,
+      "eval_samples_per_second": 11.461,
       "eval_steps_per_second": 1.455,
-      "eval_wer": 0.20268031922903176,
+      "eval_wer": 0.16684234302062942,
       "step": 7700
     },
     {
       "epoch": 11.08,
-      "learning_rate": 0.00019391069676153087,
-      "loss": 0.1328,
+      "learning_rate": 0.00023695626822157432,
+      "loss": 0.0638,
       "step": 7710
     },
     {
       "epoch": 11.09,
-      "learning_rate": 0.00019376349362119723,
-      "loss": 0.1136,
+      "learning_rate": 0.000236868804664723,
+      "loss": 0.0513,
       "step": 7720
     },
     {
       "epoch": 11.11,
-      "learning_rate": 0.0001936162904808636,
-      "loss": 0.1119,
+      "learning_rate": 0.0002367813411078717,
+      "loss": 0.0416,
       "step": 7730
     },
     {
       "epoch": 11.12,
-      "learning_rate": 0.00019346908734052992,
-      "loss": 0.1645,
+      "learning_rate": 0.00023669387755102038,
+      "loss": 0.0846,
       "step": 7740
     },
     {
       "epoch": 11.14,
-      "learning_rate": 0.00019332188420019623,
-      "loss": 0.1351,
+      "learning_rate": 0.0002366064139941691,
+      "loss": 0.0552,
       "step": 7750
     },
     {
       "epoch": 11.15,
-      "learning_rate": 0.0001931746810598626,
-      "loss": 0.1427,
+      "learning_rate": 0.00023651895043731778,
+      "loss": 0.061,
       "step": 7760
     },
     {
       "epoch": 11.16,
-      "learning_rate": 0.00019302747791952895,
-      "loss": 0.1949,
+      "learning_rate": 0.00023643148688046645,
+      "loss": 0.0928,
       "step": 7770
     },
     {
       "epoch": 11.18,
-      "learning_rate": 0.00019288027477919528,
-      "loss": 0.1581,
+      "learning_rate": 0.00023634402332361513,
+      "loss": 0.0694,
       "step": 7780
     },
     {
       "epoch": 11.19,
-      "learning_rate": 0.0001927330716388616,
-      "loss": 0.1372,
+      "learning_rate": 0.00023625655976676382,
+      "loss": 0.0618,
       "step": 7790
     },
     {
       "epoch": 11.21,
-      "learning_rate": 0.00019258586849852795,
-      "loss": 0.1415,
+      "learning_rate": 0.0002361690962099125,
+      "loss": 0.0608,
       "step": 7800
     },
     {
       "epoch": 11.21,
-      "eval_loss": 0.18760690093040466,
-      "eval_runtime": 40.0158,
-      "eval_samples_per_second": 11.42,
+      "eval_loss": 0.17372234165668488,
+      "eval_runtime": 40.0285,
+      "eval_samples_per_second": 11.417,
       "eval_steps_per_second": 1.449,
-      "eval_wer": 0.20268031922903176,
+      "eval_wer": 0.16488480650504442,
       "step": 7800
     },
     {
       "epoch": 11.22,
-      "learning_rate": 0.0001924386653581943,
-      "loss": 0.1523,
+      "learning_rate": 0.0002360816326530612,
+      "loss": 0.067,
       "step": 7810
     },
     {
       "epoch": 11.24,
-      "learning_rate": 0.00019229146221786062,
-      "loss": 0.1295,
+      "learning_rate": 0.00023599416909620988,
+      "loss": 0.0566,
       "step": 7820
     },
     {
       "epoch": 11.25,
-      "learning_rate": 0.00019214425907752695,
-      "loss": 0.1288,
+      "learning_rate": 0.0002359067055393586,
+      "loss": 0.0471,
       "step": 7830
     },
     {
       "epoch": 11.26,
-      "learning_rate": 0.0001919970559371933,
-      "loss": 0.1567,
+      "learning_rate": 0.0002358192419825073,
+      "loss": 0.0689,
       "step": 7840
     },
     {
       "epoch": 11.28,
-      "learning_rate": 0.00019184985279685967,
-      "loss": 0.1315,
+      "learning_rate": 0.00023573177842565598,
+      "loss": 0.0483,
       "step": 7850
     },
     {
       "epoch": 11.29,
-      "learning_rate": 0.00019170264965652598,
-      "loss": 0.153,
+      "learning_rate": 0.00023564431486880464,
+      "loss": 0.07,
       "step": 7860
     },
     {
       "epoch": 11.31,
-      "learning_rate": 0.0001915554465161923,
-      "loss": 0.1413,
+      "learning_rate": 0.00023555685131195332,
+      "loss": 0.0636,
       "step": 7870
     },
     {
       "epoch": 11.32,
-      "learning_rate": 0.00019140824337585867,
-      "loss": 0.1203,
+      "learning_rate": 0.000235469387755102,
+      "loss": 0.0522,
       "step": 7880
     },
     {
       "epoch": 11.34,
-      "learning_rate": 0.00019126104023552503,
-      "loss": 0.1459,
+      "learning_rate": 0.0002353819241982507,
+      "loss": 0.0678,
       "step": 7890
     },
     {
       "epoch": 11.35,
-      "learning_rate": 0.00019111383709519134,
-      "loss": 0.1289,
+      "learning_rate": 0.0002352944606413994,
+      "loss": 0.0532,
       "step": 7900
     },
     {
       "epoch": 11.35,
-      "eval_loss": 0.18372580409049988,
-      "eval_runtime": 40.1863,
-      "eval_samples_per_second": 11.372,
-      "eval_steps_per_second": 1.443,
-      "eval_wer": 0.20629423279626563,
+      "eval_loss": 0.17522485554218292,
+      "eval_runtime": 39.7919,
+      "eval_samples_per_second": 11.485,
+      "eval_steps_per_second": 1.458,
+      "eval_wer": 0.17000451739195904,
       "step": 7900
     },
     {
       "epoch": 11.36,
-      "learning_rate": 0.00019096663395485767,
-      "loss": 0.138,
+      "learning_rate": 0.00023520699708454808,
+      "loss": 0.0525,
       "step": 7910
     },
     {
       "epoch": 11.38,
-      "learning_rate": 0.00019081943081452403,
-      "loss": 0.1593,
+      "learning_rate": 0.0002351195335276968,
+      "loss": 0.0693,
       "step": 7920
     },
     {
       "epoch": 11.39,
-      "learning_rate": 0.0001906722276741904,
-      "loss": 0.1145,
+      "learning_rate": 0.00023503206997084548,
+      "loss": 0.0508,
       "step": 7930
     },
     {
       "epoch": 11.41,
-      "learning_rate": 0.0001905250245338567,
-      "loss": 0.1582,
+      "learning_rate": 0.00023494460641399417,
+      "loss": 0.0626,
       "step": 7940
     },
     {
       "epoch": 11.42,
-      "learning_rate": 0.00019037782139352303,
-      "loss": 0.1597,
+      "learning_rate": 0.00023485714285714283,
+      "loss": 0.06,
       "step": 7950
     },
     {
       "epoch": 11.44,
-      "learning_rate": 0.0001902306182531894,
-      "loss": 0.1344,
+      "learning_rate": 0.00023476967930029152,
+      "loss": 0.0545,
       "step": 7960
     },
     {
       "epoch": 11.45,
-      "learning_rate": 0.0001900834151128557,
-      "loss": 0.1387,
+      "learning_rate": 0.0002346822157434402,
+      "loss": 0.0621,
       "step": 7970
     },
     {
       "epoch": 11.47,
-      "learning_rate": 0.00018993621197252206,
-      "loss": 0.1321,
+      "learning_rate": 0.0002345947521865889,
+      "loss": 0.0467,
       "step": 7980
     },
     {
       "epoch": 11.48,
-      "learning_rate": 0.00018980372914622178,
-      "loss": 0.1783,
+      "learning_rate": 0.00023450728862973758,
+      "loss": 0.0839,
       "step": 7990
     },
     {
       "epoch": 11.49,
-      "learning_rate": 0.00018965652600588809,
-      "loss": 0.1268,
+      "learning_rate": 0.0002344198250728863,
+      "loss": 0.0557,
       "step": 8000
     },
     {
       "epoch": 11.49,
-      "eval_loss": 0.17705726623535156,
-      "eval_runtime": 39.9982,
-      "eval_samples_per_second": 11.426,
-      "eval_steps_per_second": 1.45,
-      "eval_wer": 0.20358379762084022,
+      "eval_loss": 0.18215720355510712,
+      "eval_runtime": 39.894,
+      "eval_samples_per_second": 11.455,
+      "eval_steps_per_second": 1.454,
+      "eval_wer": 0.1724137931034483,
       "step": 8000
     },
     {
       "epoch": 11.51,
-      "learning_rate": 0.00018950932286555445,
-      "loss": 0.1468,
+      "learning_rate": 0.00023433236151603498,
+      "loss": 0.0648,
       "step": 8010
     },
     {
       "epoch": 11.52,
-      "learning_rate": 0.0001893621197252208,
-      "loss": 0.1278,
+      "learning_rate": 0.00023424489795918367,
+      "loss": 0.0612,
       "step": 8020
     },
     {
       "epoch": 11.54,
-      "learning_rate": 0.00018921491658488714,
-      "loss": 0.1432,
+      "learning_rate": 0.00023415743440233236,
+      "loss": 0.0621,
       "step": 8030
     },
     {
       "epoch": 11.55,
-      "learning_rate": 0.00018906771344455345,
-      "loss": 0.1383,
+      "learning_rate": 0.00023406997084548102,
+      "loss": 0.065,
       "step": 8040
     },
     {
       "epoch": 11.57,
-      "learning_rate": 0.0001889205103042198,
-      "loss": 0.1267,
+      "learning_rate": 0.0002339825072886297,
+      "loss": 0.0541,
       "step": 8050
     },
     {
       "epoch": 11.58,
-      "learning_rate": 0.00018877330716388617,
-      "loss": 0.1315,
+      "learning_rate": 0.0002338950437317784,
+      "loss": 0.0633,
       "step": 8060
     },
     {
       "epoch": 11.59,
-      "learning_rate": 0.00018862610402355247,
-      "loss": 0.1618,
+      "learning_rate": 0.00023380758017492708,
+      "loss": 0.0753,
       "step": 8070
     },
     {
       "epoch": 11.61,
-      "learning_rate": 0.0001884789008832188,
-      "loss": 0.144,
+      "learning_rate": 0.00023372011661807577,
+      "loss": 0.0683,
       "step": 8080
     },
     {
       "epoch": 11.62,
-      "learning_rate": 0.00018833169774288517,
-      "loss": 0.1267,
+      "learning_rate": 0.00023363265306122448,
+      "loss": 0.0608,
       "step": 8090
     },
     {
       "epoch": 11.64,
-      "learning_rate": 0.00018818449460255153,
-      "loss": 0.1335,
+      "learning_rate": 0.00023354518950437317,
+      "loss": 0.0532,
       "step": 8100
     },
     {
       "epoch": 11.64,
-      "eval_loss": 0.1754847913980484,
-      "eval_runtime": 39.9481,
-      "eval_samples_per_second": 11.44,
-      "eval_steps_per_second": 1.452,
-      "eval_wer": 0.20087336244541484,
+      "eval_loss": 0.17462661862373352,
+      "eval_runtime": 39.9062,
+      "eval_samples_per_second": 11.452,
+      "eval_steps_per_second": 1.453,
+      "eval_wer": 0.16202379159765096,
       "step": 8100
     },
     {
       "epoch": 11.65,
-      "learning_rate": 0.00018803729146221783,
-      "loss": 0.1241,
+      "learning_rate": 0.00023345772594752186,
+      "loss": 0.0574,
       "step": 8110
     },
     {
       "epoch": 11.67,
-      "learning_rate": 0.00018789008832188417,
-      "loss": 0.1458,
+      "learning_rate": 0.00023337026239067055,
+      "loss": 0.0694,
       "step": 8120
     },
     {
       "epoch": 11.68,
-      "learning_rate": 0.00018774288518155053,
-      "loss": 0.1269,
+      "learning_rate": 0.0002332827988338192,
+      "loss": 0.0531,
       "step": 8130
     },
     {
       "epoch": 11.7,
-      "learning_rate": 0.0001875956820412169,
-      "loss": 0.1547,
+      "learning_rate": 0.0002331953352769679,
+      "loss": 0.0761,
       "step": 8140
     },
     {
       "epoch": 11.71,
-      "learning_rate": 0.0001874484789008832,
-      "loss": 0.138,
+      "learning_rate": 0.00023310787172011659,
+      "loss": 0.06,
       "step": 8150
     },
     {
       "epoch": 11.72,
-      "learning_rate": 0.00018730127576054953,
-      "loss": 0.1513,
+      "learning_rate": 0.00023302040816326527,
+      "loss": 0.0786,
       "step": 8160
     },
     {
       "epoch": 11.74,
-      "learning_rate": 0.0001871540726202159,
-      "loss": 0.1523,
+      "learning_rate": 0.00023293294460641396,
+      "loss": 0.0744,
       "step": 8170
     },
     {
       "epoch": 11.75,
-      "learning_rate": 0.00018700686947988225,
-      "loss": 0.1463,
+      "learning_rate": 0.00023284548104956268,
+      "loss": 0.0621,
       "step": 8180
     },
     {
       "epoch": 11.77,
-      "learning_rate": 0.00018685966633954856,
-      "loss": 0.1541,
+      "learning_rate": 0.00023275801749271136,
+      "loss": 0.0738,
       "step": 8190
     },
     {
       "epoch": 11.78,
-      "learning_rate": 0.0001867124631992149,
-      "loss": 0.1136,
+      "learning_rate": 0.00023267055393586005,
+      "loss": 0.0492,
       "step": 8200
     },
     {
       "epoch": 11.78,
-      "eval_loss": 0.19131112098693848,
-      "eval_runtime": 40.2824,
-      "eval_samples_per_second": 11.345,
-      "eval_steps_per_second": 1.44,
-      "eval_wer": 0.205089594940521,
+      "eval_loss": 0.17493364214897156,
+      "eval_runtime": 39.8456,
+      "eval_samples_per_second": 11.469,
+      "eval_steps_per_second": 1.456,
+      "eval_wer": 0.1630778497214275,
       "step": 8200
     },
     {
       "epoch": 11.8,
-      "learning_rate": 0.00018656526005888125,
-      "loss": 0.1186,
+      "learning_rate": 0.00023258309037900874,
+      "loss": 0.0569,
       "step": 8210
     },
     {
       "epoch": 11.81,
-      "learning_rate": 0.00018641805691854756,
-      "loss": 0.151,
+      "learning_rate": 0.0002324956268221574,
+      "loss": 0.0762,
       "step": 8220
     },
     {
       "epoch": 11.82,
-      "learning_rate": 0.00018627085377821392,
-      "loss": 0.1311,
+      "learning_rate": 0.0002324081632653061,
+      "loss": 0.0511,
       "step": 8230
     },
     {
       "epoch": 11.84,
-      "learning_rate": 0.00018612365063788025,
-      "loss": 0.1507,
+      "learning_rate": 0.00023232069970845478,
+      "loss": 0.0685,
       "step": 8240
     },
     {
       "epoch": 11.85,
-      "learning_rate": 0.0001859764474975466,
-      "loss": 0.1305,
+      "learning_rate": 0.00023223323615160346,
+      "loss": 0.0643,
       "step": 8250
     },
     {
       "epoch": 11.87,
-      "learning_rate": 0.00018582924435721292,
-      "loss": 0.1331,
+      "learning_rate": 0.00023214577259475218,
+      "loss": 0.0654,
       "step": 8260
     },
     {
       "epoch": 11.88,
-      "learning_rate": 0.00018568204121687928,
-      "loss": 0.1288,
+      "learning_rate": 0.00023205830903790087,
+      "loss": 0.0624,
       "step": 8270
     },
     {
       "epoch": 11.9,
-      "learning_rate": 0.0001855348380765456,
-      "loss": 0.1305,
+      "learning_rate": 0.00023197084548104955,
+      "loss": 0.0555,
       "step": 8280
     },
     {
       "epoch": 11.91,
-      "learning_rate": 0.00018538763493621197,
-      "loss": 0.1492,
+      "learning_rate": 0.00023188338192419824,
+      "loss": 0.0715,
       "step": 8290
     },
     {
       "epoch": 11.93,
-      "learning_rate": 0.00018524043179587828,
-      "loss": 0.1321,
+      "learning_rate": 0.00023179591836734693,
+      "loss": 0.0606,
       "step": 8300
     },
     {
       "epoch": 11.93,
-      "eval_loss": 0.1786966621875763,
-      "eval_runtime": 39.7362,
-      "eval_samples_per_second": 11.501,
-      "eval_steps_per_second": 1.46,
-      "eval_wer": 0.201174521909351,
+      "eval_loss": 0.1753462255001068,
+      "eval_runtime": 39.9815,
+      "eval_samples_per_second": 11.43,
+      "eval_steps_per_second": 1.451,
+      "eval_wer": 0.1683481403403102,
       "step": 8300
     },
     {
       "epoch": 11.94,
-      "learning_rate": 0.00018509322865554464,
-      "loss": 0.1387,
+      "learning_rate": 0.0002317084548104956,
+      "loss": 0.0608,
       "step": 8310
     },
     {
       "epoch": 11.95,
-      "learning_rate": 0.00018494602551521097,
-      "loss": 0.1183,
+      "learning_rate": 0.00023162099125364428,
+      "loss": 0.062,
       "step": 8320
     },
     {
       "epoch": 11.97,
-      "learning_rate": 0.00018479882237487733,
-      "loss": 0.1429,
+      "learning_rate": 0.00023153352769679297,
+      "loss": 0.0617,
       "step": 8330
     },
     {
       "epoch": 11.98,
-      "learning_rate": 0.00018465161923454364,
-      "loss": 0.142,
+      "learning_rate": 0.00023144606413994166,
+      "loss": 0.0719,
       "step": 8340
     },
     {
       "epoch": 12.0,
-      "learning_rate": 0.00018450441609421,
-      "loss": 0.1511,
+      "learning_rate": 0.00023135860058309037,
+      "loss": 0.0741,
       "step": 8350
     },
     {
       "epoch": 12.01,
-      "learning_rate": 0.00018435721295387633,
-      "loss": 0.1594,
+      "learning_rate": 0.00023127113702623906,
+      "loss": 0.0808,
       "step": 8360
     },
     {
       "epoch": 12.03,
-      "learning_rate": 0.00018421000981354266,
-      "loss": 0.1273,
+      "learning_rate": 0.00023118367346938775,
+      "loss": 0.062,
       "step": 8370
     },
     {
       "epoch": 12.04,
-      "learning_rate": 0.000184062806673209,
-      "loss": 0.1324,
+      "learning_rate": 0.00023109620991253643,
+      "loss": 0.0636,
       "step": 8380
     },
     {
       "epoch": 12.05,
-      "learning_rate": 0.00018391560353287536,
-      "loss": 0.1219,
+      "learning_rate": 0.00023100874635568512,
+      "loss": 0.0585,
       "step": 8390
     },
     {
       "epoch": 12.07,
-      "learning_rate": 0.0001837684003925417,
-      "loss": 0.1088,
+      "learning_rate": 0.00023092128279883378,
+      "loss": 0.0523,
       "step": 8400
     },
     {
       "epoch": 12.07,
-      "eval_loss": 0.19158530235290527,
-      "eval_runtime": 39.7742,
-      "eval_samples_per_second": 11.49,
-      "eval_steps_per_second": 1.458,
-      "eval_wer": 0.19966872458967022,
+      "eval_loss": 0.17258815467357635,
+      "eval_runtime": 39.8367,
+      "eval_samples_per_second": 11.472,
+      "eval_steps_per_second": 1.456,
+      "eval_wer": 0.1624755307935552,
       "step": 8400
     },
     {
       "epoch": 12.08,
-      "learning_rate": 0.00018362119725220803,
-      "loss": 0.1208,
+      "learning_rate": 0.00023083381924198247,
+      "loss": 0.064,
       "step": 8410
     },
     {
       "epoch": 12.1,
-      "learning_rate": 0.00018347399411187436,
-      "loss": 0.1286,
+      "learning_rate": 0.00023074635568513116,
+      "loss": 0.0615,
       "step": 8420
     },
     {
       "epoch": 12.11,
-      "learning_rate": 0.00018332679097154072,
-      "loss": 0.1394,
+      "learning_rate": 0.00023065889212827985,
+      "loss": 0.0614,
       "step": 8430
     },
     {
       "epoch": 12.13,
-      "learning_rate": 0.00018317958783120705,
-      "loss": 0.1601,
+      "learning_rate": 0.00023057142857142856,
+      "loss": 0.0811,
       "step": 8440
     },
     {
       "epoch": 12.14,
-      "learning_rate": 0.00018303238469087339,
-      "loss": 0.1211,
+      "learning_rate": 0.00023048396501457725,
+      "loss": 0.0522,
       "step": 8450
     },
     {
       "epoch": 12.16,
-      "learning_rate": 0.00018288518155053972,
-      "loss": 0.1519,
+      "learning_rate": 0.00023039650145772594,
+      "loss": 0.0809,
       "step": 8460
     },
     {
       "epoch": 12.17,
-      "learning_rate": 0.00018273797841020608,
-      "loss": 0.1015,
+      "learning_rate": 0.00023030903790087462,
+      "loss": 0.0437,
       "step": 8470
     },
     {
       "epoch": 12.18,
-      "learning_rate": 0.0001825907752698724,
-      "loss": 0.1214,
+      "learning_rate": 0.0002302215743440233,
+      "loss": 0.0546,
       "step": 8480
     },
     {
       "epoch": 12.2,
-      "learning_rate": 0.00018244357212953875,
-      "loss": 0.1315,
+      "learning_rate": 0.00023013411078717197,
+      "loss": 0.0621,
       "step": 8490
     },
     {
       "epoch": 12.21,
-      "learning_rate": 0.00018229636898920508,
-      "loss": 0.1319,
+      "learning_rate": 0.00023004664723032066,
+      "loss": 0.0577,
       "step": 8500
     },
     {
       "epoch": 12.21,
-      "eval_loss": 0.178669735789299,
-      "eval_runtime": 39.4909,
-      "eval_samples_per_second": 11.572,
-      "eval_steps_per_second": 1.469,
-      "eval_wer": 0.20252973949706368,
+      "eval_loss": 0.17561942338943481,
+      "eval_runtime": 39.9824,
+      "eval_samples_per_second": 11.43,
+      "eval_steps_per_second": 1.451,
+      "eval_wer": 0.17015509712392712,
       "step": 8500
     },
     {
       "epoch": 12.23,
-      "learning_rate": 0.00018214916584887144,
-      "loss": 0.1498,
+      "learning_rate": 0.00022995918367346935,
+      "loss": 0.07,
       "step": 8510
     },
     {
       "epoch": 12.24,
-      "learning_rate": 0.00018200196270853775,
-      "loss": 0.1089,
+      "learning_rate": 0.00022987172011661806,
+      "loss": 0.0501,
       "step": 8520
     },
     {
       "epoch": 12.26,
-      "learning_rate": 0.0001818547595682041,
-      "loss": 0.1212,
+      "learning_rate": 0.00022978425655976675,
+      "loss": 0.0515,
       "step": 8530
     },
     {
       "epoch": 12.27,
-      "learning_rate": 0.00018170755642787044,
-      "loss": 0.1483,
+      "learning_rate": 0.00022969679300291544,
+      "loss": 0.0658,
       "step": 8540
     },
     {
       "epoch": 12.28,
-      "learning_rate": 0.0001815603532875368,
-      "loss": 0.1295,
+      "learning_rate": 0.00022960932944606413,
+      "loss": 0.05,
       "step": 8550
     },
     {
       "epoch": 12.3,
-      "learning_rate": 0.0001814131501472031,
-      "loss": 0.1221,
+      "learning_rate": 0.00022952186588921282,
+      "loss": 0.0559,
       "step": 8560
     },
     {
       "epoch": 12.31,
-      "learning_rate": 0.00018126594700686947,
-      "loss": 0.1058,
+      "learning_rate": 0.0002294344023323615,
+      "loss": 0.0461,
       "step": 8570
     },
     {
       "epoch": 12.33,
-      "learning_rate": 0.0001811187438665358,
-      "loss": 0.1259,
+      "learning_rate": 0.00022934693877551016,
+      "loss": 0.0654,
       "step": 8580
     },
     {
       "epoch": 12.34,
-      "learning_rate": 0.00018097154072620216,
-      "loss": 0.1267,
+      "learning_rate": 0.00022925947521865885,
+      "loss": 0.0583,
       "step": 8590
     },
     {
       "epoch": 12.36,
-      "learning_rate": 0.00018082433758586847,
-      "loss": 0.1274,
+      "learning_rate": 0.00022917201166180754,
+      "loss": 0.061,
       "step": 8600
     },
     {
       "epoch": 12.36,
-      "eval_loss": 0.18227306008338928,
-      "eval_runtime": 40.0342,
-      "eval_samples_per_second": 11.415,
-      "eval_steps_per_second": 1.449,
-      "eval_wer": 0.2044872760126487,
+      "eval_loss": 0.16998374462127686,
+      "eval_runtime": 39.9068,
+      "eval_samples_per_second": 11.452,
+      "eval_steps_per_second": 1.453,
+      "eval_wer": 0.16578828489685288,
       "step": 8600
     },
     {
       "epoch": 12.37,
-      "learning_rate": 0.00018067713444553483,
-      "loss": 0.1318,
+      "learning_rate": 0.00022908454810495626,
+      "loss": 0.067,
       "step": 8610
     },
     {
       "epoch": 12.39,
-      "learning_rate": 0.00018052993130520116,
-      "loss": 0.1286,
+      "learning_rate": 0.00022899708454810494,
+      "loss": 0.0637,
       "step": 8620
     },
     {
       "epoch": 12.4,
-      "learning_rate": 0.0001803827281648675,
-      "loss": 0.1115,
+      "learning_rate": 0.00022890962099125363,
+      "loss": 0.0517,
       "step": 8630
     },
     {
       "epoch": 12.41,
-      "learning_rate": 0.00018023552502453383,
-      "loss": 0.1283,
+      "learning_rate": 0.00022882215743440232,
+      "loss": 0.0598,
       "step": 8640
     },
     {
       "epoch": 12.43,
-      "learning_rate": 0.0001800883218842002,
-      "loss": 0.1297,
+      "learning_rate": 0.000228734693877551,
+      "loss": 0.0598,
       "step": 8650
     },
     {
       "epoch": 12.44,
-      "learning_rate": 0.00017994111874386652,
-      "loss": 0.1287,
+      "learning_rate": 0.0002286472303206997,
+      "loss": 0.0656,
       "step": 8660
     },
     {
       "epoch": 12.46,
-      "learning_rate": 0.00017979391560353286,
-      "loss": 0.1334,
+      "learning_rate": 0.00022855976676384836,
+      "loss": 0.0699,
       "step": 8670
     },
     {
       "epoch": 12.47,
-      "learning_rate": 0.0001796467124631992,
-      "loss": 0.1264,
+      "learning_rate": 0.00022847230320699704,
+      "loss": 0.0524,
       "step": 8680
     },
     {
       "epoch": 12.49,
-      "learning_rate": 0.00017949950932286555,
-      "loss": 0.1287,
+      "learning_rate": 0.00022838483965014576,
+      "loss": 0.0557,
       "step": 8690
     },
     {
       "epoch": 12.5,
-      "learning_rate": 0.00017935230618253188,
-      "loss": 0.1361,
+      "learning_rate": 0.00022829737609329445,
+      "loss": 0.0546,
       "step": 8700
     },
     {
       "epoch": 12.5,
-      "eval_loss": 0.1859862059354782,
-      "eval_runtime": 39.6559,
-      "eval_samples_per_second": 11.524,
-      "eval_steps_per_second": 1.463,
-      "eval_wer": 0.20749887065201025,
+      "eval_loss": 0.16951903700828552,
+      "eval_runtime": 39.9127,
+      "eval_samples_per_second": 11.45,
+      "eval_steps_per_second": 1.453,
+      "eval_wer": 0.15750639963860863,
       "step": 8700
     },
     {
       "epoch": 12.51,
-      "learning_rate": 0.00017920510304219822,
-      "loss": 0.1718,
+      "learning_rate": 0.00022820991253644313,
+      "loss": 0.0841,
       "step": 8710
     },
     {
       "epoch": 12.53,
-      "learning_rate": 0.00017905789990186455,
-      "loss": 0.1127,
+      "learning_rate": 0.00022812244897959182,
+      "loss": 0.0452,
       "step": 8720
     },
     {
       "epoch": 12.54,
-      "learning_rate": 0.0001789106967615309,
-      "loss": 0.1558,
+      "learning_rate": 0.0002280349854227405,
+      "loss": 0.0674,
       "step": 8730
     },
     {
       "epoch": 12.56,
-      "learning_rate": 0.00017876349362119724,
-      "loss": 0.1398,
+      "learning_rate": 0.0002279475218658892,
+      "loss": 0.0652,
       "step": 8740
     },
     {
       "epoch": 12.57,
-      "learning_rate": 0.00017861629048086358,
-      "loss": 0.1351,
+      "learning_rate": 0.00022786005830903789,
+      "loss": 0.0592,
       "step": 8750
     },
     {
       "epoch": 12.59,
-      "learning_rate": 0.0001784690873405299,
-      "loss": 0.1291,
+      "learning_rate": 0.00022777259475218655,
+      "loss": 0.0628,
       "step": 8760
     },
     {
       "epoch": 12.6,
-      "learning_rate": 0.00017832188420019627,
-      "loss": 0.1339,
+      "learning_rate": 0.00022768513119533523,
+      "loss": 0.0595,
       "step": 8770
     },
     {
       "epoch": 12.61,
-      "learning_rate": 0.00017817468105986258,
-      "loss": 0.152,
+      "learning_rate": 0.00022759766763848395,
+      "loss": 0.0718,
       "step": 8780
     },
     {
       "epoch": 12.63,
-      "learning_rate": 0.00017802747791952894,
-      "loss": 0.1313,
+      "learning_rate": 0.00022751020408163264,
+      "loss": 0.0636,
       "step": 8790
     },
     {
       "epoch": 12.64,
-      "learning_rate": 0.00017788027477919527,
-      "loss": 0.1541,
+      "learning_rate": 0.00022742274052478133,
+      "loss": 0.0692,
       "step": 8800
     },
     {
       "epoch": 12.64,
-      "eval_loss": 0.1849740892648697,
-      "eval_runtime": 39.6506,
-      "eval_samples_per_second": 11.526,
-      "eval_steps_per_second": 1.463,
-      "eval_wer": 0.20283089896099984,
+      "eval_loss": 0.16482137143611908,
+      "eval_runtime": 39.9004,
+      "eval_samples_per_second": 11.454,
+      "eval_steps_per_second": 1.454,
+      "eval_wer": 0.1644330673091402,
       "step": 8800
     },
     {
       "epoch": 12.66,
-      "learning_rate": 0.00017773307163886163,
-      "loss": 0.1601,
+      "learning_rate": 0.00022733527696793,
+      "loss": 0.0734,
       "step": 8810
     },
     {
       "epoch": 12.67,
-      "learning_rate": 0.00017758586849852794,
-      "loss": 0.123,
+      "learning_rate": 0.0002272478134110787,
+      "loss": 0.0566,
       "step": 8820
     },
     {
       "epoch": 12.69,
-      "learning_rate": 0.0001774386653581943,
-      "loss": 0.134,
+      "learning_rate": 0.0002271603498542274,
+      "loss": 0.0693,
       "step": 8830
     },
     {
       "epoch": 12.7,
-      "learning_rate": 0.00017729146221786063,
-      "loss": 0.1273,
+      "learning_rate": 0.00022707288629737608,
+      "loss": 0.057,
       "step": 8840
     },
     {
       "epoch": 12.72,
-      "learning_rate": 0.000177144259077527,
-      "loss": 0.1085,
+      "learning_rate": 0.00022698542274052474,
+      "loss": 0.0512,
       "step": 8850
     },
     {
       "epoch": 12.73,
-      "learning_rate": 0.0001769970559371933,
-      "loss": 0.1438,
+      "learning_rate": 0.00022689795918367343,
+      "loss": 0.0698,
       "step": 8860
     },
     {
       "epoch": 12.74,
-      "learning_rate": 0.00017684985279685966,
-      "loss": 0.1215,
+      "learning_rate": 0.00022681049562682214,
+      "loss": 0.0621,
       "step": 8870
     },
     {
       "epoch": 12.76,
-      "learning_rate": 0.000176702649656526,
-      "loss": 0.1301,
+      "learning_rate": 0.00022672303206997083,
+      "loss": 0.0553,
       "step": 8880
     },
     {
       "epoch": 12.77,
-      "learning_rate": 0.00017655544651619235,
-      "loss": 0.118,
+      "learning_rate": 0.00022663556851311952,
+      "loss": 0.058,
       "step": 8890
     },
     {
       "epoch": 12.79,
-      "learning_rate": 0.00017640824337585866,
-      "loss": 0.0957,
+      "learning_rate": 0.0002265481049562682,
+      "loss": 0.0497,
       "step": 8900
     },
     {
       "epoch": 12.79,
-      "eval_loss": 0.1861564964056015,
-      "eval_runtime": 39.7711,
-      "eval_samples_per_second": 11.491,
-      "eval_steps_per_second": 1.458,
-      "eval_wer": 0.19469959343472368,
+      "eval_loss": 0.15961284935474396,
+      "eval_runtime": 39.8903,
+      "eval_samples_per_second": 11.456,
+      "eval_steps_per_second": 1.454,
+      "eval_wer": 0.16187321186568288,
       "step": 8900
     },
     {
       "epoch": 12.8,
-      "learning_rate": 0.00017626104023552502,
-      "loss": 0.1277,
+      "learning_rate": 0.0002264606413994169,
+      "loss": 0.0633,
       "step": 8910
     },
     {
       "epoch": 12.82,
-      "learning_rate": 0.00017611383709519135,
-      "loss": 0.1311,
+      "learning_rate": 0.00022637317784256558,
+      "loss": 0.0585,
       "step": 8920
     },
     {
       "epoch": 12.83,
-      "learning_rate": 0.00017596663395485766,
-      "loss": 0.1545,
+      "learning_rate": 0.00022628571428571427,
+      "loss": 0.0719,
       "step": 8930
     },
     {
       "epoch": 12.84,
-      "learning_rate": 0.00017581943081452402,
-      "loss": 0.1182,
+      "learning_rate": 0.00022619825072886293,
+      "loss": 0.056,
       "step": 8940
     },
     {
       "epoch": 12.86,
-      "learning_rate": 0.00017567222767419038,
-      "loss": 0.12,
+      "learning_rate": 0.00022611078717201164,
+      "loss": 0.0473,
       "step": 8950
     },
     {
       "epoch": 12.87,
-      "learning_rate": 0.00017552502453385671,
-      "loss": 0.1413,
+      "learning_rate": 0.00022602332361516033,
+      "loss": 0.066,
       "step": 8960
     },
     {
       "epoch": 12.89,
-      "learning_rate": 0.00017537782139352302,
-      "loss": 0.1131,
+      "learning_rate": 0.00022593586005830902,
+      "loss": 0.058,
       "step": 8970
     },
     {
       "epoch": 12.9,
-      "learning_rate": 0.00017523061825318938,
-      "loss": 0.124,
+      "learning_rate": 0.0002258483965014577,
+      "loss": 0.059,
       "step": 8980
     },
     {
       "epoch": 12.92,
-      "learning_rate": 0.00017508341511285574,
-      "loss": 0.122,
+      "learning_rate": 0.0002257609329446064,
+      "loss": 0.0628,
       "step": 8990
     },
     {
       "epoch": 12.93,
-      "learning_rate": 0.00017493621197252208,
-      "loss": 0.1139,
+      "learning_rate": 0.00022567346938775508,
+      "loss": 0.0524,
       "step": 9000
     },
     {
       "epoch": 12.93,
-      "eval_loss": 0.17861497402191162,
-      "eval_runtime": 39.712,
-      "eval_samples_per_second": 11.508,
-      "eval_steps_per_second": 1.461,
-      "eval_wer": 0.18310495407318175,
+      "eval_loss": 0.1611989289522171,
+      "eval_runtime": 39.9555,
+      "eval_samples_per_second": 11.438,
+      "eval_steps_per_second": 1.452,
+      "eval_wer": 0.15946393615419366,
       "step": 9000
     },
     {
       "epoch": 12.95,
-      "learning_rate": 0.00017478900883218838,
-      "loss": 0.1459,
+      "learning_rate": 0.00022558600583090377,
+      "loss": 0.0705,
       "step": 9010
     },
     {
       "epoch": 12.96,
-      "learning_rate": 0.00017464180569185474,
-      "loss": 0.1092,
+      "learning_rate": 0.00022549854227405249,
+      "loss": 0.0483,
       "step": 9020
     },
     {
       "epoch": 12.97,
-      "learning_rate": 0.0001744946025515211,
-      "loss": 0.1329,
+      "learning_rate": 0.00022541107871720112,
+      "loss": 0.0626,
       "step": 9030
     },
     {
       "epoch": 12.99,
-      "learning_rate": 0.00017434739941118744,
-      "loss": 0.1047,
+      "learning_rate": 0.00022532361516034984,
+      "loss": 0.0555,
       "step": 9040
     },
     {
       "epoch": 13.0,
-      "learning_rate": 0.00017420019627085374,
-      "loss": 0.1094,
+      "learning_rate": 0.00022524489795918367,
+      "loss": 0.057,
       "step": 9050
     },
     {
       "epoch": 13.02,
-      "learning_rate": 0.0001740529931305201,
-      "loss": 0.1347,
+      "learning_rate": 0.00022515743440233236,
+      "loss": 0.075,
       "step": 9060
     },
     {
       "epoch": 13.03,
-      "learning_rate": 0.00017390578999018646,
-      "loss": 0.1158,
+      "learning_rate": 0.00022506997084548104,
+      "loss": 0.0543,
       "step": 9070
     },
     {
       "epoch": 13.05,
-      "learning_rate": 0.00017375858684985277,
-      "loss": 0.1443,
+      "learning_rate": 0.00022498250728862973,
+      "loss": 0.0728,
       "step": 9080
     },
     {
       "epoch": 13.06,
-      "learning_rate": 0.0001736113837095191,
-      "loss": 0.1027,
+      "learning_rate": 0.0002248950437317784,
+      "loss": 0.0522,
       "step": 9090
     },
     {
       "epoch": 13.07,
-      "learning_rate": 0.00017346418056918546,
-      "loss": 0.107,
+      "learning_rate": 0.00022480758017492708,
+      "loss": 0.0501,
       "step": 9100
     },
     {
       "epoch": 13.07,
-      "eval_loss": 0.18377935886383057,
-      "eval_runtime": 39.9042,
-      "eval_samples_per_second": 11.452,
-      "eval_steps_per_second": 1.453,
-      "eval_wer": 0.1959042312904683,
+      "eval_loss": 0.1783318668603897,
+      "eval_runtime": 39.9433,
+      "eval_samples_per_second": 11.441,
+      "eval_steps_per_second": 1.452,
+      "eval_wer": 0.16413190784520404,
       "step": 9100
     },
     {
       "epoch": 13.09,
-      "learning_rate": 0.00017331697742885182,
-      "loss": 0.1134,
+      "learning_rate": 0.00022472011661807577,
+      "loss": 0.0597,
       "step": 9110
     },
     {
       "epoch": 13.1,
-      "learning_rate": 0.00017316977428851813,
-      "loss": 0.1044,
+      "learning_rate": 0.00022463265306122446,
+      "loss": 0.0493,
       "step": 9120
     },
     {
       "epoch": 13.12,
-      "learning_rate": 0.00017302257114818446,
-      "loss": 0.1194,
+      "learning_rate": 0.00022454518950437317,
+      "loss": 0.0622,
       "step": 9130
     },
     {
       "epoch": 13.13,
-      "learning_rate": 0.00017287536800785082,
-      "loss": 0.1139,
+      "learning_rate": 0.00022445772594752186,
+      "loss": 0.0528,
       "step": 9140
     },
     {
       "epoch": 13.15,
-      "learning_rate": 0.00017272816486751718,
-      "loss": 0.1271,
+      "learning_rate": 0.00022437026239067055,
+      "loss": 0.0592,
       "step": 9150
     },
     {
       "epoch": 13.16,
-      "learning_rate": 0.0001725809617271835,
-      "loss": 0.1207,
+      "learning_rate": 0.00022428279883381923,
+      "loss": 0.066,
       "step": 9160
     },
     {
       "epoch": 13.18,
-      "learning_rate": 0.00017243375858684982,
-      "loss": 0.117,
+      "learning_rate": 0.00022419533527696792,
+      "loss": 0.0549,
       "step": 9170
     },
     {
       "epoch": 13.19,
-      "learning_rate": 0.00017228655544651618,
-      "loss": 0.1332,
+      "learning_rate": 0.00022410787172011658,
+      "loss": 0.0652,
       "step": 9180
     },
     {
       "epoch": 13.2,
-      "learning_rate": 0.00017213935230618255,
-      "loss": 0.1186,
+      "learning_rate": 0.00022402040816326527,
+      "loss": 0.0613,
       "step": 9190
     },
     {
       "epoch": 13.22,
-      "learning_rate": 0.00017199214916584885,
-      "loss": 0.1103,
+      "learning_rate": 0.00022393294460641396,
+      "loss": 0.0539,
       "step": 9200
     },
     {
       "epoch": 13.22,
-      "eval_loss": 0.1748352199792862,
-      "eval_runtime": 39.9358,
-      "eval_samples_per_second": 11.443,
-      "eval_steps_per_second": 1.452,
-      "eval_wer": 0.18732118656828792,
+      "eval_loss": 0.16895711421966553,
+      "eval_runtime": 39.9239,
+      "eval_samples_per_second": 11.447,
+      "eval_steps_per_second": 1.453,
+      "eval_wer": 0.16217437132961904,
       "step": 9200
     },
     {
       "epoch": 13.23,
-      "learning_rate": 0.00017184494602551518,
-      "loss": 0.1177,
+      "learning_rate": 0.00022384548104956265,
+      "loss": 0.057,
       "step": 9210
     },
     {
       "epoch": 13.25,
-      "learning_rate": 0.00017169774288518155,
-      "loss": 0.1022,
+      "learning_rate": 0.00022375801749271136,
+      "loss": 0.0491,
       "step": 9220
     },
     {
       "epoch": 13.26,
-      "learning_rate": 0.00017155053974484785,
-      "loss": 0.1417,
+      "learning_rate": 0.00022367055393586005,
+      "loss": 0.0761,
       "step": 9230
     },
     {
       "epoch": 13.28,
-      "learning_rate": 0.0001714033366045142,
-      "loss": 0.1326,
+      "learning_rate": 0.00022358309037900874,
+      "loss": 0.063,
       "step": 9240
     },
     {
       "epoch": 13.29,
-      "learning_rate": 0.00017125613346418055,
-      "loss": 0.1207,
+      "learning_rate": 0.00022349562682215743,
+      "loss": 0.0537,
       "step": 9250
     },
     {
       "epoch": 13.3,
-      "learning_rate": 0.0001711089303238469,
-      "loss": 0.1132,
+      "learning_rate": 0.00022340816326530611,
+      "loss": 0.0535,
       "step": 9260
     },
     {
       "epoch": 13.32,
-      "learning_rate": 0.0001709617271835132,
-      "loss": 0.097,
+      "learning_rate": 0.00022332069970845477,
+      "loss": 0.0475,
       "step": 9270
     },
     {
       "epoch": 13.33,
-      "learning_rate": 0.00017081452404317957,
-      "loss": 0.1492,
+      "learning_rate": 0.00022323323615160346,
+      "loss": 0.0702,
       "step": 9280
     },
     {
       "epoch": 13.35,
-      "learning_rate": 0.0001706673209028459,
-      "loss": 0.1272,
+      "learning_rate": 0.00022314577259475215,
+      "loss": 0.0617,
       "step": 9290
     },
     {
       "epoch": 13.36,
-      "learning_rate": 0.00017052011776251227,
-      "loss": 0.1274,
+      "learning_rate": 0.00022305830903790084,
+      "loss": 0.0608,
       "step": 9300
     },
     {
       "epoch": 13.36,
-      "eval_loss": 0.17697954177856445,
-      "eval_runtime": 39.8577,
-      "eval_samples_per_second": 11.466,
-      "eval_steps_per_second": 1.455,
-      "eval_wer": 0.19771118807408522,
+      "eval_loss": 0.1596861034631729,
+      "eval_runtime": 40.0878,
+      "eval_samples_per_second": 11.4,
+      "eval_steps_per_second": 1.447,
+      "eval_wer": 0.1597650956181298,
       "step": 9300
     },
     {
       "epoch": 13.38,
-      "learning_rate": 0.00017037291462217857,
-      "loss": 0.1162,
+      "learning_rate": 0.00022297084548104955,
+      "loss": 0.0602,
       "step": 9310
     },
     {
       "epoch": 13.39,
-      "learning_rate": 0.00017022571148184493,
-      "loss": 0.1054,
+      "learning_rate": 0.00022288338192419824,
+      "loss": 0.0474,
       "step": 9320
     },
     {
       "epoch": 13.41,
-      "learning_rate": 0.00017007850834151127,
-      "loss": 0.1115,
+      "learning_rate": 0.00022279591836734693,
+      "loss": 0.053,
       "step": 9330
     },
     {
       "epoch": 13.42,
-      "learning_rate": 0.00016993130520117763,
-      "loss": 0.103,
+      "learning_rate": 0.00022270845481049562,
+      "loss": 0.0465,
       "step": 9340
     },
     {
       "epoch": 13.43,
-      "learning_rate": 0.00016978410206084393,
-      "loss": 0.1359,
+      "learning_rate": 0.0002226209912536443,
+      "loss": 0.0632,
       "step": 9350
     },
     {
       "epoch": 13.45,
-      "learning_rate": 0.0001696368989205103,
-      "loss": 0.1019,
+      "learning_rate": 0.00022253352769679297,
+      "loss": 0.0514,
       "step": 9360
     },
     {
       "epoch": 13.46,
-      "learning_rate": 0.00016948969578017663,
-      "loss": 0.1107,
+      "learning_rate": 0.00022244606413994165,
+      "loss": 0.0538,
       "step": 9370
     },
     {
       "epoch": 13.48,
-      "learning_rate": 0.00016934249263984296,
-      "loss": 0.126,
+      "learning_rate": 0.00022235860058309034,
+      "loss": 0.0593,
       "step": 9380
     },
     {
       "epoch": 13.49,
-      "learning_rate": 0.0001691952894995093,
-      "loss": 0.1239,
+      "learning_rate": 0.00022227113702623906,
+      "loss": 0.0609,
       "step": 9390
     },
     {
       "epoch": 13.51,
-      "learning_rate": 0.00016904808635917565,
-      "loss": 0.1243,
+      "learning_rate": 0.00022218367346938774,
+      "loss": 0.0627,
       "step": 9400
     },
     {
       "epoch": 13.51,
-      "eval_loss": 0.17389988899230957,
-      "eval_runtime": 39.7378,
-      "eval_samples_per_second": 11.5,
-      "eval_steps_per_second": 1.46,
-      "eval_wer": 0.19349495557897908,
+      "eval_loss": 0.16226713359355927,
+      "eval_runtime": 40.1172,
+      "eval_samples_per_second": 11.392,
+      "eval_steps_per_second": 1.446,
+      "eval_wer": 0.16262611052552325,
       "step": 9400
     },
     {
       "epoch": 13.52,
-      "learning_rate": 0.000168900883218842,
-      "loss": 0.1237,
+      "learning_rate": 0.00022209620991253643,
+      "loss": 0.0696,
       "step": 9410
     },
     {
       "epoch": 13.53,
-      "learning_rate": 0.00016875368007850832,
-      "loss": 0.1058,
+      "learning_rate": 0.00022200874635568512,
+      "loss": 0.048,
       "step": 9420
     },
     {
       "epoch": 13.55,
-      "learning_rate": 0.00016860647693817465,
-      "loss": 0.1313,
+      "learning_rate": 0.0002219212827988338,
+      "loss": 0.0624,
       "step": 9430
     },
     {
       "epoch": 13.56,
-      "learning_rate": 0.00016845927379784102,
-      "loss": 0.1101,
+      "learning_rate": 0.0002218338192419825,
+      "loss": 0.0595,
       "step": 9440
     },
     {
       "epoch": 13.58,
-      "learning_rate": 0.00016831207065750735,
-      "loss": 0.1158,
+      "learning_rate": 0.00022174635568513116,
+      "loss": 0.0583,
       "step": 9450
     },
     {
       "epoch": 13.59,
-      "learning_rate": 0.00016816486751717368,
-      "loss": 0.1189,
+      "learning_rate": 0.00022165889212827984,
+      "loss": 0.0607,
       "step": 9460
     },
     {
       "epoch": 13.61,
-      "learning_rate": 0.00016801766437684002,
-      "loss": 0.1178,
+      "learning_rate": 0.00022157142857142853,
+      "loss": 0.0548,
       "step": 9470
     },
     {
       "epoch": 13.62,
-      "learning_rate": 0.00016787046123650638,
-      "loss": 0.1273,
+      "learning_rate": 0.00022148396501457725,
+      "loss": 0.0647,
       "step": 9480
     },
     {
       "epoch": 13.64,
-      "learning_rate": 0.0001677232580961727,
-      "loss": 0.1302,
+      "learning_rate": 0.00022139650145772594,
+      "loss": 0.0683,
       "step": 9490
     },
     {
       "epoch": 13.65,
-      "learning_rate": 0.00016757605495583904,
-      "loss": 0.0975,
+      "learning_rate": 0.00022130903790087462,
+      "loss": 0.0456,
       "step": 9500
     },
     {
       "epoch": 13.65,
-      "eval_loss": 0.17656183242797852,
-      "eval_runtime": 39.8815,
-      "eval_samples_per_second": 11.459,
-      "eval_steps_per_second": 1.454,
-      "eval_wer": 0.19816292726998946,
+      "eval_loss": 0.16573207080364227,
+      "eval_runtime": 40.0704,
+      "eval_samples_per_second": 11.405,
+      "eval_steps_per_second": 1.447,
+      "eval_wer": 0.16217437132961904,
       "step": 9500
     },
     {
       "epoch": 13.66,
-      "learning_rate": 0.00016742885181550538,
-      "loss": 0.1057,
+      "learning_rate": 0.0002212215743440233,
+      "loss": 0.0506,
       "step": 9510
     },
     {
       "epoch": 13.68,
-      "learning_rate": 0.00016728164867517174,
-      "loss": 0.1182,
+      "learning_rate": 0.000221134110787172,
+      "loss": 0.0557,
       "step": 9520
     },
     {
       "epoch": 13.69,
-      "learning_rate": 0.00016713444553483804,
-      "loss": 0.111,
+      "learning_rate": 0.0002210466472303207,
+      "loss": 0.0558,
       "step": 9530
     },
     {
       "epoch": 13.71,
-      "learning_rate": 0.0001669872423945044,
-      "loss": 0.128,
+      "learning_rate": 0.00022095918367346935,
+      "loss": 0.0598,
       "step": 9540
     },
     {
       "epoch": 13.72,
-      "learning_rate": 0.00016684003925417074,
-      "loss": 0.095,
+      "learning_rate": 0.00022087172011661804,
+      "loss": 0.0459,
       "step": 9550
     },
     {
       "epoch": 13.74,
-      "learning_rate": 0.0001666928361138371,
-      "loss": 0.1286,
+      "learning_rate": 0.00022078425655976672,
+      "loss": 0.0708,
       "step": 9560
     },
     {
       "epoch": 13.75,
-      "learning_rate": 0.0001665456329735034,
-      "loss": 0.1028,
+      "learning_rate": 0.00022069679300291544,
+      "loss": 0.0507,
       "step": 9570
     },
     {
       "epoch": 13.76,
-      "learning_rate": 0.00016639842983316976,
-      "loss": 0.1336,
+      "learning_rate": 0.00022060932944606413,
+      "loss": 0.0741,
       "step": 9580
     },
     {
       "epoch": 13.78,
-      "learning_rate": 0.0001662512266928361,
-      "loss": 0.095,
+      "learning_rate": 0.00022052186588921281,
+      "loss": 0.0534,
       "step": 9590
     },
     {
       "epoch": 13.79,
-      "learning_rate": 0.00016610402355250246,
-      "loss": 0.1093,
+      "learning_rate": 0.0002204344023323615,
+      "loss": 0.0563,
       "step": 9600
     },
     {
       "epoch": 13.79,
-      "eval_loss": 0.1793922334909439,
-      "eval_runtime": 40.0085,
-      "eval_samples_per_second": 11.423,
-      "eval_steps_per_second": 1.45,
-      "eval_wer": 0.1959042312904683,
+      "eval_loss": 0.1635971963405609,
+      "eval_runtime": 40.1592,
+      "eval_samples_per_second": 11.38,
+      "eval_steps_per_second": 1.444,
+      "eval_wer": 0.16398132811323596,
       "step": 9600
     },
     {
       "epoch": 13.81,
-      "learning_rate": 0.00016595682041216876,
-      "loss": 0.1257,
+      "learning_rate": 0.0002203469387755102,
+      "loss": 0.0676,
       "step": 9610
     },
     {
       "epoch": 13.82,
-      "learning_rate": 0.00016580961727183512,
-      "loss": 0.0955,
+      "learning_rate": 0.00022025947521865888,
+      "loss": 0.044,
       "step": 9620
     },
     {
       "epoch": 13.84,
-      "learning_rate": 0.00016566241413150146,
-      "loss": 0.1234,
+      "learning_rate": 0.00022017201166180754,
+      "loss": 0.0634,
       "step": 9630
     },
     {
       "epoch": 13.85,
-      "learning_rate": 0.00016551521099116782,
-      "loss": 0.0953,
+      "learning_rate": 0.00022008454810495623,
+      "loss": 0.0501,
       "step": 9640
     },
     {
       "epoch": 13.86,
-      "learning_rate": 0.00016536800785083412,
-      "loss": 0.1147,
+      "learning_rate": 0.00021999708454810494,
+      "loss": 0.0556,
       "step": 9650
     },
     {
       "epoch": 13.88,
-      "learning_rate": 0.00016522080471050049,
-      "loss": 0.1232,
+      "learning_rate": 0.00021990962099125363,
+      "loss": 0.0659,
       "step": 9660
     },
     {
       "epoch": 13.89,
-      "learning_rate": 0.00016507360157016682,
-      "loss": 0.1264,
+      "learning_rate": 0.00021982215743440232,
+      "loss": 0.0587,
       "step": 9670
     },
     {
       "epoch": 13.91,
-      "learning_rate": 0.00016492639842983315,
-      "loss": 0.1401,
+      "learning_rate": 0.000219734693877551,
+      "loss": 0.0671,
       "step": 9680
     },
     {
       "epoch": 13.92,
-      "learning_rate": 0.00016477919528949949,
-      "loss": 0.1071,
+      "learning_rate": 0.0002196472303206997,
+      "loss": 0.056,
       "step": 9690
     },
     {
       "epoch": 13.94,
-      "learning_rate": 0.00016463199214916585,
-      "loss": 0.1123,
+      "learning_rate": 0.00021955976676384838,
+      "loss": 0.0533,
       "step": 9700
     },
     {
       "epoch": 13.94,
-      "eval_loss": 0.16937404870986938,
-      "eval_runtime": 40.1681,
-      "eval_samples_per_second": 11.377,
-      "eval_steps_per_second": 1.444,
-      "eval_wer": 0.193344375847011,
+      "eval_loss": 0.15973329544067383,
+      "eval_runtime": 39.8632,
+      "eval_samples_per_second": 11.464,
+      "eval_steps_per_second": 1.455,
+      "eval_wer": 0.1630778497214275,
       "step": 9700
     },
     {
       "epoch": 13.95,
-      "learning_rate": 0.00016448478900883218,
-      "loss": 0.13,
+      "learning_rate": 0.00021947230320699707,
+      "loss": 0.0615,
       "step": 9710
     },
     {
       "epoch": 13.97,
-      "learning_rate": 0.0001643375858684985,
-      "loss": 0.1109,
+      "learning_rate": 0.00021938483965014573,
+      "loss": 0.0534,
       "step": 9720
     },
     {
       "epoch": 13.98,
-      "learning_rate": 0.00016419038272816485,
-      "loss": 0.1241,
+      "learning_rate": 0.00021929737609329442,
+      "loss": 0.0592,
       "step": 9730
     },
     {
       "epoch": 13.99,
-      "learning_rate": 0.0001640431795878312,
-      "loss": 0.115,
+      "learning_rate": 0.00021920991253644313,
+      "loss": 0.0595,
       "step": 9740
     },
     {
       "epoch": 14.01,
-      "learning_rate": 0.00016389597644749754,
-      "loss": 0.11,
+      "learning_rate": 0.00021912244897959182,
+      "loss": 0.0589,
       "step": 9750
     },
     {
       "epoch": 14.02,
-      "learning_rate": 0.00016374877330716387,
-      "loss": 0.1115,
+      "learning_rate": 0.0002190349854227405,
+      "loss": 0.0605,
       "step": 9760
     },
     {
       "epoch": 14.04,
-      "learning_rate": 0.0001636015701668302,
-      "loss": 0.0945,
+      "learning_rate": 0.0002189475218658892,
+      "loss": 0.0444,
       "step": 9770
     },
     {
       "epoch": 14.05,
-      "learning_rate": 0.00016345436702649657,
-      "loss": 0.1122,
+      "learning_rate": 0.00021886005830903788,
+      "loss": 0.0612,
       "step": 9780
     },
     {
       "epoch": 14.07,
-      "learning_rate": 0.0001633071638861629,
-      "loss": 0.0999,
+      "learning_rate": 0.00021877259475218657,
+      "loss": 0.045,
       "step": 9790
     },
     {
       "epoch": 14.08,
-      "learning_rate": 0.00016315996074582923,
-      "loss": 0.1401,
+      "learning_rate": 0.00021868513119533526,
+      "loss": 0.0733,
       "step": 9800
     },
     {
       "epoch": 14.08,
-      "eval_loss": 0.17904061079025269,
-      "eval_runtime": 39.9612,
-      "eval_samples_per_second": 11.436,
-      "eval_steps_per_second": 1.451,
-      "eval_wer": 0.19093510013552176,
+      "eval_loss": 0.16077786684036255,
+      "eval_runtime": 39.8765,
+      "eval_samples_per_second": 11.46,
+      "eval_steps_per_second": 1.454,
+      "eval_wer": 0.1715103147116398,
       "step": 9800
     },
     {
       "epoch": 14.09,
-      "learning_rate": 0.00016301275760549557,
-      "loss": 0.1179,
+      "learning_rate": 0.00021859766763848392,
+      "loss": 0.0694,
       "step": 9810
     },
     {
       "epoch": 14.11,
-      "learning_rate": 0.00016286555446516193,
-      "loss": 0.1311,
+      "learning_rate": 0.00021851020408163264,
+      "loss": 0.0599,
       "step": 9820
     },
     {
       "epoch": 14.12,
-      "learning_rate": 0.00016271835132482823,
-      "loss": 0.1111,
+      "learning_rate": 0.00021842274052478132,
+      "loss": 0.0685,
       "step": 9830
     },
     {
       "epoch": 14.14,
-      "learning_rate": 0.0001625711481844946,
-      "loss": 0.1112,
+      "learning_rate": 0.00021833527696793,
+      "loss": 0.0596,
       "step": 9840
     },
     {
       "epoch": 14.15,
-      "learning_rate": 0.00016242394504416093,
-      "loss": 0.1189,
+      "learning_rate": 0.0002182478134110787,
+      "loss": 0.063,
       "step": 9850
     },
     {
       "epoch": 14.17,
-      "learning_rate": 0.0001622767419038273,
-      "loss": 0.1143,
+      "learning_rate": 0.0002181603498542274,
+      "loss": 0.0602,
       "step": 9860
     },
     {
       "epoch": 14.18,
-      "learning_rate": 0.0001621295387634936,
-      "loss": 0.1192,
+      "learning_rate": 0.00021807288629737608,
+      "loss": 0.0547,
       "step": 9870
     },
     {
       "epoch": 14.2,
-      "learning_rate": 0.00016198233562315996,
-      "loss": 0.1224,
+      "learning_rate": 0.00021798542274052476,
+      "loss": 0.0622,
       "step": 9880
     },
     {
       "epoch": 14.21,
-      "learning_rate": 0.0001618351324828263,
-      "loss": 0.1,
+      "learning_rate": 0.00021789795918367348,
+      "loss": 0.0522,
       "step": 9890
     },
     {
       "epoch": 14.22,
-      "learning_rate": 0.00016168792934249265,
-      "loss": 0.1255,
+      "learning_rate": 0.0002178104956268221,
+      "loss": 0.0635,
       "step": 9900
     },
     {
       "epoch": 14.22,
-      "eval_loss": 0.17239724099636078,
-      "eval_runtime": 40.3402,
-      "eval_samples_per_second": 11.329,
-      "eval_steps_per_second": 1.438,
-      "eval_wer": 0.18822466496009638,
+      "eval_loss": 0.16328591108322144,
+      "eval_runtime": 40.0655,
+      "eval_samples_per_second": 11.406,
+      "eval_steps_per_second": 1.448,
+      "eval_wer": 0.1615720524017467,
       "step": 9900
     },
     {
       "epoch": 14.24,
-      "learning_rate": 0.00016154072620215896,
-      "loss": 0.0956,
+      "learning_rate": 0.00021772303206997083,
+      "loss": 0.0453,
       "step": 9910
     },
     {
       "epoch": 14.25,
-      "learning_rate": 0.00016139352306182532,
-      "loss": 0.1031,
+      "learning_rate": 0.00021763556851311951,
+      "loss": 0.0531,
       "step": 9920
     },
     {
       "epoch": 14.27,
-      "learning_rate": 0.00016124631992149165,
-      "loss": 0.1059,
+      "learning_rate": 0.0002175481049562682,
+      "loss": 0.0613,
       "step": 9930
     },
     {
       "epoch": 14.28,
-      "learning_rate": 0.000161099116781158,
-      "loss": 0.1004,
+      "learning_rate": 0.0002174606413994169,
+      "loss": 0.0579,
       "step": 9940
     },
     {
       "epoch": 14.3,
-      "learning_rate": 0.00016095191364082432,
-      "loss": 0.1064,
+      "learning_rate": 0.00021737317784256558,
+      "loss": 0.0544,
       "step": 9950
     },
     {
       "epoch": 14.31,
-      "learning_rate": 0.00016080471050049065,
-      "loss": 0.1167,
+      "learning_rate": 0.00021728571428571427,
+      "loss": 0.0568,
       "step": 9960
     },
     {
       "epoch": 14.32,
-      "learning_rate": 0.000160657507360157,
-      "loss": 0.1056,
+      "learning_rate": 0.00021719825072886295,
+      "loss": 0.0503,
       "step": 9970
     },
     {
       "epoch": 14.34,
-      "learning_rate": 0.00016051030421982332,
-      "loss": 0.128,
+      "learning_rate": 0.00021711078717201167,
+      "loss": 0.0666,
       "step": 9980
     },
     {
       "epoch": 14.35,
-      "learning_rate": 0.00016036310107948968,
-      "loss": 0.1267,
+      "learning_rate": 0.0002170233236151603,
+      "loss": 0.0674,
       "step": 9990
     },
     {
       "epoch": 14.37,
-      "learning_rate": 0.000160215897939156,
-      "loss": 0.1126,
+      "learning_rate": 0.00021693586005830902,
+      "loss": 0.0515,
       "step": 10000
     },
     {
       "epoch": 14.37,
-      "eval_loss": 0.1660313904285431,
-      "eval_runtime": 40.0416,
-      "eval_samples_per_second": 11.413,
-      "eval_steps_per_second": 1.448,
-      "eval_wer": 0.18882698388796867,
+      "eval_loss": 0.16031676530838013,
+      "eval_runtime": 40.2452,
+      "eval_samples_per_second": 11.355,
+      "eval_steps_per_second": 1.441,
+      "eval_wer": 0.16428248757717212,
       "step": 10000
     },
     {
       "epoch": 14.38,
-      "learning_rate": 0.00016006869479882237,
-      "loss": 0.0972,
+      "learning_rate": 0.0002168483965014577,
+      "loss": 0.0453,
       "step": 10010
     },
     {
       "epoch": 14.4,
-      "learning_rate": 0.00015992149165848868,
-      "loss": 0.0785,
+      "learning_rate": 0.0002167609329446064,
+      "loss": 0.0383,
       "step": 10020
     },
     {
       "epoch": 14.41,
-      "learning_rate": 0.00015977428851815504,
-      "loss": 0.1162,
+      "learning_rate": 0.00021667346938775508,
+      "loss": 0.0582,
       "step": 10030
     },
     {
       "epoch": 14.43,
-      "learning_rate": 0.00015962708537782137,
-      "loss": 0.1037,
+      "learning_rate": 0.00021658600583090377,
+      "loss": 0.0521,
       "step": 10040
     },
     {
       "epoch": 14.44,
-      "learning_rate": 0.00015947988223748773,
-      "loss": 0.1261,
+      "learning_rate": 0.00021649854227405246,
+      "loss": 0.0611,
       "step": 10050
     },
     {
       "epoch": 14.45,
-      "learning_rate": 0.00015933267909715404,
-      "loss": 0.1,
+      "learning_rate": 0.00021641107871720115,
+      "loss": 0.0555,
       "step": 10060
     },
     {
       "epoch": 14.47,
-      "learning_rate": 0.0001591854759568204,
-      "loss": 0.1175,
+      "learning_rate": 0.00021632361516034986,
+      "loss": 0.0571,
       "step": 10070
     },
     {
       "epoch": 14.48,
-      "learning_rate": 0.00015903827281648673,
-      "loss": 0.1218,
+      "learning_rate": 0.00021623615160349852,
+      "loss": 0.0673,
       "step": 10080
     },
     {
       "epoch": 14.5,
-      "learning_rate": 0.0001588910696761531,
-      "loss": 0.0957,
+      "learning_rate": 0.0002161486880466472,
+      "loss": 0.0461,
       "step": 10090
     },
     {
       "epoch": 14.51,
-      "learning_rate": 0.0001587438665358194,
-      "loss": 0.1241,
+      "learning_rate": 0.0002160612244897959,
+      "loss": 0.0653,
       "step": 10100
     },
     {
       "epoch": 14.51,
-      "eval_loss": 0.16984562575817108,
-      "eval_runtime": 40.0158,
-      "eval_samples_per_second": 11.42,
-      "eval_steps_per_second": 1.449,
-      "eval_wer": 0.19048336093961754,
+      "eval_loss": 0.16045616567134857,
+      "eval_runtime": 39.9395,
+      "eval_samples_per_second": 11.442,
+      "eval_steps_per_second": 1.452,
+      "eval_wer": 0.15840987803041712,
       "step": 10100
     },
     {
       "epoch": 14.53,
-      "learning_rate": 0.00015859666339548576,
-      "loss": 0.0949,
+      "learning_rate": 0.00021597376093294458,
+      "loss": 0.0555,
       "step": 10110
     },
     {
       "epoch": 14.54,
-      "learning_rate": 0.0001584494602551521,
-      "loss": 0.1353,
+      "learning_rate": 0.00021588629737609327,
+      "loss": 0.0591,
       "step": 10120
     },
     {
       "epoch": 14.55,
-      "learning_rate": 0.00015830225711481843,
-      "loss": 0.1164,
+      "learning_rate": 0.00021579883381924196,
+      "loss": 0.0639,
       "step": 10130
     },
     {
       "epoch": 14.57,
-      "learning_rate": 0.00015815505397448476,
-      "loss": 0.0954,
+      "learning_rate": 0.00021571137026239065,
+      "loss": 0.0507,
       "step": 10140
     },
     {
       "epoch": 14.58,
-      "learning_rate": 0.00015800785083415112,
-      "loss": 0.1477,
+      "learning_rate": 0.00021562390670553936,
+      "loss": 0.0744,
       "step": 10150
     },
     {
       "epoch": 14.6,
-      "learning_rate": 0.00015786064769381745,
-      "loss": 0.1094,
+      "learning_rate": 0.00021553644314868805,
+      "loss": 0.0596,
       "step": 10160
     },
     {
       "epoch": 14.61,
-      "learning_rate": 0.0001577134445534838,
-      "loss": 0.0973,
+      "learning_rate": 0.0002154489795918367,
+      "loss": 0.0447,
       "step": 10170
     },
     {
       "epoch": 14.63,
-      "learning_rate": 0.00015756624141315012,
-      "loss": 0.1124,
+      "learning_rate": 0.0002153615160349854,
+      "loss": 0.0632,
       "step": 10180
     },
     {
       "epoch": 14.64,
-      "learning_rate": 0.00015741903827281648,
-      "loss": 0.1131,
+      "learning_rate": 0.0002152740524781341,
+      "loss": 0.0488,
       "step": 10190
     },
     {
       "epoch": 14.66,
-      "learning_rate": 0.00015727183513248281,
-      "loss": 0.1147,
+      "learning_rate": 0.00021518658892128278,
+      "loss": 0.0565,
       "step": 10200
     },
     {
       "epoch": 14.66,
-      "eval_loss": 0.16483080387115479,
-      "eval_runtime": 39.9302,
-      "eval_samples_per_second": 11.445,
-      "eval_steps_per_second": 1.453,
-      "eval_wer": 0.1802439391657883,
+      "eval_loss": 0.1643233746290207,
+      "eval_runtime": 40.486,
+      "eval_samples_per_second": 11.288,
+      "eval_steps_per_second": 1.433,
+      "eval_wer": 0.15810871856648095,
       "step": 10200
     },
     {
       "epoch": 14.67,
-      "learning_rate": 0.00015712463199214915,
-      "loss": 0.1111,
+      "learning_rate": 0.00021509912536443146,
+      "loss": 0.0521,
       "step": 10210
     },
     {
       "epoch": 14.68,
-      "learning_rate": 0.00015697742885181548,
-      "loss": 0.1048,
+      "learning_rate": 0.00021501166180758015,
+      "loss": 0.0523,
       "step": 10220
     },
     {
       "epoch": 14.7,
-      "learning_rate": 0.00015683022571148184,
-      "loss": 0.1285,
+      "learning_rate": 0.00021492419825072884,
+      "loss": 0.0711,
       "step": 10230
     },
     {
       "epoch": 14.71,
-      "learning_rate": 0.00015668302257114817,
-      "loss": 0.092,
+      "learning_rate": 0.00021483673469387755,
+      "loss": 0.051,
       "step": 10240
     },
     {
       "epoch": 14.73,
-      "learning_rate": 0.0001565358194308145,
-      "loss": 0.1086,
+      "learning_rate": 0.00021474927113702624,
+      "loss": 0.063,
       "step": 10250
     },
     {
       "epoch": 14.74,
-      "learning_rate": 0.00015638861629048084,
-      "loss": 0.1002,
+      "learning_rate": 0.0002146618075801749,
+      "loss": 0.0521,
       "step": 10260
     },
     {
       "epoch": 14.76,
-      "learning_rate": 0.0001562414131501472,
-      "loss": 0.1148,
+      "learning_rate": 0.0002145743440233236,
+      "loss": 0.0573,
       "step": 10270
     },
     {
       "epoch": 14.77,
-      "learning_rate": 0.0001560942100098135,
-      "loss": 0.1267,
+      "learning_rate": 0.00021448688046647228,
+      "loss": 0.0668,
       "step": 10280
     },
     {
       "epoch": 14.78,
-      "learning_rate": 0.00015594700686947987,
-      "loss": 0.1025,
+      "learning_rate": 0.00021439941690962097,
+      "loss": 0.0523,
       "step": 10290
     },
     {
       "epoch": 14.8,
-      "learning_rate": 0.0001557998037291462,
-      "loss": 0.115,
+      "learning_rate": 0.00021431195335276965,
+      "loss": 0.0576,
       "step": 10300
     },
     {
       "epoch": 14.8,
-      "eval_loss": 0.1692885160446167,
-      "eval_runtime": 40.0262,
-      "eval_samples_per_second": 11.418,
-      "eval_steps_per_second": 1.449,
-      "eval_wer": 0.18581538924860713,
+      "eval_loss": 0.15989667177200317,
+      "eval_runtime": 39.8824,
+      "eval_samples_per_second": 11.459,
+      "eval_steps_per_second": 1.454,
+      "eval_wer": 0.16081915374190633,
       "step": 10300
     },
     {
       "epoch": 14.81,
-      "learning_rate": 0.00015565260058881256,
-      "loss": 0.0927,
+      "learning_rate": 0.00021422448979591834,
+      "loss": 0.0478,
       "step": 10310
     },
     {
       "epoch": 14.83,
-      "learning_rate": 0.00015550539744847887,
-      "loss": 0.1171,
+      "learning_rate": 0.00021413702623906706,
+      "loss": 0.0601,
       "step": 10320
     },
     {
       "epoch": 14.84,
-      "learning_rate": 0.00015535819430814523,
-      "loss": 0.1267,
+      "learning_rate": 0.00021404956268221575,
+      "loss": 0.0628,
       "step": 10330
     },
     {
       "epoch": 14.86,
-      "learning_rate": 0.00015521099116781156,
-      "loss": 0.1005,
+      "learning_rate": 0.00021396209912536443,
+      "loss": 0.0493,
       "step": 10340
     },
     {
       "epoch": 14.87,
-      "learning_rate": 0.00015506378802747792,
-      "loss": 0.117,
+      "learning_rate": 0.0002138746355685131,
+      "loss": 0.0654,
       "step": 10350
     },
     {
       "epoch": 14.89,
-      "learning_rate": 0.00015491658488714423,
-      "loss": 0.0975,
+      "learning_rate": 0.00021378717201166178,
+      "loss": 0.05,
       "step": 10360
     },
     {
       "epoch": 14.9,
-      "learning_rate": 0.0001547693817468106,
-      "loss": 0.1131,
+      "learning_rate": 0.00021369970845481047,
+      "loss": 0.0567,
       "step": 10370
     },
     {
       "epoch": 14.91,
-      "learning_rate": 0.00015462217860647692,
-      "loss": 0.1209,
+      "learning_rate": 0.00021361224489795916,
+      "loss": 0.0642,
       "step": 10380
     },
     {
       "epoch": 14.93,
-      "learning_rate": 0.00015447497546614328,
-      "loss": 0.105,
+      "learning_rate": 0.00021352478134110785,
+      "loss": 0.0493,
       "step": 10390
     },
     {
       "epoch": 14.94,
-      "learning_rate": 0.0001543277723258096,
-      "loss": 0.1078,
+      "learning_rate": 0.00021343731778425653,
+      "loss": 0.0593,
       "step": 10400
     },
     {
       "epoch": 14.94,
-      "eval_loss": 0.16425766050815582,
-      "eval_runtime": 39.9939,
-      "eval_samples_per_second": 11.427,
-      "eval_steps_per_second": 1.45,
-      "eval_wer": 0.18385785273302213,
+      "eval_loss": 0.15803176164627075,
+      "eval_runtime": 40.0466,
+      "eval_samples_per_second": 11.412,
+      "eval_steps_per_second": 1.448,
+      "eval_wer": 0.1552477036590875,
       "step": 10400
     },
     {
       "epoch": 14.96,
-      "learning_rate": 0.00015418056918547595,
-      "loss": 0.1102,
+      "learning_rate": 0.00021334985422740525,
+      "loss": 0.0594,
       "step": 10410
     },
     {
       "epoch": 14.97,
-      "learning_rate": 0.00015403336604514228,
-      "loss": 0.1,
+      "learning_rate": 0.00021326239067055394,
+      "loss": 0.0525,
       "step": 10420
     },
     {
       "epoch": 14.99,
-      "learning_rate": 0.00015388616290480862,
-      "loss": 0.0944,
+      "learning_rate": 0.00021317492711370262,
+      "loss": 0.0546,
       "step": 10430
     },
     {
       "epoch": 15.0,
-      "learning_rate": 0.00015373895976447495,
-      "loss": 0.1232,
+      "learning_rate": 0.00021308746355685129,
+      "loss": 0.065,
       "step": 10440
     },
     {
       "epoch": 15.01,
-      "learning_rate": 0.0001535917566241413,
-      "loss": 0.1017,
+      "learning_rate": 0.00021299999999999997,
+      "loss": 0.0553,
       "step": 10450
     },
     {
       "epoch": 15.03,
-      "learning_rate": 0.00015344455348380764,
-      "loss": 0.0991,
+      "learning_rate": 0.00021291253644314866,
+      "loss": 0.0502,
       "step": 10460
     },
     {
       "epoch": 15.04,
-      "learning_rate": 0.00015329735034347398,
-      "loss": 0.0986,
+      "learning_rate": 0.00021282507288629735,
+      "loss": 0.0518,
       "step": 10470
     },
     {
       "epoch": 15.06,
-      "learning_rate": 0.0001531501472031403,
-      "loss": 0.1012,
+      "learning_rate": 0.00021273760932944604,
+      "loss": 0.0533,
       "step": 10480
     },
     {
       "epoch": 15.07,
-      "learning_rate": 0.00015300294406280667,
-      "loss": 0.0992,
+      "learning_rate": 0.00021265014577259472,
+      "loss": 0.0534,
       "step": 10490
     },
     {
       "epoch": 15.09,
-      "learning_rate": 0.000152855740922473,
-      "loss": 0.107,
+      "learning_rate": 0.00021256268221574344,
+      "loss": 0.0624,
       "step": 10500
     },
     {
       "epoch": 15.09,
-      "eval_loss": 0.169921875,
-      "eval_runtime": 39.8342,
-      "eval_samples_per_second": 11.473,
-      "eval_steps_per_second": 1.456,
-      "eval_wer": 0.17964162023791597,
+      "eval_loss": 0.16080322861671448,
+      "eval_runtime": 39.9501,
+      "eval_samples_per_second": 11.439,
+      "eval_steps_per_second": 1.452,
+      "eval_wer": 0.15946393615419366,
       "step": 10500
     },
     {
       "epoch": 15.1,
-      "learning_rate": 0.00015270853778213934,
-      "loss": 0.0901,
+      "learning_rate": 0.00021247521865889213,
+      "loss": 0.0495,
       "step": 10510
     },
     {
       "epoch": 15.11,
-      "learning_rate": 0.00015256133464180567,
-      "loss": 0.1056,
+      "learning_rate": 0.00021238775510204082,
+      "loss": 0.062,
       "step": 10520
     },
     {
       "epoch": 15.13,
-      "learning_rate": 0.00015241413150147203,
-      "loss": 0.1044,
+      "learning_rate": 0.00021230029154518948,
+      "loss": 0.0617,
       "step": 10530
     },
     {
       "epoch": 15.14,
-      "learning_rate": 0.00015226692836113837,
-      "loss": 0.0921,
+      "learning_rate": 0.00021221282798833816,
+      "loss": 0.0541,
       "step": 10540
     },
     {
       "epoch": 15.16,
-      "learning_rate": 0.0001521197252208047,
-      "loss": 0.1146,
+      "learning_rate": 0.00021212536443148685,
+      "loss": 0.0574,
       "step": 10550
     },
     {
       "epoch": 15.17,
-      "learning_rate": 0.00015197252208047103,
-      "loss": 0.0826,
+      "learning_rate": 0.00021203790087463554,
+      "loss": 0.0396,
       "step": 10560
     },
     {
       "epoch": 15.19,
-      "learning_rate": 0.0001518253189401374,
-      "loss": 0.1078,
+      "learning_rate": 0.00021195043731778423,
+      "loss": 0.0522,
       "step": 10570
     },
     {
       "epoch": 15.2,
-      "learning_rate": 0.0001516781157998037,
-      "loss": 0.1036,
+      "learning_rate": 0.00021186297376093294,
+      "loss": 0.0602,
       "step": 10580
     },
     {
       "epoch": 15.22,
-      "learning_rate": 0.00015153091265947006,
-      "loss": 0.0755,
+      "learning_rate": 0.00021177551020408163,
+      "loss": 0.0383,
       "step": 10590
     },
     {
       "epoch": 15.23,
-      "learning_rate": 0.0001513837095191364,
-      "loss": 0.0966,
+      "learning_rate": 0.00021168804664723032,
+      "loss": 0.0544,
       "step": 10600
     },
     {
       "epoch": 15.23,
-      "eval_loss": 0.1691877841949463,
-      "eval_runtime": 40.0386,
-      "eval_samples_per_second": 11.414,
-      "eval_steps_per_second": 1.449,
-      "eval_wer": 0.1802439391657883,
+      "eval_loss": 0.1565014123916626,
+      "eval_runtime": 39.9025,
+      "eval_samples_per_second": 11.453,
+      "eval_steps_per_second": 1.454,
+      "eval_wer": 0.16066857400993825,
       "step": 10600
     },
     {
       "epoch": 15.24,
-      "learning_rate": 0.00015123650637880275,
-      "loss": 0.087,
+      "learning_rate": 0.000211600583090379,
+      "loss": 0.0482,
       "step": 10610
     },
     {
       "epoch": 15.26,
-      "learning_rate": 0.00015108930323846906,
-      "loss": 0.1237,
+      "learning_rate": 0.00021151311953352767,
+      "loss": 0.0644,
       "step": 10620
     },
     {
       "epoch": 15.27,
-      "learning_rate": 0.00015094210009813542,
-      "loss": 0.087,
+      "learning_rate": 0.00021142565597667636,
+      "loss": 0.0465,
       "step": 10630
     },
     {
       "epoch": 15.29,
-      "learning_rate": 0.00015079489695780175,
-      "loss": 0.0968,
+      "learning_rate": 0.00021133819241982504,
+      "loss": 0.0482,
       "step": 10640
     },
     {
       "epoch": 15.3,
-      "learning_rate": 0.00015064769381746811,
-      "loss": 0.1188,
+      "learning_rate": 0.00021125072886297373,
+      "loss": 0.069,
       "step": 10650
     },
     {
       "epoch": 15.32,
-      "learning_rate": 0.00015050049067713442,
-      "loss": 0.0945,
+      "learning_rate": 0.00021116326530612242,
+      "loss": 0.0537,
       "step": 10660
     },
     {
       "epoch": 15.33,
-      "learning_rate": 0.00015035328753680078,
-      "loss": 0.1007,
+      "learning_rate": 0.00021107580174927113,
+      "loss": 0.053,
       "step": 10670
     },
     {
       "epoch": 15.34,
-      "learning_rate": 0.00015020608439646711,
-      "loss": 0.0988,
+      "learning_rate": 0.00021098833819241982,
+      "loss": 0.0516,
       "step": 10680
     },
     {
       "epoch": 15.36,
-      "learning_rate": 0.00015005888125613348,
-      "loss": 0.0876,
+      "learning_rate": 0.0002109008746355685,
+      "loss": 0.0402,
       "step": 10690
     },
     {
       "epoch": 15.37,
-      "learning_rate": 0.00014991167811579978,
-      "loss": 0.0931,
+      "learning_rate": 0.0002108134110787172,
+      "loss": 0.052,
       "step": 10700
     },
     {
       "epoch": 15.37,
-      "eval_loss": 0.16350314021110535,
-      "eval_runtime": 39.8693,
-      "eval_samples_per_second": 11.462,
-      "eval_steps_per_second": 1.455,
-      "eval_wer": 0.17753350399036288,
+      "eval_loss": 0.16009217500686646,
+      "eval_runtime": 40.0212,
+      "eval_samples_per_second": 11.419,
+      "eval_steps_per_second": 1.449,
+      "eval_wer": 0.15931335642222558,
       "step": 10700
     },
     {
       "epoch": 15.39,
-      "learning_rate": 0.00014976447497546614,
-      "loss": 0.0825,
+      "learning_rate": 0.00021072594752186586,
+      "loss": 0.0464,
       "step": 10710
     },
     {
       "epoch": 15.4,
-      "learning_rate": 0.00014961727183513248,
-      "loss": 0.1109,
+      "learning_rate": 0.00021063848396501455,
+      "loss": 0.0603,
       "step": 10720
     },
     {
       "epoch": 15.42,
-      "learning_rate": 0.0001494700686947988,
-      "loss": 0.095,
+      "learning_rate": 0.00021055102040816323,
+      "loss": 0.0535,
       "step": 10730
     },
     {
       "epoch": 15.43,
-      "learning_rate": 0.00014932286555446514,
-      "loss": 0.1073,
+      "learning_rate": 0.00021046355685131192,
+      "loss": 0.0583,
       "step": 10740
     },
     {
       "epoch": 15.45,
-      "learning_rate": 0.0001491756624141315,
-      "loss": 0.111,
+      "learning_rate": 0.00021037609329446064,
+      "loss": 0.0628,
       "step": 10750
     },
     {
       "epoch": 15.46,
-      "learning_rate": 0.00014902845927379784,
-      "loss": 0.0962,
+      "learning_rate": 0.00021028862973760933,
+      "loss": 0.0553,
       "step": 10760
     },
     {
       "epoch": 15.47,
-      "learning_rate": 0.00014888125613346417,
-      "loss": 0.096,
+      "learning_rate": 0.000210201166180758,
+      "loss": 0.0526,
       "step": 10770
     },
     {
       "epoch": 15.49,
-      "learning_rate": 0.0001487340529931305,
-      "loss": 0.0717,
+      "learning_rate": 0.0002101137026239067,
+      "loss": 0.0417,
       "step": 10780
     },
     {
       "epoch": 15.5,
-      "learning_rate": 0.00014858684985279686,
-      "loss": 0.0893,
+      "learning_rate": 0.0002100262390670554,
+      "loss": 0.0498,
       "step": 10790
     },
     {
       "epoch": 15.52,
-      "learning_rate": 0.0001484396467124632,
-      "loss": 0.1068,
+      "learning_rate": 0.00020993877551020405,
+      "loss": 0.0577,
       "step": 10800
     },
     {
       "epoch": 15.52,
-      "eval_loss": 0.16544069349765778,
-      "eval_runtime": 39.8983,
-      "eval_samples_per_second": 11.454,
-      "eval_steps_per_second": 1.454,
-      "eval_wer": 0.17407017015509713,
-      "step": 10800
+      "eval_loss": 0.1494068205356598,
+      "eval_runtime": 40.6793,
+      "eval_samples_per_second": 11.234,
+      "eval_steps_per_second": 1.426,
+      "eval_wer": 0.1552477036590875,
+      "step": 10800
     },
     {
       "epoch": 15.53,
-      "learning_rate": 0.00014829244357212953,
-      "loss": 0.0968,
+      "learning_rate": 0.00020985131195335274,
+      "loss": 0.047,
       "step": 10810
     },
     {
       "epoch": 15.55,
-      "learning_rate": 0.00014814524043179586,
-      "loss": 0.0907,
+      "learning_rate": 0.00020976384839650143,
+      "loss": 0.0467,
       "step": 10820
     },
     {
       "epoch": 15.56,
-      "learning_rate": 0.00014799803729146222,
-      "loss": 0.0836,
+      "learning_rate": 0.0002096763848396501,
+      "loss": 0.0468,
       "step": 10830
     },
     {
       "epoch": 15.57,
-      "learning_rate": 0.00014785083415112856,
-      "loss": 0.0866,
+      "learning_rate": 0.00020958892128279883,
+      "loss": 0.0484,
       "step": 10840
     },
     {
       "epoch": 15.59,
-      "learning_rate": 0.00014770363101079486,
-      "loss": 0.1004,
+      "learning_rate": 0.00020950145772594752,
+      "loss": 0.0553,
       "step": 10850
     },
     {
       "epoch": 15.6,
-      "learning_rate": 0.00014755642787046122,
-      "loss": 0.0972,
+      "learning_rate": 0.0002094139941690962,
+      "loss": 0.0478,
       "step": 10860
     },
     {
       "epoch": 15.62,
-      "learning_rate": 0.00014740922473012756,
-      "loss": 0.1118,
+      "learning_rate": 0.0002093265306122449,
+      "loss": 0.061,
       "step": 10870
     },
     {
       "epoch": 15.63,
-      "learning_rate": 0.00014726202158979392,
-      "loss": 0.1021,
+      "learning_rate": 0.00020923906705539358,
+      "loss": 0.0492,
       "step": 10880
     },
     {
       "epoch": 15.65,
-      "learning_rate": 0.00014711481844946022,
-      "loss": 0.0791,
+      "learning_rate": 0.00020915160349854224,
+      "loss": 0.0462,
       "step": 10890
     },
     {
       "epoch": 15.66,
-      "learning_rate": 0.00014696761530912658,
-      "loss": 0.1224,
+      "learning_rate": 0.00020906413994169093,
+      "loss": 0.0762,
       "step": 10900
     },
     {
       "epoch": 15.66,
-      "eval_loss": 0.16307871043682098,
-      "eval_runtime": 40.0248,
-      "eval_samples_per_second": 11.418,
-      "eval_steps_per_second": 1.449,
-      "eval_wer": 0.1767806053305225,
+      "eval_loss": 0.1616104245185852,
+      "eval_runtime": 40.2584,
+      "eval_samples_per_second": 11.352,
+      "eval_steps_per_second": 1.441,
+      "eval_wer": 0.15991567535009787,
       "step": 10900
     },
     {
       "epoch": 15.68,
-      "learning_rate": 0.00014682041216879292,
-      "loss": 0.0825,
+      "learning_rate": 0.00020897667638483962,
+      "loss": 0.0435,
       "step": 10910
     },
     {
       "epoch": 15.69,
-      "learning_rate": 0.00014667320902845928,
-      "loss": 0.0958,
+      "learning_rate": 0.0002088892128279883,
+      "loss": 0.0551,
       "step": 10920
     },
     {
       "epoch": 15.7,
-      "learning_rate": 0.00014652600588812558,
-      "loss": 0.112,
+      "learning_rate": 0.00020880174927113702,
+      "loss": 0.0621,
       "step": 10930
     },
     {
       "epoch": 15.72,
-      "learning_rate": 0.00014637880274779195,
-      "loss": 0.0934,
+      "learning_rate": 0.0002087142857142857,
+      "loss": 0.0501,
       "step": 10940
     },
     {
       "epoch": 15.73,
-      "learning_rate": 0.00014623159960745828,
-      "loss": 0.0917,
+      "learning_rate": 0.0002086268221574344,
+      "loss": 0.0519,
       "step": 10950
     },
     {
       "epoch": 15.75,
-      "learning_rate": 0.00014608439646712464,
-      "loss": 0.0813,
+      "learning_rate": 0.00020853935860058308,
+      "loss": 0.0415,
       "step": 10960
     },
     {
       "epoch": 15.76,
-      "learning_rate": 0.00014593719332679095,
-      "loss": 0.1391,
+      "learning_rate": 0.00020845189504373177,
+      "loss": 0.0717,
       "step": 10970
     },
     {
       "epoch": 15.78,
-      "learning_rate": 0.0001457899901864573,
-      "loss": 0.1232,
+      "learning_rate": 0.00020836443148688043,
+      "loss": 0.0691,
       "step": 10980
     },
     {
       "epoch": 15.79,
-      "learning_rate": 0.00014564278704612364,
-      "loss": 0.1132,
+      "learning_rate": 0.00020827696793002912,
+      "loss": 0.0477,
       "step": 10990
     },
     {
       "epoch": 15.8,
-      "learning_rate": 0.00014549558390578997,
-      "loss": 0.0999,
+      "learning_rate": 0.0002081895043731778,
+      "loss": 0.0509,
       "step": 11000
     },
     {
       "epoch": 15.8,
-      "eval_loss": 0.15900734066963196,
-      "eval_runtime": 40.0131,
-      "eval_samples_per_second": 11.421,
-      "eval_steps_per_second": 1.45,
-      "eval_wer": 0.17663002559855442,
+      "eval_loss": 0.1517663598060608,
+      "eval_runtime": 39.8657,
+      "eval_samples_per_second": 11.463,
+      "eval_steps_per_second": 1.455,
+      "eval_wer": 0.15359132660743863,
       "step": 11000
     },
     {
       "epoch": 15.82,
-      "learning_rate": 0.0001453483807654563,
-      "loss": 0.079,
+      "learning_rate": 0.00020810204081632652,
+      "loss": 0.0382,
       "step": 11010
     },
     {
       "epoch": 15.83,
-      "learning_rate": 0.00014520117762512264,
-      "loss": 0.102,
+      "learning_rate": 0.0002080145772594752,
+      "loss": 0.0558,
       "step": 11020
     },
     {
       "epoch": 15.85,
-      "learning_rate": 0.000145053974484789,
-      "loss": 0.088,
+      "learning_rate": 0.0002079271137026239,
+      "loss": 0.0464,
       "step": 11030
     },
     {
       "epoch": 15.86,
-      "learning_rate": 0.00014490677134445533,
-      "loss": 0.0976,
+      "learning_rate": 0.00020783965014577259,
+      "loss": 0.0541,
       "step": 11040
     },
     {
       "epoch": 15.88,
-      "learning_rate": 0.00014475956820412167,
-      "loss": 0.1118,
+      "learning_rate": 0.00020775218658892127,
+      "loss": 0.067,
       "step": 11050
     },
     {
       "epoch": 15.89,
-      "learning_rate": 0.000144612365063788,
-      "loss": 0.0847,
+      "learning_rate": 0.00020766472303206996,
+      "loss": 0.0431,
       "step": 11060
     },
     {
       "epoch": 15.91,
-      "learning_rate": 0.00014446516192345436,
-      "loss": 0.0962,
+      "learning_rate": 0.00020758600583090377,
+      "loss": 0.0541,
       "step": 11070
     },
     {
       "epoch": 15.92,
-      "learning_rate": 0.0001443179587831207,
-      "loss": 0.1053,
+      "learning_rate": 0.00020749854227405246,
+      "loss": 0.0622,
       "step": 11080
     },
     {
       "epoch": 15.93,
-      "learning_rate": 0.00014417075564278703,
-      "loss": 0.1131,
+      "learning_rate": 0.00020741107871720114,
+      "loss": 0.0577,
       "step": 11090
     },
     {
       "epoch": 15.95,
-      "learning_rate": 0.00014402355250245336,
-      "loss": 0.0879,
+      "learning_rate": 0.00020732361516034983,
+      "loss": 0.0542,
       "step": 11100
     },
     {
       "epoch": 15.95,
-      "eval_loss": 0.16153839230537415,
-      "eval_runtime": 40.5046,
-      "eval_samples_per_second": 11.283,
-      "eval_steps_per_second": 1.432,
-      "eval_wer": 0.17196205390754404,
+      "eval_loss": 0.15586353838443756,
+      "eval_runtime": 39.8742,
+      "eval_samples_per_second": 11.461,
+      "eval_steps_per_second": 1.455,
+      "eval_wer": 0.15645234151483212,
       "step": 11100
     },
     {
       "epoch": 15.96,
-      "learning_rate": 0.00014387634936211972,
-      "loss": 0.1013,
+      "learning_rate": 0.00020723615160349855,
+      "loss": 0.0576,
       "step": 11110
     },
     {
       "epoch": 15.98,
-      "learning_rate": 0.00014372914622178605,
-      "loss": 0.1083,
+      "learning_rate": 0.00020714868804664723,
+      "loss": 0.0741,
       "step": 11120
     },
     {
       "epoch": 15.99,
-      "learning_rate": 0.0001435819430814524,
-      "loss": 0.1037,
+      "learning_rate": 0.0002070612244897959,
+      "loss": 0.0568,
       "step": 11130
     },
     {
       "epoch": 16.01,
-      "learning_rate": 0.00014343473994111872,
-      "loss": 0.0954,
+      "learning_rate": 0.00020697376093294458,
+      "loss": 0.0524,
       "step": 11140
     },
     {
       "epoch": 16.02,
-      "learning_rate": 0.00014328753680078506,
-      "loss": 0.0845,
+      "learning_rate": 0.00020688629737609327,
+      "loss": 0.0527,
       "step": 11150
     },
     {
       "epoch": 16.03,
-      "learning_rate": 0.00014314033366045142,
-      "loss": 0.0827,
+      "learning_rate": 0.00020679883381924196,
+      "loss": 0.047,
       "step": 11160
     },
     {
       "epoch": 16.05,
-      "learning_rate": 0.00014299313052011775,
-      "loss": 0.1113,
+      "learning_rate": 0.00020671137026239065,
+      "loss": 0.0621,
       "step": 11170
     },
     {
       "epoch": 16.06,
-      "learning_rate": 0.00014284592737978408,
-      "loss": 0.0934,
+      "learning_rate": 0.00020662390670553933,
+      "loss": 0.0552,
       "step": 11180
     },
     {
       "epoch": 16.08,
-      "learning_rate": 0.00014269872423945042,
-      "loss": 0.0903,
+      "learning_rate": 0.00020653644314868802,
+      "loss": 0.0441,
       "step": 11190
     },
     {
       "epoch": 16.09,
-      "learning_rate": 0.00014255152109911678,
-      "loss": 0.0761,
+      "learning_rate": 0.00020644897959183674,
+      "loss": 0.041,
       "step": 11200
     },
     {
       "epoch": 16.09,
-      "eval_loss": 0.16535413265228271,
-      "eval_runtime": 40.0395,
-      "eval_samples_per_second": 11.414,
-      "eval_steps_per_second": 1.449,
-      "eval_wer": 0.17497364854690559,
+      "eval_loss": 0.16197079420089722,
+      "eval_runtime": 40.3539,
+      "eval_samples_per_second": 11.325,
+      "eval_steps_per_second": 1.437,
+      "eval_wer": 0.15840987803041712,
       "step": 11200
     },
     {
       "epoch": 16.11,
-      "learning_rate": 0.0001424043179587831,
-      "loss": 0.0709,
+      "learning_rate": 0.00020636151603498543,
+      "loss": 0.0362,
       "step": 11210
     },
     {
       "epoch": 16.12,
-      "learning_rate": 0.00014225711481844944,
-      "loss": 0.1006,
+      "learning_rate": 0.00020627405247813409,
+      "loss": 0.0615,
       "step": 11220
     },
     {
       "epoch": 16.14,
-      "learning_rate": 0.00014210991167811578,
-      "loss": 0.1218,
+      "learning_rate": 0.00020618658892128277,
+      "loss": 0.0602,
       "step": 11230
     },
     {
       "epoch": 16.15,
-      "learning_rate": 0.00014196270853778214,
-      "loss": 0.0994,
+      "learning_rate": 0.00020609912536443146,
+      "loss": 0.0569,
       "step": 11240
     },
     {
       "epoch": 16.16,
-      "learning_rate": 0.00014181550539744847,
-      "loss": 0.0986,
+      "learning_rate": 0.00020601166180758015,
+      "loss": 0.0459,
       "step": 11250
     },
     {
       "epoch": 16.18,
-      "learning_rate": 0.0001416683022571148,
-      "loss": 0.0789,
+      "learning_rate": 0.00020592419825072884,
+      "loss": 0.0427,
       "step": 11260
     },
     {
       "epoch": 16.19,
-      "learning_rate": 0.00014152109911678114,
-      "loss": 0.1026,
+      "learning_rate": 0.00020583673469387753,
+      "loss": 0.0593,
       "step": 11270
     },
     {
       "epoch": 16.21,
-      "learning_rate": 0.0001413738959764475,
-      "loss": 0.0982,
+      "learning_rate": 0.00020574927113702624,
+      "loss": 0.056,
       "step": 11280
     },
     {
       "epoch": 16.22,
-      "learning_rate": 0.00014122669283611383,
-      "loss": 0.1112,
+      "learning_rate": 0.00020566180758017493,
+      "loss": 0.0566,
       "step": 11290
     },
     {
       "epoch": 16.24,
-      "learning_rate": 0.00014107948969578016,
-      "loss": 0.0821,
+      "learning_rate": 0.00020557434402332362,
+      "loss": 0.0449,
       "step": 11300
     },
     {
       "epoch": 16.24,
-      "eval_loss": 0.15813738107681274,
-      "eval_runtime": 39.8582,
-      "eval_samples_per_second": 11.466,
-      "eval_steps_per_second": 1.455,
-      "eval_wer": 0.17452190935100134,
+      "eval_loss": 0.15112736821174622,
+      "eval_runtime": 39.9719,
+      "eval_samples_per_second": 11.433,
+      "eval_steps_per_second": 1.451,
+      "eval_wer": 0.15298900767956633,
       "step": 11300
     },
     {
       "epoch": 16.25,
-      "learning_rate": 0.0001409322865554465,
-      "loss": 0.0935,
+      "learning_rate": 0.00020548688046647228,
+      "loss": 0.052,
       "step": 11310
     },
     {
       "epoch": 16.26,
-      "learning_rate": 0.00014078508341511283,
-      "loss": 0.1029,
+      "learning_rate": 0.00020539941690962096,
+      "loss": 0.0581,
       "step": 11320
     },
     {
       "epoch": 16.28,
-      "learning_rate": 0.0001406378802747792,
-      "loss": 0.1002,
+      "learning_rate": 0.00020531195335276965,
+      "loss": 0.0511,
       "step": 11330
     },
     {
       "epoch": 16.29,
-      "learning_rate": 0.00014049067713444553,
-      "loss": 0.095,
+      "learning_rate": 0.00020522448979591834,
+      "loss": 0.0511,
       "step": 11340
     },
     {
       "epoch": 16.31,
-      "learning_rate": 0.00014034347399411186,
-      "loss": 0.1028,
+      "learning_rate": 0.00020513702623906703,
+      "loss": 0.056,
       "step": 11350
     },
     {
       "epoch": 16.32,
-      "learning_rate": 0.0001401962708537782,
-      "loss": 0.079,
+      "learning_rate": 0.00020504956268221572,
+      "loss": 0.0385,
       "step": 11360
     },
     {
       "epoch": 16.34,
-      "learning_rate": 0.00014004906771344455,
-      "loss": 0.0941,
+      "learning_rate": 0.00020496209912536443,
+      "loss": 0.0542,
       "step": 11370
     },
     {
       "epoch": 16.35,
-      "learning_rate": 0.00013990186457311089,
-      "loss": 0.0839,
+      "learning_rate": 0.00020487463556851312,
+      "loss": 0.0441,
       "step": 11380
     },
     {
       "epoch": 16.36,
-      "learning_rate": 0.00013975466143277722,
-      "loss": 0.0976,
+      "learning_rate": 0.0002047871720116618,
+      "loss": 0.056,
       "step": 11390
     },
     {
       "epoch": 16.38,
-      "learning_rate": 0.00013960745829244355,
-      "loss": 0.0984,
+      "learning_rate": 0.00020469970845481047,
+      "loss": 0.0541,
       "step": 11400
     },
     {
       "epoch": 16.38,
-      "eval_loss": 0.16124248504638672,
-      "eval_runtime": 40.0383,
-      "eval_samples_per_second": 11.414,
-      "eval_steps_per_second": 1.449,
-      "eval_wer": 0.16940219846408675,
+      "eval_loss": 0.16502057015895844,
+      "eval_runtime": 39.9246,
+      "eval_samples_per_second": 11.447,
+      "eval_steps_per_second": 1.453,
+      "eval_wer": 0.16036741454600212,
       "step": 11400
     },
     {
       "epoch": 16.39,
-      "learning_rate": 0.0001394602551521099,
-      "loss": 0.0889,
+      "learning_rate": 0.00020461224489795916,
+      "loss": 0.052,
       "step": 11410
     },
     {
       "epoch": 16.41,
-      "learning_rate": 0.00013931305201177625,
-      "loss": 0.0951,
+      "learning_rate": 0.00020452478134110784,
+      "loss": 0.055,
       "step": 11420
     },
     {
       "epoch": 16.42,
-      "learning_rate": 0.00013916584887144258,
-      "loss": 0.075,
+      "learning_rate": 0.00020443731778425653,
+      "loss": 0.0402,
       "step": 11430
     },
     {
       "epoch": 16.44,
-      "learning_rate": 0.0001390186457311089,
-      "loss": 0.1025,
+      "learning_rate": 0.00020434985422740522,
+      "loss": 0.0544,
       "step": 11440
     },
     {
       "epoch": 16.45,
-      "learning_rate": 0.00013887144259077525,
-      "loss": 0.0838,
+      "learning_rate": 0.00020426239067055393,
+      "loss": 0.0474,
       "step": 11450
     },
     {
       "epoch": 16.47,
-      "learning_rate": 0.0001387242394504416,
-      "loss": 0.093,
+      "learning_rate": 0.00020417492711370262,
+      "loss": 0.0534,
       "step": 11460
     },
     {
       "epoch": 16.48,
-      "learning_rate": 0.00013857703631010794,
-      "loss": 0.1044,
+      "learning_rate": 0.0002040874635568513,
+      "loss": 0.0624,
       "step": 11470
     },
     {
       "epoch": 16.49,
-      "learning_rate": 0.00013842983316977427,
-      "loss": 0.0944,
+      "learning_rate": 0.000204,
+      "loss": 0.0476,
       "step": 11480
     },
     {
       "epoch": 16.51,
-      "learning_rate": 0.0001382826300294406,
-      "loss": 0.0967,
+      "learning_rate": 0.00020391253644314866,
+      "loss": 0.0496,
       "step": 11490
     },
     {
       "epoch": 16.52,
-      "learning_rate": 0.00013813542688910697,
-      "loss": 0.0815,
+      "learning_rate": 0.00020382507288629735,
+      "loss": 0.0467,
       "step": 11500
     },
     {
       "epoch": 16.52,
-      "eval_loss": 0.16426511108875275,
-      "eval_runtime": 39.87,
-      "eval_samples_per_second": 11.462,
-      "eval_steps_per_second": 1.455,
-      "eval_wer": 0.17361843095919288,
+      "eval_loss": 0.1587601602077484,
+      "eval_runtime": 40.0875,
+      "eval_samples_per_second": 11.4,
+      "eval_steps_per_second": 1.447,
+      "eval_wer": 0.1572052401746725,
       "step": 11500
     },
     {
       "epoch": 16.54,
-      "learning_rate": 0.0001379882237487733,
-      "loss": 0.1004,
+      "learning_rate": 0.00020373760932944603,
+      "loss": 0.0546,
       "step": 11510
     },
     {
       "epoch": 16.55,
-      "learning_rate": 0.00013784102060843963,
-      "loss": 0.0916,
+      "learning_rate": 0.00020365014577259472,
+      "loss": 0.0515,
       "step": 11520
     },
     {
       "epoch": 16.57,
-      "learning_rate": 0.00013769381746810597,
-      "loss": 0.0947,
+      "learning_rate": 0.0002035626822157434,
+      "loss": 0.0493,
       "step": 11530
     },
     {
       "epoch": 16.58,
-      "learning_rate": 0.00013754661432777233,
-      "loss": 0.0905,
+      "learning_rate": 0.00020347521865889213,
+      "loss": 0.0529,
       "step": 11540
     },
     {
       "epoch": 16.59,
-      "learning_rate": 0.00013739941118743866,
-      "loss": 0.0864,
+      "learning_rate": 0.00020338775510204081,
+      "loss": 0.0473,
       "step": 11550
     },
     {
       "epoch": 16.61,
-      "learning_rate": 0.000137252208047105,
-      "loss": 0.0852,
+      "learning_rate": 0.0002033002915451895,
+      "loss": 0.0439,
       "step": 11560
     },
     {
       "epoch": 16.62,
-      "learning_rate": 0.00013710500490677133,
-      "loss": 0.1071,
+      "learning_rate": 0.0002032128279883382,
+      "loss": 0.0563,
       "step": 11570
     },
     {
       "epoch": 16.64,
-      "learning_rate": 0.0001369578017664377,
-      "loss": 0.0939,
+      "learning_rate": 0.00020312536443148685,
+      "loss": 0.0501,
       "step": 11580
     },
     {
       "epoch": 16.65,
-      "learning_rate": 0.00013681059862610402,
-      "loss": 0.1154,
+      "learning_rate": 0.00020303790087463554,
+      "loss": 0.0624,
       "step": 11590
     },
     {
       "epoch": 16.67,
-      "learning_rate": 0.00013666339548577036,
-      "loss": 0.0898,
+      "learning_rate": 0.00020295043731778423,
+      "loss": 0.0507,
       "step": 11600
     },
     {
       "epoch": 16.67,
-      "eval_loss": 0.16324785351753235,
-      "eval_runtime": 39.9344,
-      "eval_samples_per_second": 11.444,
-      "eval_steps_per_second": 1.452,
-      "eval_wer": 0.17723234452642675,
+      "eval_loss": 0.16117839515209198,
+      "eval_runtime": 39.882,
+      "eval_samples_per_second": 11.459,
+      "eval_steps_per_second": 1.454,
+      "eval_wer": 0.16262611052552325,
       "step": 11600
     },
     {
       "epoch": 16.68,
-      "learning_rate": 0.0001365161923454367,
-      "loss": 0.0818,
+      "learning_rate": 0.00020286297376093291,
+      "loss": 0.0458,
       "step": 11610
     },
     {
       "epoch": 16.7,
-      "learning_rate": 0.00013636898920510302,
-      "loss": 0.0951,
+      "learning_rate": 0.0002027755102040816,
+      "loss": 0.0498,
       "step": 11620
     },
     {
       "epoch": 16.71,
-      "learning_rate": 0.00013622178606476938,
-      "loss": 0.1037,
+      "learning_rate": 0.00020268804664723032,
+      "loss": 0.0548,
       "step": 11630
     },
     {
       "epoch": 16.72,
-      "learning_rate": 0.00013607458292443572,
-      "loss": 0.0855,
+      "learning_rate": 0.000202600583090379,
+      "loss": 0.0446,
       "step": 11640
     },
     {
       "epoch": 16.74,
-      "learning_rate": 0.00013592737978410205,
-      "loss": 0.0795,
+      "learning_rate": 0.0002025131195335277,
+      "loss": 0.0469,
       "step": 11650
     },
     {
       "epoch": 16.75,
-      "learning_rate": 0.00013578017664376838,
-      "loss": 0.0877,
+      "learning_rate": 0.00020242565597667638,
+      "loss": 0.0471,
       "step": 11660
     },
     {
       "epoch": 16.77,
-      "learning_rate": 0.00013563297350343474,
-      "loss": 0.1126,
+      "learning_rate": 0.00020233819241982504,
+      "loss": 0.0673,
       "step": 11670
     },
     {
       "epoch": 16.78,
-      "learning_rate": 0.00013548577036310108,
-      "loss": 0.0809,
+      "learning_rate": 0.00020225072886297373,
+      "loss": 0.0393,
       "step": 11680
     },
     {
       "epoch": 16.8,
-      "learning_rate": 0.0001353385672227674,
-      "loss": 0.1051,
+      "learning_rate": 0.00020216326530612242,
+      "loss": 0.0569,
       "step": 11690
     },
     {
       "epoch": 16.81,
-      "learning_rate": 0.00013519136408243374,
-      "loss": 0.0958,
+      "learning_rate": 0.0002020758017492711,
+      "loss": 0.0558,
       "step": 11700
     },
     {
       "epoch": 16.81,
-      "eval_loss": 0.159054696559906,
-      "eval_runtime": 40.0796,
-      "eval_samples_per_second": 11.402,
-      "eval_steps_per_second": 1.447,
-      "eval_wer": 0.17467248908296942,
+      "eval_loss": 0.16593004763126373,
+      "eval_runtime": 39.9159,
+      "eval_samples_per_second": 11.449,
+      "eval_steps_per_second": 1.453,
+      "eval_wer": 0.16096973347387442,
       "step": 11700
     },
     {
       "epoch": 16.82,
-      "learning_rate": 0.0001350441609421001,
-      "loss": 0.0972,
+      "learning_rate": 0.00020198833819241982,
+      "loss": 0.0504,
       "step": 11710
     },
     {
       "epoch": 16.84,
-      "learning_rate": 0.00013489695780176644,
-      "loss": 0.0908,
+      "learning_rate": 0.0002019008746355685,
+      "loss": 0.0514,
       "step": 11720
     },
     {
       "epoch": 16.85,
-      "learning_rate": 0.00013474975466143274,
-      "loss": 0.1005,
+      "learning_rate": 0.0002018134110787172,
+      "loss": 0.0549,
       "step": 11730
     },
     {
       "epoch": 16.87,
-      "learning_rate": 0.0001346025515210991,
-      "loss": 0.1101,
+      "learning_rate": 0.00020172594752186588,
+      "loss": 0.0633,
       "step": 11740
     },
     {
       "epoch": 16.88,
-      "learning_rate": 0.00013445534838076544,
-      "loss": 0.1086,
+      "learning_rate": 0.00020163848396501457,
+      "loss": 0.0622,
       "step": 11750
     },
     {
       "epoch": 16.9,
-      "learning_rate": 0.0001343081452404318,
-      "loss": 0.0864,
+      "learning_rate": 0.00020155102040816323,
+      "loss": 0.041,
       "step": 11760
     },
     {
       "epoch": 16.91,
-      "learning_rate": 0.0001341609421000981,
-      "loss": 0.1002,
+      "learning_rate": 0.00020146355685131192,
+      "loss": 0.0537,
       "step": 11770
     },
     {
       "epoch": 16.93,
-      "learning_rate": 0.00013401373895976447,
-      "loss": 0.1145,
+      "learning_rate": 0.0002013760932944606,
+      "loss": 0.0601,
       "step": 11780
     },
     {
       "epoch": 16.94,
-      "learning_rate": 0.0001338665358194308,
-      "loss": 0.0848,
+      "learning_rate": 0.0002012886297376093,
+      "loss": 0.0484,
       "step": 11790
     },
     {
       "epoch": 16.95,
-      "learning_rate": 0.00013371933267909716,
-      "loss": 0.0986,
+      "learning_rate": 0.000201201166180758,
+      "loss": 0.0538,
       "step": 11800
     },
     {
       "epoch": 16.95,
-      "eval_loss": 0.16650956869125366,
-      "eval_runtime": 40.0741,
-      "eval_samples_per_second": 11.404,
-      "eval_steps_per_second": 1.447,
-      "eval_wer": 0.17331727149525675,
+      "eval_loss": 0.16603288054466248,
+      "eval_runtime": 39.9783,
+      "eval_samples_per_second": 11.431,
+      "eval_steps_per_second": 1.451,
+      "eval_wer": 0.15901219695828941,
       "step": 11800
     },
     {
       "epoch": 16.97,
-      "learning_rate": 0.00013357212953876347,
-      "loss": 0.0942,
+      "learning_rate": 0.0002011137026239067,
+      "loss": 0.045,
       "step": 11810
     },
     {
       "epoch": 16.98,
-      "learning_rate": 0.00013342492639842983,
-      "loss": 0.0868,
+      "learning_rate": 0.0002010262390670554,
+      "loss": 0.0527,
       "step": 11820
     },
     {
       "epoch": 17.0,
-      "learning_rate": 0.00013327772325809616,
-      "loss": 0.0857,
+      "learning_rate": 0.00020093877551020407,
+      "loss": 0.0446,
       "step": 11830
     },
     {
       "epoch": 17.01,
-      "learning_rate": 0.00013313052011776252,
-      "loss": 0.0911,
+      "learning_rate": 0.00020085131195335276,
+      "loss": 0.0503,
       "step": 11840
     },
     {
       "epoch": 17.03,
-      "learning_rate": 0.00013298331697742883,
-      "loss": 0.0743,
+      "learning_rate": 0.00020076384839650142,
+      "loss": 0.0435,
       "step": 11850
     },
     {
       "epoch": 17.04,
-      "learning_rate": 0.0001328361138370952,
-      "loss": 0.0727,
+      "learning_rate": 0.0002006763848396501,
+      "loss": 0.0375,
       "step": 11860
     },
     {
       "epoch": 17.05,
-      "learning_rate": 0.00013268891069676152,
-      "loss": 0.0905,
+      "learning_rate": 0.0002005889212827988,
+      "loss": 0.0511,
       "step": 11870
     },
     {
       "epoch": 17.07,
-      "learning_rate": 0.00013254170755642785,
-      "loss": 0.0724,
+      "learning_rate": 0.00020050145772594751,
+      "loss": 0.0388,
       "step": 11880
     },
     {
       "epoch": 17.08,
-      "learning_rate": 0.0001323945044160942,
-      "loss": 0.0872,
+      "learning_rate": 0.0002004139941690962,
+      "loss": 0.0468,
       "step": 11890
     },
     {
       "epoch": 17.1,
-      "learning_rate": 0.00013224730127576052,
-      "loss": 0.0843,
+      "learning_rate": 0.0002003265306122449,
+      "loss": 0.0465,
       "step": 11900
     },
     {
       "epoch": 17.1,
-      "eval_loss": 0.16180233657360077,
-      "eval_runtime": 40.2757,
-      "eval_samples_per_second": 11.347,
-      "eval_steps_per_second": 1.44,
-      "eval_wer": 0.17256437283541634,
+      "eval_loss": 0.16712433099746704,
+      "eval_runtime": 39.8942,
+      "eval_samples_per_second": 11.455,
+      "eval_steps_per_second": 1.454,
+      "eval_wer": 0.15359132660743863,
       "step": 11900
     },
     {
       "epoch": 17.11,
-      "learning_rate": 0.00013210009813542688,
-      "loss": 0.1044,
+      "learning_rate": 0.00020023906705539358,
+      "loss": 0.0585,
       "step": 11910
     },
     {
       "epoch": 17.13,
-      "learning_rate": 0.00013195289499509321,
-      "loss": 0.1075,
+      "learning_rate": 0.00020015160349854227,
+      "loss": 0.0591,
       "step": 11920
     },
     {
       "epoch": 17.14,
-      "learning_rate": 0.00013180569185475955,
-      "loss": 0.0632,
+      "learning_rate": 0.00020006413994169095,
+      "loss": 0.0358,
       "step": 11930
     },
     {
       "epoch": 17.16,
-      "learning_rate": 0.00013165848871442588,
-      "loss": 0.1181,
+      "learning_rate": 0.00019997667638483961,
+      "loss": 0.0715,
       "step": 11940
     },
     {
       "epoch": 17.17,
-      "learning_rate": 0.00013151128557409224,
-      "loss": 0.0798,
+      "learning_rate": 0.0001998892128279883,
+      "loss": 0.0445,
       "step": 11950
     },
     {
       "epoch": 17.18,
-      "learning_rate": 0.00013136408243375857,
-      "loss": 0.0976,
+      "learning_rate": 0.000199801749271137,
+      "loss": 0.0582,
       "step": 11960
     },
     {
       "epoch": 17.2,
-      "learning_rate": 0.0001312168792934249,
-      "loss": 0.1028,
+      "learning_rate": 0.0001997142857142857,
+      "loss": 0.0634,
       "step": 11970
     },
     {
       "epoch": 17.21,
-      "learning_rate": 0.00013106967615309124,
-      "loss": 0.0819,
+      "learning_rate": 0.0001996268221574344,
+      "loss": 0.0422,
       "step": 11980
     },
     {
       "epoch": 17.23,
-      "learning_rate": 0.0001309224730127576,
-      "loss": 0.0977,
+      "learning_rate": 0.00019953935860058308,
+      "loss": 0.0569,
       "step": 11990
     },
     {
       "epoch": 17.24,
-      "learning_rate": 0.00013077526987242394,
-      "loss": 0.0888,
+      "learning_rate": 0.00019945189504373177,
+      "loss": 0.0453,
       "step": 12000
     },
     {
       "epoch": 17.24,
-      "eval_loss": 0.1597016453742981,
-      "eval_runtime": 40.0997,
-      "eval_samples_per_second": 11.397,
-      "eval_steps_per_second": 1.446,
-      "eval_wer": 0.16985393765999096,
+      "eval_loss": 0.16343770921230316,
+      "eval_runtime": 40.2019,
+      "eval_samples_per_second": 11.368,
+      "eval_steps_per_second": 1.443,
+      "eval_wer": 0.15359132660743863,
       "step": 12000
     },
     {
       "epoch": 17.26,
-      "learning_rate": 0.00013062806673209027,
-      "loss": 0.0903,
+      "learning_rate": 0.00019936443148688046,
+      "loss": 0.0483,
       "step": 12010
     },
     {
       "epoch": 17.27,
-      "learning_rate": 0.0001304808635917566,
-      "loss": 0.0879,
+      "learning_rate": 0.00019927696793002914,
+      "loss": 0.0479,
       "step": 12020
     },
     {
       "epoch": 17.28,
-      "learning_rate": 0.00013033366045142294,
-      "loss": 0.0902,
+      "learning_rate": 0.0001991895043731778,
+      "loss": 0.0434,
       "step": 12030
     },
     {
       "epoch": 17.3,
-      "learning_rate": 0.0001301864573110893,
-      "loss": 0.0922,
+      "learning_rate": 0.0001991020408163265,
+      "loss": 0.0618,
       "step": 12040
     },
     {
       "epoch": 17.31,
-      "learning_rate": 0.00013003925417075563,
-      "loss": 0.0705,
+      "learning_rate": 0.00019901457725947518,
+      "loss": 0.0422,
       "step": 12050
     },
     {
       "epoch": 17.33,
-      "learning_rate": 0.00012989205103042196,
-      "loss": 0.0861,
+      "learning_rate": 0.0001989271137026239,
+      "loss": 0.0444,
       "step": 12060
     },
     {
       "epoch": 17.34,
-      "learning_rate": 0.0001297448478900883,
-      "loss": 0.0923,
+      "learning_rate": 0.00019883965014577258,
+      "loss": 0.0555,
       "step": 12070
     },
     {
       "epoch": 17.36,
-      "learning_rate": 0.00012959764474975466,
-      "loss": 0.0776,
+      "learning_rate": 0.00019876093294460642,
+      "loss": 0.0392,
       "step": 12080
     },
     {
       "epoch": 17.37,
-      "learning_rate": 0.000129450441609421,
-      "loss": 0.1092,
+      "learning_rate": 0.00019867346938775508,
+      "loss": 0.0604,
       "step": 12090
     },
     {
       "epoch": 17.39,
-      "learning_rate": 0.00012930323846908732,
-      "loss": 0.0851,
+      "learning_rate": 0.00019858600583090377,
+      "loss": 0.0459,
       "step": 12100
     },
     {
       "epoch": 17.39,
-      "eval_loss": 0.15801562368869781,
-      "eval_runtime": 40.1138,
-      "eval_samples_per_second": 11.393,
-      "eval_steps_per_second": 1.446,
-      "eval_wer": 0.1704562565878633,
+      "eval_loss": 0.16481797397136688,
+      "eval_runtime": 39.9889,
+      "eval_samples_per_second": 11.428,
+      "eval_steps_per_second": 1.45,
+      "eval_wer": 0.15856045776238517,
       "step": 12100
     },
     {
       "epoch": 17.4,
-      "learning_rate": 0.00012915603532875366,
-      "loss": 0.0997,
+      "learning_rate": 0.00019849854227405245,
+      "loss": 0.0508,
       "step": 12110
     },
     {
       "epoch": 17.41,
-      "learning_rate": 0.00012900883218842002,
-      "loss": 0.0931,
+      "learning_rate": 0.00019841107871720114,
+      "loss": 0.0506,
       "step": 12120
     },
     {
       "epoch": 17.43,
-      "learning_rate": 0.00012886162904808635,
-      "loss": 0.0745,
+      "learning_rate": 0.00019832361516034983,
+      "loss": 0.0386,
       "step": 12130
     },
     {
       "epoch": 17.44,
-      "learning_rate": 0.00012871442590775268,
-      "loss": 0.0965,
+      "learning_rate": 0.00019823615160349852,
+      "loss": 0.0586,
       "step": 12140
     },
     {
       "epoch": 17.46,
-      "learning_rate": 0.00012856722276741902,
-      "loss": 0.0909,
+      "learning_rate": 0.00019814868804664723,
+      "loss": 0.0486,
       "step": 12150
     },
     {
       "epoch": 17.47,
-      "learning_rate": 0.00012842001962708538,
-      "loss": 0.1066,
+      "learning_rate": 0.00019806122448979592,
+      "loss": 0.0655,
       "step": 12160
     },
     {
       "epoch": 17.49,
-      "learning_rate": 0.0001282728164867517,
-      "loss": 0.0924,
+      "learning_rate": 0.0001979737609329446,
+      "loss": 0.052,
       "step": 12170
     },
     {
       "epoch": 17.5,
-      "learning_rate": 0.00012812561334641804,
-      "loss": 0.0757,
+      "learning_rate": 0.00019788629737609327,
+      "loss": 0.0415,
       "step": 12180
     },
     {
       "epoch": 17.51,
-      "learning_rate": 0.00012797841020608438,
-      "loss": 0.0849,
+      "learning_rate": 0.00019779883381924196,
+      "loss": 0.0468,
       "step": 12190
     },
     {
       "epoch": 17.53,
-      "learning_rate": 0.0001278459273797841,
-      "loss": 0.0928,
+      "learning_rate": 0.00019771137026239064,
+      "loss": 0.0481,
       "step": 12200
     },
     {
       "epoch": 17.53,
-      "eval_loss": 0.16045060753822327,
-      "eval_runtime": 40.0909,
-      "eval_samples_per_second": 11.399,
-      "eval_steps_per_second": 1.447,
-      "eval_wer": 0.1695527781960548,
+      "eval_loss": 0.16811503469944,
+      "eval_runtime": 39.855,
+      "eval_samples_per_second": 11.467,
+      "eval_steps_per_second": 1.455,
+      "eval_wer": 0.1537419063394067,
       "step": 12200
     },
     {
       "epoch": 17.54,
-      "learning_rate": 0.00012769872423945043,
-      "loss": 0.1008,
+      "learning_rate": 0.00019762390670553933,
+      "loss": 0.0529,
       "step": 12210
     },
     {
       "epoch": 17.56,
-      "learning_rate": 0.00012755152109911676,
-      "loss": 0.1008,
+      "learning_rate": 0.00019753644314868802,
+      "loss": 0.0568,
       "step": 12220
     },
     {
       "epoch": 17.57,
-      "learning_rate": 0.0001274043179587831,
-      "loss": 0.0863,
+      "learning_rate": 0.0001974489795918367,
+      "loss": 0.0454,
       "step": 12230
     },
     {
       "epoch": 17.59,
-      "learning_rate": 0.00012725711481844946,
-      "loss": 0.0996,
+      "learning_rate": 0.00019736151603498542,
+      "loss": 0.0558,
       "step": 12240
     },
     {
       "epoch": 17.6,
-      "learning_rate": 0.0001271099116781158,
-      "loss": 0.067,
+      "learning_rate": 0.0001972740524781341,
+      "loss": 0.0351,
       "step": 12250
     },
     {
       "epoch": 17.61,
-      "learning_rate": 0.00012696270853778212,
-      "loss": 0.0988,
+      "learning_rate": 0.0001971865889212828,
+      "loss": 0.0534,
       "step": 12260
     },
     {
       "epoch": 17.63,
-      "learning_rate": 0.00012681550539744846,
-      "loss": 0.0883,
+      "learning_rate": 0.00019709912536443146,
+      "loss": 0.0451,
       "step": 12270
     },
     {
       "epoch": 17.64,
-      "learning_rate": 0.0001266683022571148,
-      "loss": 0.0811,
+      "learning_rate": 0.00019701166180758015,
+      "loss": 0.0438,
       "step": 12280
     },
     {
       "epoch": 17.66,
-      "learning_rate": 0.00012652109911678115,
-      "loss": 0.0999,
+      "learning_rate": 0.00019692419825072884,
+      "loss": 0.055,
       "step": 12290
     },
     {
       "epoch": 17.67,
-      "learning_rate": 0.00012637389597644749,
-      "loss": 0.0886,
+      "learning_rate": 0.00019683673469387752,
+      "loss": 0.0506,
       "step": 12300
     },
     {
       "epoch": 17.67,
-      "eval_loss": 0.1619402915239334,
-      "eval_runtime": 40.3366,
-      "eval_samples_per_second": 11.33,
-      "eval_steps_per_second": 1.438,
-      "eval_wer": 0.16940219846408675,
+      "eval_loss": 0.16215792298316956,
+      "eval_runtime": 40.1021,
+      "eval_samples_per_second": 11.396,
+      "eval_steps_per_second": 1.446,
+      "eval_wer": 0.15750639963860863,
       "step": 12300
     },
     {
       "epoch": 17.69,
-      "learning_rate": 0.00012622669283611382,
-      "loss": 0.0936,
+      "learning_rate": 0.0001967492711370262,
+      "loss": 0.0472,
       "step": 12310
     },
     {
       "epoch": 17.7,
-      "learning_rate": 0.00012607948969578015,
-      "loss": 0.1139,
+      "learning_rate": 0.0001966618075801749,
+      "loss": 0.0624,
       "step": 12320
     },
     {
       "epoch": 17.72,
-      "learning_rate": 0.0001259322865554465,
-      "loss": 0.0851,
+      "learning_rate": 0.00019657434402332361,
+      "loss": 0.0492,
       "step": 12330
     },
     {
       "epoch": 17.73,
-      "learning_rate": 0.00012578508341511285,
-      "loss": 0.1101,
+      "learning_rate": 0.0001964868804664723,
+      "loss": 0.0607,
       "step": 12340
     },
     {
       "epoch": 17.74,
-      "learning_rate": 0.00012563788027477918,
-      "loss": 0.0792,
+      "learning_rate": 0.000196399416909621,
+      "loss": 0.0454,
       "step": 12350
     },
     {
       "epoch": 17.76,
-      "learning_rate": 0.0001254906771344455,
-      "loss": 0.0789,
+      "learning_rate": 0.00019631195335276965,
+      "loss": 0.0457,
       "step": 12360
     },
     {
       "epoch": 17.77,
-      "learning_rate": 0.00012534347399411187,
-      "loss": 0.1043,
+      "learning_rate": 0.00019622448979591834,
+      "loss": 0.0618,
       "step": 12370
     },
     {
       "epoch": 17.79,
-      "learning_rate": 0.0001251962708537782,
-      "loss": 0.0701,
+      "learning_rate": 0.00019613702623906703,
+      "loss": 0.034,
       "step": 12380
     },
     {
       "epoch": 17.8,
-      "learning_rate": 0.00012504906771344454,
-      "loss": 0.0817,
+      "learning_rate": 0.00019604956268221571,
+      "loss": 0.0467,
       "step": 12390
     },
     {
       "epoch": 17.82,
-      "learning_rate": 0.00012490186457311087,
-      "loss": 0.0931,
+      "learning_rate": 0.0001959620991253644,
+      "loss": 0.057,
       "step": 12400
     },
     {
       "epoch": 17.82,
-      "eval_loss": 0.15292048454284668,
-      "eval_runtime": 40.0068,
-      "eval_samples_per_second": 11.423,
+      "eval_loss": 0.15522630512714386,
+      "eval_runtime": 39.9906,
+      "eval_samples_per_second": 11.428,
       "eval_steps_per_second": 1.45,
-      "eval_wer": 0.16819756060834212,
+      "eval_wer": 0.15298900767956633,
       "step": 12400
     },
     {
       "epoch": 17.83,
-      "learning_rate": 0.0001247546614327772,
-      "loss": 0.1021,
+      "learning_rate": 0.00019587463556851312,
+      "loss": 0.0523,
       "step": 12410
     },
     {
       "epoch": 17.84,
-      "learning_rate": 0.00012460745829244357,
-      "loss": 0.0851,
+      "learning_rate": 0.0001957871720116618,
+      "loss": 0.0505,
       "step": 12420
     },
     {
       "epoch": 17.86,
-      "learning_rate": 0.0001244602551521099,
-      "loss": 0.07,
+      "learning_rate": 0.0001956997084548105,
+      "loss": 0.0351,
       "step": 12430
     },
     {
       "epoch": 17.87,
-      "learning_rate": 0.00012431305201177623,
-      "loss": 0.108,
+      "learning_rate": 0.00019561224489795918,
+      "loss": 0.0668,
       "step": 12440
     },
     {
       "epoch": 17.89,
-      "learning_rate": 0.00012416584887144257,
-      "loss": 0.0826,
+      "learning_rate": 0.00019552478134110784,
+      "loss": 0.053,
       "step": 12450
     },
     {
       "epoch": 17.9,
-      "learning_rate": 0.00012401864573110893,
-      "loss": 0.0834,
+      "learning_rate": 0.00019543731778425653,
+      "loss": 0.0562,
       "step": 12460
     },
     {
       "epoch": 17.92,
-      "learning_rate": 0.00012387144259077526,
-      "loss": 0.082,
+      "learning_rate": 0.00019534985422740522,
+      "loss": 0.0597,
       "step": 12470
     },
     {
       "epoch": 17.93,
-      "learning_rate": 0.0001237242394504416,
-      "loss": 0.0886,
+      "learning_rate": 0.0001952623906705539,
+      "loss": 0.06,
       "step": 12480
     },
     {
       "epoch": 17.95,
-      "learning_rate": 0.00012357703631010793,
-      "loss": 0.0873,
+      "learning_rate": 0.0001951749271137026,
+      "loss": 0.0595,
       "step": 12490
     },
     {
       "epoch": 17.96,
-      "learning_rate": 0.0001234298331697743,
-      "loss": 0.0747,
+      "learning_rate": 0.0001950874635568513,
+      "loss": 0.0492,
       "step": 12500
     },
     {
       "epoch": 17.96,
-      "eval_loss": 0.14587688446044922,
-      "eval_runtime": 40.2304,
-      "eval_samples_per_second": 11.36,
-      "eval_steps_per_second": 1.442,
-      "eval_wer": 0.16277669025749134,
+      "eval_loss": 0.15482190251350403,
+      "eval_runtime": 39.803,
+      "eval_samples_per_second": 11.482,
+      "eval_steps_per_second": 1.457,
+      "eval_wer": 0.15148321035988557,
       "step": 12500
     },
     {
       "epoch": 17.97,
-      "learning_rate": 0.00012328263002944062,
-      "loss": 0.0806,
+      "learning_rate": 0.000195,
+      "loss": 0.0518,
       "step": 12510
     },
     {
       "epoch": 17.99,
-      "learning_rate": 0.00012313542688910696,
-      "loss": 0.0835,
+      "learning_rate": 0.00019491253644314868,
+      "loss": 0.0581,
       "step": 12520
     },
     {
       "epoch": 18.0,
-      "learning_rate": 0.0001229882237487733,
-      "loss": 0.0926,
+      "learning_rate": 0.00019482507288629737,
+      "loss": 0.0601,
       "step": 12530
     },
     {
       "epoch": 18.02,
-      "learning_rate": 0.00012284102060843965,
-      "loss": 0.0802,
+      "learning_rate": 0.00019473760932944603,
+      "loss": 0.0456,
       "step": 12540
     },
     {
       "epoch": 18.03,
-      "learning_rate": 0.00012269381746810598,
-      "loss": 0.0666,
+      "learning_rate": 0.00019465014577259472,
+      "loss": 0.039,
       "step": 12550
     },
     {
       "epoch": 18.05,
-      "learning_rate": 0.00012254661432777232,
-      "loss": 0.0892,
+      "learning_rate": 0.0001945626822157434,
+      "loss": 0.0476,
       "step": 12560
     },
     {
       "epoch": 18.06,
-      "learning_rate": 0.00012239941118743865,
-      "loss": 0.0741,
+      "learning_rate": 0.0001944752186588921,
+      "loss": 0.0403,
       "step": 12570
     },
     {
       "epoch": 18.07,
-      "learning_rate": 0.00012225220804710498,
-      "loss": 0.0788,
+      "learning_rate": 0.0001943877551020408,
+      "loss": 0.0417,
       "step": 12580
     },
     {
       "epoch": 18.09,
-      "learning_rate": 0.00012210500490677134,
-      "loss": 0.0796,
+      "learning_rate": 0.0001943002915451895,
+      "loss": 0.0489,
       "step": 12590
     },
     {
       "epoch": 18.1,
-      "learning_rate": 0.00012195780176643766,
-      "loss": 0.1014,
+      "learning_rate": 0.0001942128279883382,
+      "loss": 0.0555,
       "step": 12600
     },
     {
       "epoch": 18.1,
-      "eval_loss": 0.15678279101848602,
-      "eval_runtime": 40.1685,
-      "eval_samples_per_second": 11.377,
-      "eval_steps_per_second": 1.444,
-      "eval_wer": 0.16608944436078904,
+      "eval_loss": 0.1575390100479126,
+      "eval_runtime": 39.9057,
+      "eval_samples_per_second": 11.452,
+      "eval_steps_per_second": 1.453,
+      "eval_wer": 0.1532901671435025,
       "step": 12600
     },
     {
       "epoch": 18.12,
-      "learning_rate": 0.00012181059862610401,
-      "loss": 0.0859,
+      "learning_rate": 0.00019412536443148688,
+      "loss": 0.0471,
       "step": 12610
     },
     {
       "epoch": 18.13,
-      "learning_rate": 0.00012166339548577034,
-      "loss": 0.0854,
+      "learning_rate": 0.00019403790087463556,
+      "loss": 0.0501,
       "step": 12620
     },
     {
       "epoch": 18.15,
-      "learning_rate": 0.00012151619234543669,
-      "loss": 0.078,
+      "learning_rate": 0.00019395043731778422,
+      "loss": 0.0415,
       "step": 12630
     },
     {
       "epoch": 18.16,
-      "learning_rate": 0.00012136898920510302,
-      "loss": 0.0905,
+      "learning_rate": 0.0001938629737609329,
+      "loss": 0.0538,
       "step": 12640
     },
     {
       "epoch": 18.18,
-      "learning_rate": 0.00012122178606476937,
-      "loss": 0.0968,
+      "learning_rate": 0.0001937755102040816,
+      "loss": 0.052,
       "step": 12650
     },
     {
       "epoch": 18.19,
-      "learning_rate": 0.0001210745829244357,
-      "loss": 0.0987,
+      "learning_rate": 0.0001936880466472303,
+      "loss": 0.0515,
       "step": 12660
     },
     {
       "epoch": 18.2,
-      "learning_rate": 0.00012092737978410205,
-      "loss": 0.0743,
+      "learning_rate": 0.000193600583090379,
+      "loss": 0.0416,
       "step": 12670
     },
     {
       "epoch": 18.22,
-      "learning_rate": 0.00012078017664376838,
-      "loss": 0.0799,
+      "learning_rate": 0.0001935131195335277,
+      "loss": 0.0406,
       "step": 12680
     },
     {
       "epoch": 18.23,
-      "learning_rate": 0.00012063297350343472,
-      "loss": 0.087,
+      "learning_rate": 0.00019342565597667638,
+      "loss": 0.0448,
       "step": 12690
     },
     {
       "epoch": 18.25,
-      "learning_rate": 0.00012048577036310106,
-      "loss": 0.0729,
+      "learning_rate": 0.00019333819241982507,
+      "loss": 0.0364,
       "step": 12700
     },
     {
       "epoch": 18.25,
-      "eval_loss": 0.1656101793050766,
-      "eval_runtime": 39.9566,
-      "eval_samples_per_second": 11.437,
+      "eval_loss": 0.1611556112766266,
+      "eval_runtime": 39.9322,
+      "eval_samples_per_second": 11.444,
       "eval_steps_per_second": 1.452,
-      "eval_wer": 0.17120915524770366,
+      "eval_wer": 0.1492245143803644,
       "step": 12700
     },
     {
       "epoch": 18.26,
-      "learning_rate": 0.0001203385672227674,
-      "loss": 0.0898,
+      "learning_rate": 0.00019325072886297375,
+      "loss": 0.0508,
       "step": 12710
     },
     {
       "epoch": 18.28,
-      "learning_rate": 0.00012019136408243375,
-      "loss": 0.0798,
+      "learning_rate": 0.00019316326530612242,
+      "loss": 0.0462,
       "step": 12720
     },
     {
       "epoch": 18.29,
-      "learning_rate": 0.00012004416094210008,
-      "loss": 0.0952,
+      "learning_rate": 0.0001930758017492711,
+      "loss": 0.0587,
       "step": 12730
     },
     {
       "epoch": 18.3,
-      "learning_rate": 0.00011989695780176643,
-      "loss": 0.0874,
+      "learning_rate": 0.0001929883381924198,
+      "loss": 0.05,
       "step": 12740
     },
     {
       "epoch": 18.32,
-      "learning_rate": 0.00011974975466143276,
-      "loss": 0.0713,
+      "learning_rate": 0.00019290087463556848,
+      "loss": 0.0434,
       "step": 12750
     },
     {
       "epoch": 18.33,
-      "learning_rate": 0.0001196025515210991,
-      "loss": 0.0865,
+      "learning_rate": 0.0001928134110787172,
+      "loss": 0.0483,
       "step": 12760
     },
     {
       "epoch": 18.35,
-      "learning_rate": 0.00011945534838076544,
-      "loss": 0.073,
+      "learning_rate": 0.00019272594752186588,
+      "loss": 0.0404,
       "step": 12770
     },
     {
       "epoch": 18.36,
-      "learning_rate": 0.00011930814524043179,
-      "loss": 0.0902,
+      "learning_rate": 0.00019263848396501457,
+      "loss": 0.0468,
       "step": 12780
     },
     {
       "epoch": 18.38,
-      "learning_rate": 0.00011916094210009812,
-      "loss": 0.0766,
+      "learning_rate": 0.00019255102040816326,
+      "loss": 0.0445,
       "step": 12790
     },
     {
       "epoch": 18.39,
-      "learning_rate": 0.00011901373895976447,
-      "loss": 0.0884,
+      "learning_rate": 0.00019246355685131195,
+      "loss": 0.0517,
       "step": 12800
     },
     {
       "epoch": 18.39,
-      "eval_loss": 0.15617236495018005,
-      "eval_runtime": 40.4371,
-      "eval_samples_per_second": 11.302,
-      "eval_steps_per_second": 1.434,
-      "eval_wer": 0.16789640114440596,
+      "eval_loss": 0.1672176569700241,
+      "eval_runtime": 39.9361,
+      "eval_samples_per_second": 11.443,
+      "eval_steps_per_second": 1.452,
+      "eval_wer": 0.1511820508959494,
       "step": 12800
     },
     {
       "epoch": 18.41,
-      "learning_rate": 0.0001188665358194308,
-      "loss": 0.0802,
+      "learning_rate": 0.0001923760932944606,
+      "loss": 0.0504,
       "step": 12810
     },
     {
       "epoch": 18.42,
-      "learning_rate": 0.00011871933267909715,
-      "loss": 0.0701,
+      "learning_rate": 0.0001922886297376093,
+      "loss": 0.0403,
       "step": 12820
     },
     {
       "epoch": 18.43,
-      "learning_rate": 0.00011857212953876348,
-      "loss": 0.0678,
+      "learning_rate": 0.00019220116618075798,
+      "loss": 0.042,
       "step": 12830
     },
     {
       "epoch": 18.45,
-      "learning_rate": 0.00011842492639842981,
-      "loss": 0.08,
+      "learning_rate": 0.0001921137026239067,
+      "loss": 0.0442,
       "step": 12840
     },
     {
       "epoch": 18.46,
-      "learning_rate": 0.00011827772325809616,
-      "loss": 0.0661,
+      "learning_rate": 0.00019202623906705538,
+      "loss": 0.0394,
       "step": 12850
     },
     {
       "epoch": 18.48,
-      "learning_rate": 0.0001181305201177625,
-      "loss": 0.0891,
+      "learning_rate": 0.00019193877551020407,
+      "loss": 0.0494,
       "step": 12860
     },
     {
       "epoch": 18.49,
-      "learning_rate": 0.00011798331697742884,
-      "loss": 0.0792,
+      "learning_rate": 0.00019185131195335276,
+      "loss": 0.0417,
       "step": 12870
     },
     {
       "epoch": 18.51,
-      "learning_rate": 0.00011783611383709517,
-      "loss": 0.0976,
+      "learning_rate": 0.00019176384839650145,
+      "loss": 0.0476,
       "step": 12880
     },
     {
       "epoch": 18.52,
-      "learning_rate": 0.00011768891069676152,
-      "loss": 0.0936,
+      "learning_rate": 0.00019167638483965014,
+      "loss": 0.0571,
       "step": 12890
     },
     {
       "epoch": 18.53,
-      "learning_rate": 0.00011754170755642785,
-      "loss": 0.0823,
+      "learning_rate": 0.0001915889212827988,
+      "loss": 0.0444,
       "step": 12900
     },
     {
       "epoch": 18.53,
-      "eval_loss": 0.15830214321613312,
-      "eval_runtime": 40.3114,
-      "eval_samples_per_second": 11.337,
-      "eval_steps_per_second": 1.439,
-      "eval_wer": 0.16970335792802288,
+      "eval_loss": 0.15720723569393158,
+      "eval_runtime": 39.8966,
+      "eval_samples_per_second": 11.455,
+      "eval_steps_per_second": 1.454,
+      "eval_wer": 0.15163379009185363,
       "step": 12900
     },
     {
       "epoch": 18.55,
-      "learning_rate": 0.0001173945044160942,
-      "loss": 0.0743,
+      "learning_rate": 0.00019150145772594749,
+      "loss": 0.0466,
       "step": 12910
     },
     {
       "epoch": 18.56,
-      "learning_rate": 0.00011724730127576054,
-      "loss": 0.0868,
+      "learning_rate": 0.00019141399416909617,
+      "loss": 0.0483,
       "step": 12920
     },
     {
       "epoch": 18.58,
-      "learning_rate": 0.00011710009813542688,
-      "loss": 0.0812,
+      "learning_rate": 0.0001913265306122449,
+      "loss": 0.0495,
       "step": 12930
     },
     {
       "epoch": 18.59,
-      "learning_rate": 0.00011695289499509322,
-      "loss": 0.0652,
+      "learning_rate": 0.00019123906705539358,
+      "loss": 0.0393,
       "step": 12940
     },
     {
       "epoch": 18.61,
-      "learning_rate": 0.00011680569185475956,
-      "loss": 0.0738,
+      "learning_rate": 0.00019115160349854226,
+      "loss": 0.0394,
       "step": 12950
     },
     {
       "epoch": 18.62,
-      "learning_rate": 0.0001166584887144259,
-      "loss": 0.1027,
+      "learning_rate": 0.00019106413994169095,
+      "loss": 0.0639,
       "step": 12960
     },
     {
       "epoch": 18.64,
-      "learning_rate": 0.00011651128557409224,
-      "loss": 0.0777,
+      "learning_rate": 0.00019097667638483964,
+      "loss": 0.0442,
       "step": 12970
     },
     {
       "epoch": 18.65,
-      "learning_rate": 0.00011636408243375858,
-      "loss": 0.1243,
+      "learning_rate": 0.00019088921282798833,
+      "loss": 0.0807,
       "step": 12980
     },
     {
       "epoch": 18.66,
-      "learning_rate": 0.00011621687929342491,
-      "loss": 0.092,
+      "learning_rate": 0.000190801749271137,
+      "loss": 0.0581,
       "step": 12990
     },
     {
       "epoch": 18.68,
-      "learning_rate": 0.00011606967615309126,
-      "loss": 0.066,
+      "learning_rate": 0.00019071428571428568,
+      "loss": 0.0413,
       "step": 13000
     },
     {
       "epoch": 18.68,
-      "eval_loss": 0.1592216044664383,
-      "eval_runtime": 40.6807,
-      "eval_samples_per_second": 11.234,
-      "eval_steps_per_second": 1.426,
-      "eval_wer": 0.17166089444360788,
+      "eval_loss": 0.1595200002193451,
+      "eval_runtime": 39.866,
+      "eval_samples_per_second": 11.463,
+      "eval_steps_per_second": 1.455,
+      "eval_wer": 0.14877277518446017,
       "step": 13000
     },
     {
       "epoch": 18.69,
-      "learning_rate": 0.00011592247301275759,
-      "loss": 0.08,
+      "learning_rate": 0.0001906268221574344,
+      "loss": 0.0487,
       "step": 13010
     },
     {
       "epoch": 18.71,
-      "learning_rate": 0.00011577526987242394,
-      "loss": 0.0768,
+      "learning_rate": 0.00019053935860058308,
+      "loss": 0.0526,
       "step": 13020
     },
     {
       "epoch": 18.72,
-      "learning_rate": 0.00011562806673209027,
-      "loss": 0.078,
+      "learning_rate": 0.00019045189504373177,
+      "loss": 0.0432,
       "step": 13030
     },
     {
       "epoch": 18.74,
-      "learning_rate": 0.00011548086359175662,
-      "loss": 0.0931,
+      "learning_rate": 0.00019036443148688045,
+      "loss": 0.0537,
       "step": 13040
     },
     {
       "epoch": 18.75,
-      "learning_rate": 0.00011533366045142295,
-      "loss": 0.0784,
+      "learning_rate": 0.00019027696793002914,
+      "loss": 0.0455,
       "step": 13050
     },
     {
       "epoch": 18.76,
-      "learning_rate": 0.0001151864573110893,
-      "loss": 0.0902,
+      "learning_rate": 0.00019018950437317783,
+      "loss": 0.051,
       "step": 13060
     },
     {
       "epoch": 18.78,
-      "learning_rate": 0.00011503925417075563,
-      "loss": 0.0885,
+      "learning_rate": 0.00019010204081632652,
+      "loss": 0.05,
       "step": 13070
     },
     {
       "epoch": 18.79,
-      "learning_rate": 0.00011489205103042198,
-      "loss": 0.0786,
+      "learning_rate": 0.00019001457725947518,
+      "loss": 0.0452,
       "step": 13080
     },
     {
       "epoch": 18.81,
-      "learning_rate": 0.00011474484789008831,
-      "loss": 0.0763,
+      "learning_rate": 0.00018992711370262387,
+      "loss": 0.0445,
       "step": 13090
     },
     {
       "epoch": 18.82,
-      "learning_rate": 0.00011459764474975466,
-      "loss": 0.0849,
+      "learning_rate": 0.00018983965014577258,
+      "loss": 0.0621,
       "step": 13100
     },
     {
       "epoch": 18.82,
-      "eval_loss": 0.15664438903331757,
-      "eval_runtime": 40.0732,
-      "eval_samples_per_second": 11.404,
-      "eval_steps_per_second": 1.447,
-      "eval_wer": 0.16684234302062942,
+      "eval_loss": 0.15519918501377106,
+      "eval_runtime": 40.5387,
+      "eval_samples_per_second": 11.273,
+      "eval_steps_per_second": 1.431,
+      "eval_wer": 0.14817045625658787,
       "step": 13100
     },
     {
       "epoch": 18.84,
-      "learning_rate": 0.00011445044160942099,
-      "loss": 0.0843,
+      "learning_rate": 0.00018975218658892127,
+      "loss": 0.0501,
       "step": 13110
     },
     {
       "epoch": 18.85,
-      "learning_rate": 0.00011430323846908734,
-      "loss": 0.0776,
+      "learning_rate": 0.00018966472303206996,
+      "loss": 0.049,
       "step": 13120
     },
     {
       "epoch": 18.86,
-      "learning_rate": 0.00011415603532875367,
-      "loss": 0.0597,
+      "learning_rate": 0.00018957725947521865,
+      "loss": 0.039,
       "step": 13130
     },
     {
       "epoch": 18.88,
-      "learning_rate": 0.00011400883218842,
-      "loss": 0.0803,
+      "learning_rate": 0.00018948979591836733,
+      "loss": 0.0542,
       "step": 13140
     },
     {
       "epoch": 18.89,
-      "learning_rate": 0.00011386162904808635,
-      "loss": 0.0802,
+      "learning_rate": 0.00018940233236151602,
+      "loss": 0.0485,
       "step": 13150
     },
     {
       "epoch": 18.91,
-      "learning_rate": 0.00011371442590775269,
-      "loss": 0.0719,
+      "learning_rate": 0.0001893148688046647,
+      "loss": 0.0452,
       "step": 13160
     },
     {
       "epoch": 18.92,
-      "learning_rate": 0.00011356722276741903,
-      "loss": 0.0707,
+      "learning_rate": 0.00018922740524781337,
+      "loss": 0.0457,
       "step": 13170
     },
     {
       "epoch": 18.94,
-      "learning_rate": 0.00011342001962708537,
-      "loss": 0.0741,
+      "learning_rate": 0.00018913994169096206,
+      "loss": 0.0465,
       "step": 13180
     },
     {
       "epoch": 18.95,
-      "learning_rate": 0.00011327281648675171,
-      "loss": 0.0834,
+      "learning_rate": 0.00018905247813411077,
+      "loss": 0.0556,
       "step": 13190
     },
     {
       "epoch": 18.97,
-      "learning_rate": 0.00011312561334641805,
-      "loss": 0.0706,
+      "learning_rate": 0.00018896501457725946,
+      "loss": 0.04,
       "step": 13200
     },
     {
       "epoch": 18.97,
-      "eval_loss": 0.16251307725906372,
-      "eval_runtime": 39.9205,
-      "eval_samples_per_second": 11.448,
+      "eval_loss": 0.15459690988063812,
+      "eval_runtime": 39.9064,
+      "eval_samples_per_second": 11.452,
       "eval_steps_per_second": 1.453,
-      "eval_wer": 0.16774582141243788,
+      "eval_wer": 0.14696581840084325,
       "step": 13200
     },
     {
       "epoch": 18.98,
-      "learning_rate": 0.00011297841020608439,
-      "loss": 0.0606,
+      "learning_rate": 0.00018887755102040815,
+      "loss": 0.0382,
       "step": 13210
     },
     {
       "epoch": 18.99,
-      "learning_rate": 0.00011283120706575073,
-      "loss": 0.0619,
+      "learning_rate": 0.00018879008746355684,
+      "loss": 0.0354,
       "step": 13220
     },
     {
       "epoch": 19.01,
-      "learning_rate": 0.00011268400392541707,
-      "loss": 0.0671,
+      "learning_rate": 0.00018870262390670553,
+      "loss": 0.0414,
       "step": 13230
     },
     {
       "epoch": 19.02,
-      "learning_rate": 0.00011253680078508341,
-      "loss": 0.0765,
+      "learning_rate": 0.0001886151603498542,
+      "loss": 0.0463,
       "step": 13240
     },
     {
       "epoch": 19.04,
-      "learning_rate": 0.00011238959764474975,
-      "loss": 0.0798,
+      "learning_rate": 0.0001885276967930029,
+      "loss": 0.0497,
       "step": 13250
     },
     {
       "epoch": 19.05,
-      "learning_rate": 0.00011224239450441609,
-      "loss": 0.0822,
+      "learning_rate": 0.00018844023323615156,
+      "loss": 0.0504,
       "step": 13260
     },
     {
       "epoch": 19.07,
-      "learning_rate": 0.00011209519136408243,
-      "loss": 0.0743,
+      "learning_rate": 0.00018835276967930028,
+      "loss": 0.0424,
       "step": 13270
     },
     {
       "epoch": 19.08,
-      "learning_rate": 0.00011194798822374877,
-      "loss": 0.0842,
+      "learning_rate": 0.00018826530612244896,
+      "loss": 0.0491,
       "step": 13280
     },
     {
       "epoch": 19.09,
-      "learning_rate": 0.0001118007850834151,
-      "loss": 0.08,
+      "learning_rate": 0.00018817784256559765,
+      "loss": 0.0485,
       "step": 13290
     },
     {
       "epoch": 19.11,
-      "learning_rate": 0.00011165358194308145,
-      "loss": 0.0693,
+      "learning_rate": 0.00018809037900874634,
+      "loss": 0.0394,
       "step": 13300
     },
     {
       "epoch": 19.11,
-      "eval_loss": 0.16297851502895355,
-      "eval_runtime": 40.169,
-      "eval_samples_per_second": 11.377,
-      "eval_steps_per_second": 1.444,
-      "eval_wer": 0.16910103900015058,
+      "eval_loss": 0.16309237480163574,
+      "eval_runtime": 39.9257,
+      "eval_samples_per_second": 11.446,
+      "eval_steps_per_second": 1.453,
+      "eval_wer": 0.15283842794759825,
       "step": 13300
     },
     {
       "epoch": 19.12,
-      "learning_rate": 0.00011150637880274778,
-      "loss": 0.0847,
+      "learning_rate": 0.00018800291545189503,
+      "loss": 0.0537,
       "step": 13310
     },
     {
       "epoch": 19.14,
-      "learning_rate": 0.00011135917566241413,
-      "loss": 0.0775,
+      "learning_rate": 0.00018791545189504372,
+      "loss": 0.0444,
       "step": 13320
     },
     {
       "epoch": 19.15,
-      "learning_rate": 0.00011121197252208046,
-      "loss": 0.0876,
+      "learning_rate": 0.0001878279883381924,
+      "loss": 0.0577,
       "step": 13330
     },
     {
       "epoch": 19.17,
-      "learning_rate": 0.00011106476938174681,
-      "loss": 0.0912,
+      "learning_rate": 0.00018774052478134112,
+      "loss": 0.056,
       "step": 13340
     },
     {
       "epoch": 19.18,
-      "learning_rate": 0.00011091756624141314,
-      "loss": 0.0635,
+      "learning_rate": 0.00018765306122448975,
+      "loss": 0.0365,
       "step": 13350
     },
     {
       "epoch": 19.2,
-      "learning_rate": 0.00011077036310107949,
-      "loss": 0.0762,
+      "learning_rate": 0.00018756559766763847,
+      "loss": 0.048,
       "step": 13360
     },
     {
       "epoch": 19.21,
-      "learning_rate": 0.00011062315996074582,
-      "loss": 0.0885,
+      "learning_rate": 0.00018747813411078716,
+      "loss": 0.0582,
       "step": 13370
     },
     {
       "epoch": 19.22,
-      "learning_rate": 0.00011047595682041217,
-      "loss": 0.0821,
+      "learning_rate": 0.00018739067055393584,
+      "loss": 0.056,
       "step": 13380
     },
     {
       "epoch": 19.24,
-      "learning_rate": 0.0001103287536800785,
-      "loss": 0.0731,
+      "learning_rate": 0.00018730320699708453,
+      "loss": 0.0473,
       "step": 13390
     },
     {
       "epoch": 19.25,
-      "learning_rate": 0.00011018155053974485,
-      "loss": 0.0736,
+      "learning_rate": 0.00018721574344023322,
+      "loss": 0.0465,
       "step": 13400
     },
     {
       "epoch": 19.25,
-      "eval_loss": 0.16010314226150513,
-      "eval_runtime": 40.0705,
-      "eval_samples_per_second": 11.405,
-      "eval_steps_per_second": 1.447,
-      "eval_wer": 0.16714350248456558,
+      "eval_loss": 0.16543611884117126,
+      "eval_runtime": 40.0496,
+      "eval_samples_per_second": 11.411,
+      "eval_steps_per_second": 1.448,
+      "eval_wer": 0.14967625357626863,
       "step": 13400
     },
     {
       "epoch": 19.27,
-      "learning_rate": 0.00011003434739941118,
-      "loss": 0.0817,
+      "learning_rate": 0.0001871282798833819,
+      "loss": 0.0526,
       "step": 13410
     },
     {
       "epoch": 19.28,
-      "learning_rate": 0.00010988714425907753,
-      "loss": 0.0842,
+      "learning_rate": 0.0001870408163265306,
+      "loss": 0.0448,
       "step": 13420
     },
     {
       "epoch": 19.3,
-      "learning_rate": 0.00010973994111874386,
-      "loss": 0.0852,
+      "learning_rate": 0.0001869533527696793,
+      "loss": 0.0571,
       "step": 13430
     },
     {
       "epoch": 19.31,
-      "learning_rate": 0.00010959273797841018,
-      "loss": 0.0867,
+      "learning_rate": 0.00018686588921282794,
+      "loss": 0.0514,
       "step": 13440
     },
     {
       "epoch": 19.32,
-      "learning_rate": 0.00010944553483807654,
-      "loss": 0.0984,
+      "learning_rate": 0.00018677842565597666,
+      "loss": 0.0644,
       "step": 13450
     },
     {
       "epoch": 19.34,
-      "learning_rate": 0.00010929833169774286,
-      "loss": 0.0867,
+      "learning_rate": 0.00018669096209912535,
+      "loss": 0.0626,
       "step": 13460
     },
     {
       "epoch": 19.35,
-      "learning_rate": 0.00010915112855740922,
-      "loss": 0.0619,
+      "learning_rate": 0.00018660349854227403,
+      "loss": 0.0391,
       "step": 13470
     },
     {
       "epoch": 19.37,
-      "learning_rate": 0.00010900392541707554,
-      "loss": 0.089,
+      "learning_rate": 0.00018651603498542272,
+      "loss": 0.0568,
       "step": 13480
     },
     {
       "epoch": 19.38,
-      "learning_rate": 0.0001088567222767419,
-      "loss": 0.0666,
+      "learning_rate": 0.0001864285714285714,
+      "loss": 0.0375,
       "step": 13490
     },
     {
       "epoch": 19.4,
-      "learning_rate": 0.00010870951913640822,
-      "loss": 0.0877,
+      "learning_rate": 0.0001863411078717201,
+      "loss": 0.0529,
       "step": 13500
     },
     {
       "epoch": 19.4,
-      "eval_loss": 0.16647975146770477,
-      "eval_runtime": 39.9295,
-      "eval_samples_per_second": 11.445,
-      "eval_steps_per_second": 1.453,
-      "eval_wer": 0.1689504592681825,
+      "eval_loss": 0.16998930275440216,
+      "eval_runtime": 40.128,
+      "eval_samples_per_second": 11.389,
+      "eval_steps_per_second": 1.445,
+      "eval_wer": 0.15268784821563017,
       "step": 13500
     },
     {
       "epoch": 19.41,
-      "learning_rate": 0.00010856231599607458,
-      "loss": 0.0905,
+      "learning_rate": 0.00018625364431486879,
+      "loss": 0.0572,
       "step": 13510
     },
     {
       "epoch": 19.43,
-      "learning_rate": 0.0001084151128557409,
-      "loss": 0.0656,
+      "learning_rate": 0.0001861661807580175,
+      "loss": 0.0376,
       "step": 13520
     },
     {
       "epoch": 19.44,
-      "learning_rate": 0.00010826790971540726,
-      "loss": 0.0834,
+      "learning_rate": 0.00018607871720116616,
+      "loss": 0.0473,
       "step": 13530
     },
     {
       "epoch": 19.45,
-      "learning_rate": 0.00010812070657507358,
-      "loss": 0.0769,
+      "learning_rate": 0.00018599125364431485,
+      "loss": 0.0473,
       "step": 13540
     },
     {
       "epoch": 19.47,
-      "learning_rate": 0.00010797350343473995,
-      "loss": 0.0981,
+      "learning_rate": 0.00018590379008746354,
+      "loss": 0.0561,
       "step": 13550
     },
     {
       "epoch": 19.48,
-      "learning_rate": 0.00010782630029440627,
-      "loss": 0.1006,
+      "learning_rate": 0.00018581632653061223,
+      "loss": 0.0615,
       "step": 13560
     },
     {
       "epoch": 19.5,
-      "learning_rate": 0.00010767909715407263,
-      "loss": 0.0693,
+      "learning_rate": 0.0001857288629737609,
+      "loss": 0.0379,
       "step": 13570
     },
     {
       "epoch": 19.51,
-      "learning_rate": 0.00010753189401373895,
-      "loss": 0.0747,
+      "learning_rate": 0.0001856413994169096,
+      "loss": 0.0445,
       "step": 13580
     },
     {
       "epoch": 19.53,
-      "learning_rate": 0.00010738469087340528,
-      "loss": 0.077,
+      "learning_rate": 0.0001855539358600583,
+      "loss": 0.0518,
       "step": 13590
     },
     {
       "epoch": 19.54,
-      "learning_rate": 0.00010723748773307163,
-      "loss": 0.0735,
+      "learning_rate": 0.000185466472303207,
+      "loss": 0.0432,
       "step": 13600
     },
     {
       "epoch": 19.54,
-      "eval_loss": 0.16026227176189423,
-      "eval_runtime": 39.8849,
-      "eval_samples_per_second": 11.458,
-      "eval_steps_per_second": 1.454,
-      "eval_wer": 0.1675952416804698,
+      "eval_loss": 0.15767106413841248,
+      "eval_runtime": 39.8653,
+      "eval_samples_per_second": 11.464,
+      "eval_steps_per_second": 1.455,
+      "eval_wer": 0.14696581840084325,
       "step": 13600
     },
     {
       "epoch": 19.55,
-      "learning_rate": 0.00010709028459273796,
-      "loss": 0.1046,
+      "learning_rate": 0.0001853790087463557,
+      "loss": 0.0629,
       "step": 13610
     },
     {
       "epoch": 19.57,
-      "learning_rate": 0.0001069430814524043,
-      "loss": 0.0685,
+      "learning_rate": 0.00018529154518950435,
+      "loss": 0.0393,
       "step": 13620
     },
     {
       "epoch": 19.58,
-      "learning_rate": 0.00010679587831207064,
-      "loss": 0.0856,
+      "learning_rate": 0.00018520408163265304,
+      "loss": 0.0519,
       "step": 13630
     },
     {
       "epoch": 19.6,
-      "learning_rate": 0.00010664867517173699,
-      "loss": 0.0621,
+      "learning_rate": 0.00018511661807580173,
+      "loss": 0.0367,
       "step": 13640
     },
     {
       "epoch": 19.61,
-      "learning_rate": 0.00010650147203140332,
-      "loss": 0.0743,
+      "learning_rate": 0.00018502915451895042,
+      "loss": 0.0431,
       "step": 13650
     },
     {
       "epoch": 19.63,
-      "learning_rate": 0.00010635426889106967,
-      "loss": 0.0899,
+      "learning_rate": 0.0001849416909620991,
+      "loss": 0.0576,
       "step": 13660
     },
     {
       "epoch": 19.64,
-      "learning_rate": 0.000106207065750736,
-      "loss": 0.0761,
+      "learning_rate": 0.0001848542274052478,
+      "loss": 0.0461,
       "step": 13670
     },
     {
       "epoch": 19.66,
-      "learning_rate": 0.00010605986261040235,
-      "loss": 0.0887,
+      "learning_rate": 0.00018476676384839648,
+      "loss": 0.045,
       "step": 13680
     },
     {
       "epoch": 19.67,
-      "learning_rate": 0.00010591265947006868,
-      "loss": 0.082,
+      "learning_rate": 0.0001846793002915452,
+      "loss": 0.0508,
       "step": 13690
     },
     {
       "epoch": 19.68,
-      "learning_rate": 0.00010576545632973503,
-      "loss": 0.0804,
+      "learning_rate": 0.00018459183673469388,
+      "loss": 0.0459,
       "step": 13700
     },
     {
       "epoch": 19.68,
-      "eval_loss": 0.16200566291809082,
-      "eval_runtime": 40.0981,
-      "eval_samples_per_second": 11.397,
+      "eval_loss": 0.158447265625,
+      "eval_runtime": 40.1118,
+      "eval_samples_per_second": 11.393,
       "eval_steps_per_second": 1.446,
-      "eval_wer": 0.17060683631983134,
+      "eval_wer": 0.1525372684836621,
       "step": 13700
     },
     {
       "epoch": 19.7,
-      "learning_rate": 0.00010561825318940136,
-      "loss": 0.08,
+      "learning_rate": 0.00018450437317784254,
+      "loss": 0.057,
       "step": 13710
     },
     {
       "epoch": 19.71,
-      "learning_rate": 0.00010547105004906771,
-      "loss": 0.0683,
+      "learning_rate": 0.00018441690962099123,
+      "loss": 0.0471,
       "step": 13720
     },
     {
       "epoch": 19.73,
-      "learning_rate": 0.00010532384690873404,
-      "loss": 0.0769,
+      "learning_rate": 0.00018432944606413992,
+      "loss": 0.0489,
       "step": 13730
     },
     {
       "epoch": 19.74,
-      "learning_rate": 0.00010517664376840037,
-      "loss": 0.0768,
+      "learning_rate": 0.0001842419825072886,
+      "loss": 0.0507,
       "step": 13740
     },
     {
       "epoch": 19.76,
-      "learning_rate": 0.00010502944062806672,
-      "loss": 0.0666,
+      "learning_rate": 0.0001841545189504373,
+      "loss": 0.0386,
       "step": 13750
     },
     {
       "epoch": 19.77,
-      "learning_rate": 0.00010488223748773305,
-      "loss": 0.0676,
+      "learning_rate": 0.00018406705539358598,
+      "loss": 0.0393,
       "step": 13760
     },
     {
       "epoch": 19.78,
-      "learning_rate": 0.0001047350343473994,
-      "loss": 0.0745,
+      "learning_rate": 0.0001839795918367347,
+      "loss": 0.0411,
       "step": 13770
     },
     {
       "epoch": 19.8,
-      "learning_rate": 0.00010458783120706574,
-      "loss": 0.0802,
+      "learning_rate": 0.00018389212827988339,
+      "loss": 0.056,
       "step": 13780
     },
     {
       "epoch": 19.81,
-      "learning_rate": 0.00010444062806673208,
-      "loss": 0.0733,
+      "learning_rate": 0.00018380466472303207,
+      "loss": 0.0394,
       "step": 13790
     },
     {
       "epoch": 19.83,
-      "learning_rate": 0.00010429342492639842,
-      "loss": 0.0871,
+      "learning_rate": 0.00018371720116618074,
+      "loss": 0.0529,
       "step": 13800
     },
     {
       "epoch": 19.83,
-      "eval_loss": 0.1567193865776062,
-      "eval_runtime": 39.8991,
-      "eval_samples_per_second": 11.454,
+      "eval_loss": 0.1573006957769394,
+      "eval_runtime": 39.8821,
+      "eval_samples_per_second": 11.459,
       "eval_steps_per_second": 1.454,
-      "eval_wer": 0.1644330673091402,
+      "eval_wer": 0.14967625357626863,
       "step": 13800
     },
     {
       "epoch": 19.84,
-      "learning_rate": 0.00010414622178606476,
-      "loss": 0.0767,
+      "learning_rate": 0.00018362973760932942,
+      "loss": 0.0512,
       "step": 13810
     },
     {
       "epoch": 19.86,
-      "learning_rate": 0.0001039990186457311,
-      "loss": 0.0782,
+      "learning_rate": 0.0001835422740524781,
+      "loss": 0.05,
       "step": 13820
     },
     {
       "epoch": 19.87,
-      "learning_rate": 0.00010385181550539744,
-      "loss": 0.077,
+      "learning_rate": 0.0001834548104956268,
+      "loss": 0.0485,
       "step": 13830
     },
     {
       "epoch": 19.89,
-      "learning_rate": 0.00010370461236506378,
-      "loss": 0.0592,
+      "learning_rate": 0.0001833673469387755,
+      "loss": 0.0368,
       "step": 13840
     },
     {
       "epoch": 19.9,
-      "learning_rate": 0.00010355740922473012,
-      "loss": 0.0768,
+      "learning_rate": 0.00018327988338192417,
+      "loss": 0.0512,
       "step": 13850
     },
     {
       "epoch": 19.91,
-      "learning_rate": 0.00010341020608439646,
-      "loss": 0.0797,
+      "learning_rate": 0.0001831924198250729,
+      "loss": 0.05,
       "step": 13860
     },
     {
       "epoch": 19.93,
-      "learning_rate": 0.0001032630029440628,
-      "loss": 0.0721,
+      "learning_rate": 0.00018310495626822158,
+      "loss": 0.0443,
       "step": 13870
     },
     {
       "epoch": 19.94,
-      "learning_rate": 0.00010311579980372914,
-      "loss": 0.0699,
+      "learning_rate": 0.00018301749271137027,
+      "loss": 0.0381,
       "step": 13880
     },
     {
       "epoch": 19.96,
-      "learning_rate": 0.00010296859666339547,
-      "loss": 0.0656,
+      "learning_rate": 0.00018293002915451893,
+      "loss": 0.0396,
       "step": 13890
     },
     {
       "epoch": 19.97,
-      "learning_rate": 0.00010282139352306182,
-      "loss": 0.0822,
+      "learning_rate": 0.00018284256559766761,
+      "loss": 0.0467,
       "step": 13900
     },
     {
       "epoch": 19.97,
-      "eval_loss": 0.1556914746761322,
-      "eval_runtime": 40.153,
-      "eval_samples_per_second": 11.381,
-      "eval_steps_per_second": 1.444,
-      "eval_wer": 0.16578828489685288,
+      "eval_loss": 0.1563096046447754,
+      "eval_runtime": 39.8443,
+      "eval_samples_per_second": 11.47,
+      "eval_steps_per_second": 1.456,
+      "eval_wer": 0.15208552928775787,
       "step": 13900
     },
     {
       "epoch": 19.99,
-      "learning_rate": 0.00010267419038272815,
-      "loss": 0.0944,
+      "learning_rate": 0.0001827551020408163,
+      "loss": 0.0617,
       "step": 13910
     },
     {
       "epoch": 20.0,
-      "learning_rate": 0.0001025269872423945,
-      "loss": 0.0612,
+      "learning_rate": 0.000182667638483965,
+      "loss": 0.0334,
       "step": 13920
     },
     {
       "epoch": 20.01,
-      "learning_rate": 0.00010237978410206083,
-      "loss": 0.082,
+      "learning_rate": 0.00018258017492711368,
+      "loss": 0.0511,
       "step": 13930
     },
     {
       "epoch": 20.03,
-      "learning_rate": 0.00010224730127576054,
-      "loss": 0.0691,
+      "learning_rate": 0.00018249271137026237,
+      "loss": 0.0419,
       "step": 13940
     },
     {
       "epoch": 20.04,
-      "learning_rate": 0.00010210009813542687,
-      "loss": 0.068,
+      "learning_rate": 0.00018240524781341108,
+      "loss": 0.0376,
       "step": 13950
     },
     {
       "epoch": 20.06,
-      "learning_rate": 0.00010195289499509322,
-      "loss": 0.0663,
+      "learning_rate": 0.00018231778425655977,
+      "loss": 0.0431,
       "step": 13960
     },
     {
       "epoch": 20.07,
-      "learning_rate": 0.00010180569185475955,
-      "loss": 0.0654,
+      "learning_rate": 0.00018223032069970846,
+      "loss": 0.0356,
       "step": 13970
     },
     {
       "epoch": 20.09,
-      "learning_rate": 0.0001016584887144259,
-      "loss": 0.091,
+      "learning_rate": 0.00018214285714285712,
+      "loss": 0.0566,
       "step": 13980
     },
     {
       "epoch": 20.1,
-      "learning_rate": 0.00010151128557409223,
-      "loss": 0.075,
+      "learning_rate": 0.0001820553935860058,
+      "loss": 0.0478,
       "step": 13990
     },
     {
       "epoch": 20.11,
-      "learning_rate": 0.00010136408243375858,
-      "loss": 0.0784,
+      "learning_rate": 0.0001819679300291545,
+      "loss": 0.0516,
       "step": 14000
     },
     {
       "epoch": 20.11,
-      "eval_loss": 0.16241466999053955,
-      "eval_runtime": 39.9865,
-      "eval_samples_per_second": 11.429,
+      "eval_loss": 0.15548963844776154,
+      "eval_runtime": 40.0025,
+      "eval_samples_per_second": 11.424,
       "eval_steps_per_second": 1.45,
-      "eval_wer": 0.16187321186568288,
+      "eval_wer": 0.15268784821563017,
       "step": 14000
     },
     {
       "epoch": 20.13,
-      "learning_rate": 0.00010121687929342491,
-      "loss": 0.0876,
+      "learning_rate": 0.00018188046647230318,
+      "loss": 0.0554,
       "step": 14010
     },
     {
       "epoch": 20.14,
-      "learning_rate": 0.00010106967615309126,
-      "loss": 0.0809,
+      "learning_rate": 0.00018179300291545187,
+      "loss": 0.0498,
       "step": 14020
     },
     {
       "epoch": 20.16,
-      "learning_rate": 0.00010092247301275759,
-      "loss": 0.0855,
+      "learning_rate": 0.00018170553935860058,
+      "loss": 0.0542,
       "step": 14030
     },
     {
       "epoch": 20.17,
-      "learning_rate": 0.00010077526987242394,
-      "loss": 0.0702,
+      "learning_rate": 0.00018161807580174927,
+      "loss": 0.043,
       "step": 14040
     },
     {
       "epoch": 20.19,
-      "learning_rate": 0.00010062806673209027,
-      "loss": 0.0846,
+      "learning_rate": 0.00018153061224489796,
+      "loss": 0.0511,
       "step": 14050
     },
     {
       "epoch": 20.2,
-      "learning_rate": 0.00010048086359175662,
-      "loss": 0.0724,
+      "learning_rate": 0.00018144314868804665,
+      "loss": 0.0495,
       "step": 14060
     },
     {
       "epoch": 20.22,
-      "learning_rate": 0.00010033366045142295,
-      "loss": 0.0664,
+      "learning_rate": 0.0001813556851311953,
+      "loss": 0.0342,
       "step": 14070
     },
     {
       "epoch": 20.23,
-      "learning_rate": 0.0001001864573110893,
-      "loss": 0.0825,
+      "learning_rate": 0.000181268221574344,
+      "loss": 0.056,
       "step": 14080
     },
     {
       "epoch": 20.24,
-      "learning_rate": 0.00010003925417075563,
-      "loss": 0.0651,
+      "learning_rate": 0.00018118075801749268,
+      "loss": 0.0438,
       "step": 14090
     },
     {
       "epoch": 20.26,
-      "learning_rate": 9.989205103042197e-05,
-      "loss": 0.0822,
+      "learning_rate": 0.00018109329446064137,
+      "loss": 0.0506,
       "step": 14100
     },
     {
       "epoch": 20.26,
-      "eval_loss": 0.16037575900554657,
-      "eval_runtime": 39.914,
-      "eval_samples_per_second": 11.45,
+      "eval_loss": 0.15269336104393005,
+      "eval_runtime": 39.9095,
+      "eval_samples_per_second": 11.451,
       "eval_steps_per_second": 1.453,
-      "eval_wer": 0.1617226321337148,
+      "eval_wer": 0.14681523866887516,
       "step": 14100
     },
     {
       "epoch": 20.27,
-      "learning_rate": 9.974484789008831e-05,
-      "loss": 0.0743,
+      "learning_rate": 0.00018100583090379006,
+      "loss": 0.0443,
       "step": 14110
     },
     {
       "epoch": 20.29,
-      "learning_rate": 9.959764474975465e-05,
-      "loss": 0.0709,
+      "learning_rate": 0.00018091836734693877,
+      "loss": 0.04,
       "step": 14120
     },
     {
       "epoch": 20.3,
-      "learning_rate": 9.945044160942099e-05,
-      "loss": 0.068,
+      "learning_rate": 0.00018083090379008746,
+      "loss": 0.046,
       "step": 14130
     },
     {
       "epoch": 20.32,
-      "learning_rate": 9.930323846908733e-05,
-      "loss": 0.067,
+      "learning_rate": 0.00018074344023323615,
+      "loss": 0.0455,
       "step": 14140
     },
     {
       "epoch": 20.33,
-      "learning_rate": 9.915603532875367e-05,
-      "loss": 0.0749,
+      "learning_rate": 0.00018065597667638484,
+      "loss": 0.0471,
       "step": 14150
     },
     {
       "epoch": 20.34,
-      "learning_rate": 9.900883218842e-05,
-      "loss": 0.0676,
+      "learning_rate": 0.0001805685131195335,
+      "loss": 0.0436,
       "step": 14160
     },
     {
       "epoch": 20.36,
-      "learning_rate": 9.886162904808635e-05,
-      "loss": 0.0652,
+      "learning_rate": 0.0001804810495626822,
+      "loss": 0.0417,
       "step": 14170
     },
     {
       "epoch": 20.37,
-      "learning_rate": 9.871442590775269e-05,
-      "loss": 0.0841,
+      "learning_rate": 0.00018039358600583088,
+      "loss": 0.0517,
       "step": 14180
     },
     {
       "epoch": 20.39,
-      "learning_rate": 9.856722276741903e-05,
-      "loss": 0.0779,
+      "learning_rate": 0.00018030612244897956,
+      "loss": 0.0492,
       "step": 14190
     },
     {
       "epoch": 20.4,
-      "learning_rate": 9.842001962708537e-05,
-      "loss": 0.0694,
+      "learning_rate": 0.00018021865889212828,
+      "loss": 0.0445,
       "step": 14200
     },
     {
       "epoch": 20.4,
-      "eval_loss": 0.14927269518375397,
-      "eval_runtime": 40.1956,
-      "eval_samples_per_second": 11.369,
-      "eval_steps_per_second": 1.443,
-      "eval_wer": 0.16142147266977866,
+      "eval_loss": 0.15262174606323242,
+      "eval_runtime": 40.2286,
+      "eval_samples_per_second": 11.36,
+      "eval_steps_per_second": 1.442,
+      "eval_wer": 0.15027857250414095,
       "step": 14200
     },
     {
       "epoch": 20.42,
-      "learning_rate": 9.827281648675171e-05,
-      "loss": 0.0796,
+      "learning_rate": 0.00018013119533527697,
+      "loss": 0.0505,
       "step": 14210
     },
     {
       "epoch": 20.43,
-      "learning_rate": 9.812561334641805e-05,
-      "loss": 0.0779,
+      "learning_rate": 0.00018004373177842565,
+      "loss": 0.0436,
       "step": 14220
     },
     {
       "epoch": 20.45,
-      "learning_rate": 9.79784102060844e-05,
-      "loss": 0.0788,
+      "learning_rate": 0.00017995626822157434,
+      "loss": 0.046,
       "step": 14230
     },
     {
       "epoch": 20.46,
-      "learning_rate": 9.783120706575073e-05,
-      "loss": 0.0567,
+      "learning_rate": 0.00017986880466472303,
+      "loss": 0.0341,
       "step": 14240
     },
     {
       "epoch": 20.47,
-      "learning_rate": 9.768400392541706e-05,
-      "loss": 0.0715,
+      "learning_rate": 0.0001797813411078717,
+      "loss": 0.0454,
       "step": 14250
     },
     {
       "epoch": 20.49,
-      "learning_rate": 9.753680078508341e-05,
-      "loss": 0.0744,
+      "learning_rate": 0.00017969387755102038,
+      "loss": 0.046,
       "step": 14260
     },
     {
       "epoch": 20.5,
-      "learning_rate": 9.738959764474974e-05,
-      "loss": 0.0641,
+      "learning_rate": 0.00017960641399416907,
+      "loss": 0.0421,
       "step": 14270
     },
     {
       "epoch": 20.52,
-      "learning_rate": 9.724239450441609e-05,
-      "loss": 0.0776,
+      "learning_rate": 0.00017951895043731775,
+      "loss": 0.048,
       "step": 14280
     },
     {
       "epoch": 20.53,
-      "learning_rate": 9.709519136408242e-05,
-      "loss": 0.0637,
+      "learning_rate": 0.00017943148688046647,
+      "loss": 0.0355,
       "step": 14290
     },
     {
       "epoch": 20.55,
-      "learning_rate": 9.694798822374877e-05,
-      "loss": 0.0775,
+      "learning_rate": 0.00017934402332361516,
+      "loss": 0.0467,
       "step": 14300
     },
     {
       "epoch": 20.55,
-      "eval_loss": 0.1541454941034317,
-      "eval_runtime": 39.9553,
-      "eval_samples_per_second": 11.438,
-      "eval_steps_per_second": 1.452,
-      "eval_wer": 0.16081915374190633,
+      "eval_loss": 0.14778903126716614,
+      "eval_runtime": 40.0483,
+      "eval_samples_per_second": 11.411,
+      "eval_steps_per_second": 1.448,
+      "eval_wer": 0.14907393464839633,
       "step": 14300
     },
     {
       "epoch": 20.56,
-      "learning_rate": 9.68007850834151e-05,
-      "loss": 0.0805,
+      "learning_rate": 0.00017925655976676384,
+      "loss": 0.0495,
       "step": 14310
     },
     {
       "epoch": 20.57,
-      "learning_rate": 9.665358194308145e-05,
-      "loss": 0.0722,
+      "learning_rate": 0.00017916909620991253,
+      "loss": 0.0415,
       "step": 14320
     },
     {
       "epoch": 20.59,
-      "learning_rate": 9.650637880274778e-05,
-      "loss": 0.0682,
+      "learning_rate": 0.00017908163265306122,
+      "loss": 0.0429,
       "step": 14330
     },
     {
       "epoch": 20.6,
-      "learning_rate": 9.635917566241413e-05,
-      "loss": 0.0724,
+      "learning_rate": 0.00017899416909620988,
+      "loss": 0.0447,
       "step": 14340
     },
     {
       "epoch": 20.62,
-      "learning_rate": 9.621197252208046e-05,
-      "loss": 0.0834,
+      "learning_rate": 0.00017890670553935857,
+      "loss": 0.0496,
       "step": 14350
     },
     {
       "epoch": 20.63,
-      "learning_rate": 9.606476938174681e-05,
-      "loss": 0.0795,
+      "learning_rate": 0.00017881924198250726,
+      "loss": 0.0504,
       "step": 14360
     },
     {
       "epoch": 20.65,
-      "learning_rate": 9.591756624141314e-05,
-      "loss": 0.0874,
+      "learning_rate": 0.00017873177842565595,
+      "loss": 0.053,
       "step": 14370
     },
     {
       "epoch": 20.66,
-      "learning_rate": 9.577036310107949e-05,
-      "loss": 0.0689,
+      "learning_rate": 0.00017864431486880466,
+      "loss": 0.0482,
       "step": 14380
     },
     {
       "epoch": 20.68,
-      "learning_rate": 9.562315996074582e-05,
-      "loss": 0.0669,
+      "learning_rate": 0.00017855685131195335,
+      "loss": 0.0393,
       "step": 14390
     },
     {
       "epoch": 20.69,
-      "learning_rate": 9.547595682041216e-05,
-      "loss": 0.0747,
+      "learning_rate": 0.00017846938775510204,
+      "loss": 0.0475,
       "step": 14400
     },
     {
       "epoch": 20.69,
-      "eval_loss": 0.15958380699157715,
-      "eval_runtime": 40.1805,
-      "eval_samples_per_second": 11.374,
-      "eval_steps_per_second": 1.443,
-      "eval_wer": 0.1624755307935552,
+      "eval_loss": 0.1577850878238678,
+      "eval_runtime": 40.5152,
+      "eval_samples_per_second": 11.28,
+      "eval_steps_per_second": 1.432,
+      "eval_wer": 0.15057973196807709,
       "step": 14400
     },
     {
       "epoch": 20.7,
-      "learning_rate": 9.53287536800785e-05,
-      "loss": 0.0737,
+      "learning_rate": 0.00017838192419825072,
+      "loss": 0.0475,
       "step": 14410
     },
     {
       "epoch": 20.72,
-      "learning_rate": 9.518155053974484e-05,
-      "loss": 0.0717,
+      "learning_rate": 0.0001782944606413994,
+      "loss": 0.04,
       "step": 14420
     },
     {
       "epoch": 20.73,
-      "learning_rate": 9.503434739941118e-05,
-      "loss": 0.0923,
+      "learning_rate": 0.00017820699708454807,
+      "loss": 0.0525,
       "step": 14430
     },
     {
       "epoch": 20.75,
-      "learning_rate": 9.488714425907752e-05,
-      "loss": 0.0623,
+      "learning_rate": 0.00017811953352769676,
+      "loss": 0.0412,
       "step": 14440
     },
     {
       "epoch": 20.76,
-      "learning_rate": 9.473994111874386e-05,
-      "loss": 0.0658,
+      "learning_rate": 0.00017803206997084545,
+      "loss": 0.0376,
       "step": 14450
     },
     {
       "epoch": 20.78,
-      "learning_rate": 9.45927379784102e-05,
-      "loss": 0.0658,
+      "learning_rate": 0.00017794460641399416,
+      "loss": 0.0419,
       "step": 14460
     },
     {
       "epoch": 20.79,
-      "learning_rate": 9.444553483807654e-05,
-      "loss": 0.068,
+      "learning_rate": 0.00017785714285714285,
+      "loss": 0.0365,
       "step": 14470
     },
     {
       "epoch": 20.8,
-      "learning_rate": 9.429833169774288e-05,
-      "loss": 0.0697,
+      "learning_rate": 0.00017776967930029154,
+      "loss": 0.0439,
       "step": 14480
     },
     {
       "epoch": 20.82,
-      "learning_rate": 9.415112855740923e-05,
-      "loss": 0.0618,
+      "learning_rate": 0.00017768221574344023,
+      "loss": 0.038,
       "step": 14490
     },
     {
       "epoch": 20.83,
-      "learning_rate": 9.400392541707556e-05,
-      "loss": 0.0701,
+      "learning_rate": 0.00017759475218658891,
+      "loss": 0.0416,
       "step": 14500
     },
     {
       "epoch": 20.83,
-      "eval_loss": 0.15933384001255035,
-      "eval_runtime": 40.0625,
-      "eval_samples_per_second": 11.407,
-      "eval_steps_per_second": 1.448,
-      "eval_wer": 0.16081915374190633,
+      "eval_loss": 0.14831580221652985,
+      "eval_runtime": 39.9501,
+      "eval_samples_per_second": 11.439,
+      "eval_steps_per_second": 1.452,
+      "eval_wer": 0.15148321035988557,
       "step": 14500
     },
     {
       "epoch": 20.85,
-      "learning_rate": 9.38567222767419e-05,
-      "loss": 0.0737,
+      "learning_rate": 0.0001775072886297376,
+      "loss": 0.0488,
       "step": 14510
     },
     {
       "epoch": 20.86,
-      "learning_rate": 9.370951913640824e-05,
-      "loss": 0.0772,
+      "learning_rate": 0.00017741982507288626,
+      "loss": 0.0511,
       "step": 14520
     },
     {
       "epoch": 20.88,
-      "learning_rate": 9.356231599607459e-05,
-      "loss": 0.0846,
+      "learning_rate": 0.00017733236151603495,
+      "loss": 0.0605,
       "step": 14530
     },
     {
       "epoch": 20.89,
-      "learning_rate": 9.341511285574092e-05,
-      "loss": 0.0514,
+      "learning_rate": 0.00017724489795918364,
+      "loss": 0.027,
       "step": 14540
     },
     {
       "epoch": 20.91,
-      "learning_rate": 9.326790971540725e-05,
-      "loss": 0.0853,
+      "learning_rate": 0.00017715743440233235,
+      "loss": 0.0499,
       "step": 14550
     },
     {
       "epoch": 20.92,
-      "learning_rate": 9.31207065750736e-05,
-      "loss": 0.0715,
+      "learning_rate": 0.00017706997084548104,
+      "loss": 0.0462,
       "step": 14560
     },
     {
       "epoch": 20.93,
-      "learning_rate": 9.297350343473993e-05,
-      "loss": 0.0635,
+      "learning_rate": 0.00017698250728862973,
+      "loss": 0.0376,
       "step": 14570
     },
     {
       "epoch": 20.95,
-      "learning_rate": 9.282630029440628e-05,
-      "loss": 0.0978,
+      "learning_rate": 0.00017689504373177842,
+      "loss": 0.0632,
       "step": 14580
     },
     {
       "epoch": 20.96,
-      "learning_rate": 9.267909715407261e-05,
-      "loss": 0.0803,
+      "learning_rate": 0.0001768075801749271,
+      "loss": 0.0459,
       "step": 14590
     },
     {
       "epoch": 20.98,
-      "learning_rate": 9.253189401373896e-05,
-      "loss": 0.0698,
+      "learning_rate": 0.0001767201166180758,
+      "loss": 0.0467,
       "step": 14600
     },
     {
       "epoch": 20.98,
-      "eval_loss": 0.15506555140018463,
-      "eval_runtime": 40.1892,
-      "eval_samples_per_second": 11.371,
-      "eval_steps_per_second": 1.443,
-      "eval_wer": 0.16081915374190633,
+      "eval_loss": 0.14795160293579102,
+      "eval_runtime": 40.0839,
+      "eval_samples_per_second": 11.401,
+      "eval_steps_per_second": 1.447,
+      "eval_wer": 0.1519349495557898,
       "step": 14600
     },
     {
       "epoch": 20.99,
-      "learning_rate": 9.23846908734053e-05,
-      "loss": 0.0791,
+      "learning_rate": 0.00017663265306122445,
+      "loss": 0.0478,
       "step": 14610
     },
     {
       "epoch": 21.01,
-      "learning_rate": 9.223748773307164e-05,
-      "loss": 0.0568,
+      "learning_rate": 0.00017654518950437314,
+      "loss": 0.0346,
       "step": 14620
     },
     {
       "epoch": 21.02,
-      "learning_rate": 9.209028459273797e-05,
-      "loss": 0.0696,
+      "learning_rate": 0.00017645772594752186,
+      "loss": 0.0437,
       "step": 14630
     },
     {
       "epoch": 21.03,
-      "learning_rate": 9.194308145240432e-05,
-      "loss": 0.082,
+      "learning_rate": 0.00017637026239067055,
+      "loss": 0.0533,
       "step": 14640
     },
     {
       "epoch": 21.05,
-      "learning_rate": 9.179587831207065e-05,
-      "loss": 0.0648,
+      "learning_rate": 0.00017628279883381923,
+      "loss": 0.0414,
       "step": 14650
     },
     {
       "epoch": 21.06,
-      "learning_rate": 9.1648675171737e-05,
-      "loss": 0.056,
+      "learning_rate": 0.00017619533527696792,
+      "loss": 0.033,
       "step": 14660
     },
     {
       "epoch": 21.08,
-      "learning_rate": 9.150147203140332e-05,
-      "loss": 0.0694,
+      "learning_rate": 0.0001761078717201166,
+      "loss": 0.0384,
       "step": 14670
     },
     {
       "epoch": 21.09,
-      "learning_rate": 9.135426889106968e-05,
-      "loss": 0.0662,
+      "learning_rate": 0.0001760204081632653,
+      "loss": 0.0456,
       "step": 14680
     },
     {
       "epoch": 21.11,
-      "learning_rate": 9.1207065750736e-05,
-      "loss": 0.0671,
+      "learning_rate": 0.00017593294460641398,
+      "loss": 0.0376,
       "step": 14690
     },
     {
       "epoch": 21.12,
-      "learning_rate": 9.105986261040233e-05,
-      "loss": 0.0785,
+      "learning_rate": 0.00017584548104956265,
+      "loss": 0.0527,
       "step": 14700
     },
     {
       "epoch": 21.12,
-      "eval_loss": 0.15715524554252625,
-      "eval_runtime": 40.3412,
-      "eval_samples_per_second": 11.328,
-      "eval_steps_per_second": 1.438,
-      "eval_wer": 0.1636801686492998,
+      "eval_loss": 0.152811199426651,
+      "eval_runtime": 39.8549,
+      "eval_samples_per_second": 11.467,
+      "eval_steps_per_second": 1.455,
+      "eval_wer": 0.14907393464839633,
       "step": 14700
     },
     {
       "epoch": 21.14,
-      "learning_rate": 9.091265947006868e-05,
-      "loss": 0.0661,
+      "learning_rate": 0.00017575801749271133,
+      "loss": 0.0395,
       "step": 14710
     },
     {
       "epoch": 21.15,
-      "learning_rate": 9.076545632973502e-05,
-      "loss": 0.0702,
+      "learning_rate": 0.00017567055393586005,
+      "loss": 0.0435,
       "step": 14720
     },
     {
       "epoch": 21.16,
-      "learning_rate": 9.061825318940136e-05,
-      "loss": 0.0671,
+      "learning_rate": 0.00017558309037900874,
+      "loss": 0.0414,
       "step": 14730
     },
     {
       "epoch": 21.18,
-      "learning_rate": 9.04710500490677e-05,
-      "loss": 0.0707,
+      "learning_rate": 0.00017549562682215742,
+      "loss": 0.0364,
       "step": 14740
     },
     {
       "epoch": 21.19,
-      "learning_rate": 9.032384690873404e-05,
-      "loss": 0.0559,
+      "learning_rate": 0.0001754081632653061,
+      "loss": 0.0349,
       "step": 14750
     },
     {
       "epoch": 21.21,
-      "learning_rate": 9.017664376840038e-05,
-      "loss": 0.068,
+      "learning_rate": 0.0001753206997084548,
+      "loss": 0.0436,
       "step": 14760
     },
     {
       "epoch": 21.22,
-      "learning_rate": 9.002944062806672e-05,
-      "loss": 0.0817,
+      "learning_rate": 0.0001752332361516035,
+      "loss": 0.0513,
       "step": 14770
     },
     {
       "epoch": 21.24,
-      "learning_rate": 8.988223748773306e-05,
-      "loss": 0.0792,
+      "learning_rate": 0.00017514577259475218,
+      "loss": 0.0489,
       "step": 14780
     },
     {
       "epoch": 21.25,
-      "learning_rate": 8.97350343473994e-05,
-      "loss": 0.054,
+      "learning_rate": 0.00017505830903790084,
+      "loss": 0.0302,
       "step": 14790
     },
     {
       "epoch": 21.26,
-      "learning_rate": 8.958783120706574e-05,
-      "loss": 0.0975,
+      "learning_rate": 0.00017497084548104952,
+      "loss": 0.0625,
       "step": 14800
     },
     {
       "epoch": 21.26,
-      "eval_loss": 0.16008907556533813,
-      "eval_runtime": 40.1457,
-      "eval_samples_per_second": 11.384,
-      "eval_steps_per_second": 1.445,
-      "eval_wer": 0.16322842945339558,
+      "eval_loss": 0.1483238786458969,
+      "eval_runtime": 40.1794,
+      "eval_samples_per_second": 11.374,
+      "eval_steps_per_second": 1.444,
+      "eval_wer": 0.14907393464839633,
       "step": 14800
     },
     {
       "epoch": 21.28,
-      "learning_rate": 8.944062806673208e-05,
-      "loss": 0.0653,
+      "learning_rate": 0.00017488338192419824,
+      "loss": 0.0435,
       "step": 14810
     },
     {
       "epoch": 21.29,
-      "learning_rate": 8.929342492639842e-05,
-      "loss": 0.0675,
+      "learning_rate": 0.00017479591836734693,
+      "loss": 0.0369,
       "step": 14820
     },
     {
       "epoch": 21.31,
-      "learning_rate": 8.914622178606476e-05,
-      "loss": 0.0676,
+      "learning_rate": 0.00017470845481049562,
+      "loss": 0.0447,
       "step": 14830
     },
     {
       "epoch": 21.32,
-      "learning_rate": 8.89990186457311e-05,
-      "loss": 0.0631,
+      "learning_rate": 0.0001746209912536443,
+      "loss": 0.0332,
       "step": 14840
     },
     {
       "epoch": 21.34,
-      "learning_rate": 8.885181550539743e-05,
-      "loss": 0.0625,
+      "learning_rate": 0.000174533527696793,
+      "loss": 0.0375,
       "step": 14850
     },
     {
       "epoch": 21.35,
-      "learning_rate": 8.870461236506378e-05,
-      "loss": 0.071,
+      "learning_rate": 0.00017444606413994168,
+      "loss": 0.0468,
       "step": 14860
     },
     {
       "epoch": 21.36,
-      "learning_rate": 8.855740922473011e-05,
-      "loss": 0.0684,
+      "learning_rate": 0.00017435860058309037,
+      "loss": 0.0434,
       "step": 14870
     },
     {
       "epoch": 21.38,
-      "learning_rate": 8.841020608439646e-05,
-      "loss": 0.0656,
+      "learning_rate": 0.00017427113702623903,
+      "loss": 0.047,
       "step": 14880
     },
     {
       "epoch": 21.39,
-      "learning_rate": 8.826300294406279e-05,
-      "loss": 0.0609,
+      "learning_rate": 0.00017418367346938774,
+      "loss": 0.0402,
       "step": 14890
     },
     {
       "epoch": 21.41,
-      "learning_rate": 8.811579980372914e-05,
-      "loss": 0.0697,
+      "learning_rate": 0.00017409620991253643,
+      "loss": 0.043,
       "step": 14900
     },
     {
       "epoch": 21.41,
-      "eval_loss": 0.15438880026340485,
-      "eval_runtime": 40.0849,
-      "eval_samples_per_second": 11.401,
+      "eval_loss": 0.15278911590576172,
+      "eval_runtime": 40.0861,
+      "eval_samples_per_second": 11.4,
       "eval_steps_per_second": 1.447,
-      "eval_wer": 0.15931335642222558,
+      "eval_wer": 0.14862219545249208,
       "step": 14900
     },
     {
       "epoch": 21.42,
-      "learning_rate": 8.796859666339547e-05,
-      "loss": 0.0698,
+      "learning_rate": 0.00017400874635568512,
+      "loss": 0.0458,
       "step": 14910
     },
     {
       "epoch": 21.44,
-      "learning_rate": 8.782139352306182e-05,
-      "loss": 0.0645,
+      "learning_rate": 0.0001739212827988338,
+      "loss": 0.0451,
       "step": 14920
     },
     {
       "epoch": 21.45,
-      "learning_rate": 8.767419038272815e-05,
-      "loss": 0.0664,
+      "learning_rate": 0.0001738338192419825,
+      "loss": 0.0435,
       "step": 14930
     },
     {
       "epoch": 21.47,
-      "learning_rate": 8.75269872423945e-05,
-      "loss": 0.0682,
+      "learning_rate": 0.00017374635568513118,
+      "loss": 0.043,
       "step": 14940
     },
     {
       "epoch": 21.48,
-      "learning_rate": 8.737978410206083e-05,
-      "loss": 0.0693,
+      "learning_rate": 0.00017365889212827987,
+      "loss": 0.0461,
       "step": 14950
     },
     {
       "epoch": 21.49,
-      "learning_rate": 8.723258096172718e-05,
-      "loss": 0.0717,
+      "learning_rate": 0.00017357142857142859,
+      "loss": 0.0498,
       "step": 14960
     },
     {
       "epoch": 21.51,
-      "learning_rate": 8.708537782139351e-05,
-      "loss": 0.0818,
+      "learning_rate": 0.00017348396501457722,
+      "loss": 0.0465,
       "step": 14970
     },
     {
       "epoch": 21.52,
-      "learning_rate": 8.693817468105985e-05,
-      "loss": 0.0579,
+      "learning_rate": 0.00017339650145772593,
+      "loss": 0.0386,
       "step": 14980
     },
     {
       "epoch": 21.54,
-      "learning_rate": 8.679097154072619e-05,
-      "loss": 0.0598,
+      "learning_rate": 0.00017330903790087462,
+      "loss": 0.0388,
       "step": 14990
     },
     {
       "epoch": 21.55,
-      "learning_rate": 8.664376840039253e-05,
-      "loss": 0.0638,
+      "learning_rate": 0.0001732215743440233,
+      "loss": 0.0362,
       "step": 15000
     },
     {
       "epoch": 21.55,
-      "eval_loss": 0.14896854758262634,
-      "eval_runtime": 40.2901,
-      "eval_samples_per_second": 11.343,
+      "eval_loss": 0.15051406621932983,
+      "eval_runtime": 40.2706,
+      "eval_samples_per_second": 11.348,
       "eval_steps_per_second": 1.44,
-      "eval_wer": 0.1572052401746725,
+      "eval_wer": 0.15283842794759825,
       "step": 15000
     },
     {
       "epoch": 21.57,
-      "learning_rate": 8.649656526005887e-05,
-      "loss": 0.07,
+      "learning_rate": 0.000173134110787172,
+      "loss": 0.0421,
       "step": 15010
     },
     {
       "epoch": 21.58,
-      "learning_rate": 8.63493621197252e-05,
-      "loss": 0.0689,
+      "learning_rate": 0.00017304664723032069,
+      "loss": 0.0505,
       "step": 15020
     },
     {
       "epoch": 21.59,
-      "learning_rate": 8.620215897939155e-05,
-      "loss": 0.064,
+      "learning_rate": 0.00017295918367346937,
+      "loss": 0.0425,
       "step": 15030
     },
     {
       "epoch": 21.61,
-      "learning_rate": 8.605495583905789e-05,
-      "loss": 0.0582,
+      "learning_rate": 0.00017287172011661806,
+      "loss": 0.0369,
       "step": 15040
     },
     {
       "epoch": 21.62,
-      "learning_rate": 8.590775269872423e-05,
-      "loss": 0.0626,
+      "learning_rate": 0.00017278425655976678,
+      "loss": 0.0457,
       "step": 15050
     },
     {
       "epoch": 21.64,
-      "learning_rate": 8.576054955839057e-05,
-      "loss": 0.0564,
+      "learning_rate": 0.0001726967930029154,
+      "loss": 0.0358,
       "step": 15060
     },
     {
       "epoch": 21.65,
-      "learning_rate": 8.561334641805691e-05,
-      "loss": 0.0642,
+      "learning_rate": 0.00017260932944606413,
+      "loss": 0.0433,
       "step": 15070
     },
     {
       "epoch": 21.67,
-      "learning_rate": 8.546614327772325e-05,
-      "loss": 0.056,
+      "learning_rate": 0.0001725218658892128,
+      "loss": 0.0411,
       "step": 15080
     },
     {
       "epoch": 21.68,
-      "learning_rate": 8.53189401373896e-05,
-      "loss": 0.064,
+      "learning_rate": 0.0001724344023323615,
+      "loss": 0.0382,
       "step": 15090
     },
     {
       "epoch": 21.7,
-      "learning_rate": 8.517173699705593e-05,
-      "loss": 0.0665,
+      "learning_rate": 0.0001723469387755102,
+      "loss": 0.0418,
       "step": 15100
     },
     {
       "epoch": 21.7,
-      "eval_loss": 0.15601423382759094,
-      "eval_runtime": 40.1857,
-      "eval_samples_per_second": 11.372,
-      "eval_steps_per_second": 1.443,
-      "eval_wer": 0.1558500225869598,
+      "eval_loss": 0.16807501018047333,
+      "eval_runtime": 40.3622,
+      "eval_samples_per_second": 11.322,
+      "eval_steps_per_second": 1.437,
+      "eval_wer": 0.15404306580334287,
       "step": 15100
     },
     {
       "epoch": 21.71,
-      "learning_rate": 8.502453385672227e-05,
-      "loss": 0.0687,
+      "learning_rate": 0.00017225947521865888,
+      "loss": 0.043,
       "step": 15110
     },
     {
       "epoch": 21.72,
-      "learning_rate": 8.487733071638861e-05,
-      "loss": 0.0617,
+      "learning_rate": 0.00017217201166180756,
+      "loss": 0.0402,
       "step": 15120
     },
     {
       "epoch": 21.74,
-      "learning_rate": 8.473012757605494e-05,
-      "loss": 0.0851,
+      "learning_rate": 0.00017208454810495625,
+      "loss": 0.059,
       "step": 15130
     },
     {
       "epoch": 21.75,
-      "learning_rate": 8.458292443572129e-05,
-      "loss": 0.0644,
+      "learning_rate": 0.00017199708454810497,
+      "loss": 0.039,
       "step": 15140
     },
     {
       "epoch": 21.77,
-      "learning_rate": 8.443572129538762e-05,
-      "loss": 0.073,
+      "learning_rate": 0.00017190962099125363,
+      "loss": 0.049,
       "step": 15150
     },
     {
       "epoch": 21.78,
-      "learning_rate": 8.428851815505397e-05,
-      "loss": 0.0676,
+      "learning_rate": 0.00017182215743440232,
+      "loss": 0.047,
       "step": 15160
     },
     {
       "epoch": 21.8,
-      "learning_rate": 8.41413150147203e-05,
-      "loss": 0.0772,
+      "learning_rate": 0.000171734693877551,
+      "loss": 0.0545,
       "step": 15170
     },
     {
       "epoch": 21.81,
-      "learning_rate": 8.399411187438665e-05,
-      "loss": 0.0654,
+      "learning_rate": 0.0001716472303206997,
+      "loss": 0.0409,
       "step": 15180
     },
     {
       "epoch": 21.82,
-      "learning_rate": 8.384690873405298e-05,
-      "loss": 0.0832,
+      "learning_rate": 0.00017155976676384838,
+      "loss": 0.0512,
       "step": 15190
     },
     {
       "epoch": 21.84,
-      "learning_rate": 8.369970559371933e-05,
-      "loss": 0.0764,
+      "learning_rate": 0.00017147230320699707,
+      "loss": 0.0521,
       "step": 15200
     },
     {
       "epoch": 21.84,
-      "eval_loss": 0.15133680403232574,
-      "eval_runtime": 40.219,
-      "eval_samples_per_second": 11.363,
-      "eval_steps_per_second": 1.442,
-      "eval_wer": 0.15795813883451287,
+      "eval_loss": 0.1461421400308609,
+      "eval_runtime": 39.9835,
+      "eval_samples_per_second": 11.43,
+      "eval_steps_per_second": 1.451,
+      "eval_wer": 0.1472669778647794,
       "step": 15200
     },
     {
       "epoch": 21.85,
-      "learning_rate": 8.355250245338566e-05,
-      "loss": 0.0564,
+      "learning_rate": 0.00017138483965014576,
+      "loss": 0.0358,
       "step": 15210
     },
     {
       "epoch": 21.87,
-      "learning_rate": 8.340529931305201e-05,
-      "loss": 0.0689,
+      "learning_rate": 0.00017129737609329447,
+      "loss": 0.0419,
       "step": 15220
     },
     {
       "epoch": 21.88,
-      "learning_rate": 8.325809617271834e-05,
-      "loss": 0.0747,
+      "learning_rate": 0.00017120991253644316,
+      "loss": 0.0488,
       "step": 15230
     },
     {
       "epoch": 21.9,
-      "learning_rate": 8.311089303238469e-05,
-      "loss": 0.0566,
+      "learning_rate": 0.00017112244897959182,
+      "loss": 0.0375,
       "step": 15240
     },
     {
       "epoch": 21.91,
-      "learning_rate": 8.296368989205102e-05,
-      "loss": 0.0712,
+      "learning_rate": 0.0001710349854227405,
+      "loss": 0.0492,
       "step": 15250
     },
     {
       "epoch": 21.93,
-      "learning_rate": 8.281648675171737e-05,
-      "loss": 0.0637,
+      "learning_rate": 0.0001709475218658892,
+      "loss": 0.045,
       "step": 15260
     },
     {
       "epoch": 21.94,
-      "learning_rate": 8.26692836113837e-05,
-      "loss": 0.0638,
+      "learning_rate": 0.00017086005830903788,
+      "loss": 0.0392,
       "step": 15270
     },
     {
       "epoch": 21.95,
-      "learning_rate": 8.252208047105004e-05,
-      "loss": 0.0679,
+      "learning_rate": 0.00017077259475218657,
+      "loss": 0.0451,
       "step": 15280
     },
     {
       "epoch": 21.97,
-      "learning_rate": 8.237487733071638e-05,
-      "loss": 0.0616,
+      "learning_rate": 0.00017068513119533526,
+      "loss": 0.0383,
       "step": 15290
     },
     {
       "epoch": 21.98,
-      "learning_rate": 8.222767419038272e-05,
-      "loss": 0.0642,
+      "learning_rate": 0.00017059766763848395,
+      "loss": 0.0437,
       "step": 15300
     },
     {
       "epoch": 21.98,
-      "eval_loss": 0.149861678481102,
-      "eval_runtime": 40.2796,
-      "eval_samples_per_second": 11.346,
-      "eval_steps_per_second": 1.44,
-      "eval_wer": 0.1552477036590875,
+      "eval_loss": 0.14696332812309265,
+      "eval_runtime": 40.0769,
+      "eval_samples_per_second": 11.403,
+      "eval_steps_per_second": 1.447,
+      "eval_wer": 0.14425538322541787,
       "step": 15300
     },
     {
       "epoch": 22.0,
-      "learning_rate": 8.208047105004906e-05,
-      "loss": 0.0694,
+      "learning_rate": 0.00017051020408163266,
+      "loss": 0.0477,
       "step": 15310
     },
     {
       "epoch": 22.01,
-      "learning_rate": 8.19332679097154e-05,
-      "loss": 0.0605,
+      "learning_rate": 0.00017042274052478135,
+      "loss": 0.0389,
       "step": 15320
     },
     {
       "epoch": 22.03,
-      "learning_rate": 8.178606476938174e-05,
-      "loss": 0.0637,
+      "learning_rate": 0.00017033527696793,
+      "loss": 0.0445,
       "step": 15330
     },
     {
       "epoch": 22.04,
-      "learning_rate": 8.163886162904808e-05,
-      "loss": 0.0606,
+      "learning_rate": 0.0001702478134110787,
+      "loss": 0.0441,
       "step": 15340
     },
     {
       "epoch": 22.05,
-      "learning_rate": 8.149165848871443e-05,
-      "loss": 0.0583,
+      "learning_rate": 0.00017016034985422739,
+      "loss": 0.0376,
       "step": 15350
     },
     {
       "epoch": 22.07,
-      "learning_rate": 8.134445534838076e-05,
-      "loss": 0.0483,
+      "learning_rate": 0.00017007288629737607,
+      "loss": 0.0399,
       "step": 15360
     },
     {
       "epoch": 22.08,
-      "learning_rate": 8.11972522080471e-05,
-      "loss": 0.0672,
+      "learning_rate": 0.00016998542274052476,
+      "loss": 0.047,
       "step": 15370
     },
     {
       "epoch": 22.1,
-      "learning_rate": 8.105004906771344e-05,
-      "loss": 0.0558,
+      "learning_rate": 0.00016989795918367345,
+      "loss": 0.0426,
       "step": 15380
     },
     {
       "epoch": 22.11,
-      "learning_rate": 8.090284592737979e-05,
-      "loss": 0.0598,
+      "learning_rate": 0.00016981049562682216,
+      "loss": 0.0458,
       "step": 15390
     },
     {
       "epoch": 22.13,
-      "learning_rate": 8.075564278704612e-05,
-      "loss": 0.0577,
+      "learning_rate": 0.00016972303206997085,
+      "loss": 0.0432,
       "step": 15400
     },
     {
       "epoch": 22.13,
-      "eval_loss": 0.15612779557704926,
-      "eval_runtime": 40.1084,
-      "eval_samples_per_second": 11.394,
-      "eval_steps_per_second": 1.446,
-      "eval_wer": 0.15871103749435325,
+      "eval_loss": 0.15462301671504974,
+      "eval_runtime": 40.0637,
+      "eval_samples_per_second": 11.407,
+      "eval_steps_per_second": 1.448,
+      "eval_wer": 0.1525372684836621,
       "step": 15400
     },
     {
       "epoch": 22.14,
-      "learning_rate": 8.060843964671247e-05,
-      "loss": 0.0704,
+      "learning_rate": 0.00016963556851311954,
+      "loss": 0.0564,
       "step": 15410
     },
     {
       "epoch": 22.16,
-      "learning_rate": 8.04612365063788e-05,
-      "loss": 0.0887,
+      "learning_rate": 0.0001695481049562682,
+      "loss": 0.0718,
       "step": 15420
     },
     {
       "epoch": 22.17,
-      "learning_rate": 8.031403336604513e-05,
-      "loss": 0.0569,
+      "learning_rate": 0.0001694606413994169,
+      "loss": 0.0464,
       "step": 15430
     },
     {
       "epoch": 22.18,
-      "learning_rate": 8.016683022571148e-05,
-      "loss": 0.0672,
+      "learning_rate": 0.00016937317784256558,
+      "loss": 0.0498,
       "step": 15440
     },
     {
       "epoch": 22.2,
-      "learning_rate": 8.001962708537781e-05,
-      "loss": 0.0557,
+      "learning_rate": 0.00016928571428571427,
+      "loss": 0.0444,
       "step": 15450
     },
     {
       "epoch": 22.21,
-      "learning_rate": 7.987242394504416e-05,
-      "loss": 0.041,
+      "learning_rate": 0.00016919825072886295,
+      "loss": 0.0321,
       "step": 15460
     },
     {
       "epoch": 22.23,
-      "learning_rate": 7.97252208047105e-05,
-      "loss": 0.0759,
+      "learning_rate": 0.00016911078717201164,
+      "loss": 0.0566,
       "step": 15470
     },
     {
       "epoch": 22.24,
-      "learning_rate": 7.957801766437684e-05,
-      "loss": 0.0464,
+      "learning_rate": 0.00016902332361516036,
+      "loss": 0.0392,
       "step": 15480
     },
     {
       "epoch": 22.26,
-      "learning_rate": 7.943081452404317e-05,
-      "loss": 0.0764,
+      "learning_rate": 0.00016893586005830904,
+      "loss": 0.052,
       "step": 15490
     },
     {
       "epoch": 22.27,
-      "learning_rate": 7.928361138370952e-05,
-      "loss": 0.0551,
+      "learning_rate": 0.00016884839650145773,
+      "loss": 0.04,
       "step": 15500
     },
     {
       "epoch": 22.27,
-      "eval_loss": 0.1587643325328827,
-      "eval_runtime": 40.2333,
-      "eval_samples_per_second": 11.359,
-      "eval_steps_per_second": 1.442,
-      "eval_wer": 0.1570546604427044,
+      "eval_loss": 0.16076627373695374,
+      "eval_runtime": 40.1413,
+      "eval_samples_per_second": 11.385,
+      "eval_steps_per_second": 1.445,
+      "eval_wer": 0.15464538473121517,
       "step": 15500
     },
     {
       "epoch": 22.28,
-      "learning_rate": 7.913640824337585e-05,
-      "loss": 0.0516,
+      "learning_rate": 0.0001687609329446064,
+      "loss": 0.0389,
       "step": 15510
     },
     {
       "epoch": 22.3,
-      "learning_rate": 7.89892051030422e-05,
-      "loss": 0.0619,
+      "learning_rate": 0.00016867346938775508,
+      "loss": 0.0462,
       "step": 15520
     },
     {
       "epoch": 22.31,
-      "learning_rate": 7.884200196270853e-05,
-      "loss": 0.0618,
+      "learning_rate": 0.00016858600583090377,
+      "loss": 0.0418,
       "step": 15530
     },
     {
       "epoch": 22.33,
-      "learning_rate": 7.869479882237488e-05,
-      "loss": 0.0654,
+      "learning_rate": 0.00016849854227405246,
+      "loss": 0.0472,
       "step": 15540
     },
     {
       "epoch": 22.34,
-      "learning_rate": 7.854759568204122e-05,
-      "loss": 0.0652,
+      "learning_rate": 0.00016841107871720114,
+      "loss": 0.0482,
       "step": 15550
     },
     {
       "epoch": 22.36,
-      "learning_rate": 7.840039254170756e-05,
-      "loss": 0.0614,
+      "learning_rate": 0.00016832361516034983,
+      "loss": 0.0414,
       "step": 15560
     },
     {
       "epoch": 22.37,
-      "learning_rate": 7.82531894013739e-05,
-      "loss": 0.0753,
+      "learning_rate": 0.00016823615160349855,
+      "loss": 0.0522,
       "step": 15570
     },
     {
       "epoch": 22.39,
-      "learning_rate": 7.810598626104022e-05,
-      "loss": 0.0618,
+      "learning_rate": 0.00016814868804664723,
+      "loss": 0.0446,
       "step": 15580
     },
     {
       "epoch": 22.4,
-      "learning_rate": 7.795878312070658e-05,
-      "loss": 0.0634,
+      "learning_rate": 0.0001680612244897959,
+      "loss": 0.0412,
       "step": 15590
     },
     {
       "epoch": 22.41,
-      "learning_rate": 7.78115799803729e-05,
-      "loss": 0.0707,
+      "learning_rate": 0.00016797376093294458,
+      "loss": 0.0489,
       "step": 15600
     },
     {
       "epoch": 22.41,
-      "eval_loss": 0.1521938592195511,
-      "eval_runtime": 40.3124,
-      "eval_samples_per_second": 11.336,
-      "eval_steps_per_second": 1.439,
-      "eval_wer": 0.15810871856648095,
+      "eval_loss": 0.15332674980163574,
+      "eval_runtime": 40.0542,
+      "eval_samples_per_second": 11.41,
+      "eval_steps_per_second": 1.448,
+      "eval_wer": 0.15042915223610903,
       "step": 15600
     },
     {
       "epoch": 22.43,
-      "learning_rate": 7.766437684003926e-05,
-      "loss": 0.0505,
+      "learning_rate": 0.00016788629737609327,
+      "loss": 0.0315,
       "step": 15610
     },
     {
       "epoch": 22.44,
-      "learning_rate": 7.751717369970558e-05,
-      "loss": 0.0701,
+      "learning_rate": 0.00016779883381924196,
+      "loss": 0.0503,
       "step": 15620
     },
     {
       "epoch": 22.46,
-      "learning_rate": 7.736997055937194e-05,
-      "loss": 0.058,
+      "learning_rate": 0.00016771137026239065,
+      "loss": 0.0387,
       "step": 15630
     },
     {
       "epoch": 22.47,
-      "learning_rate": 7.722276741903826e-05,
-      "loss": 0.0668,
+      "learning_rate": 0.00016762390670553934,
+      "loss": 0.0459,
       "step": 15640
     },
     {
       "epoch": 22.49,
-      "learning_rate": 7.707556427870462e-05,
-      "loss": 0.0538,
+      "learning_rate": 0.00016753644314868805,
+      "loss": 0.0387,
       "step": 15650
     },
     {
       "epoch": 22.5,
-      "learning_rate": 7.692836113837094e-05,
-      "loss": 0.0613,
+      "learning_rate": 0.00016744897959183674,
+      "loss": 0.0415,
       "step": 15660
     },
     {
       "epoch": 22.51,
-      "learning_rate": 7.67811579980373e-05,
-      "loss": 0.0562,
+      "learning_rate": 0.00016736151603498543,
+      "loss": 0.0416,
       "step": 15670
     },
     {
       "epoch": 22.53,
-      "learning_rate": 7.663395485770362e-05,
-      "loss": 0.0581,
+      "learning_rate": 0.0001672740524781341,
+      "loss": 0.0442,
       "step": 15680
     },
     {
       "epoch": 22.54,
-      "learning_rate": 7.648675171736998e-05,
-      "loss": 0.0557,
+      "learning_rate": 0.00016718658892128277,
+      "loss": 0.0342,
       "step": 15690
     },
     {
       "epoch": 22.56,
-      "learning_rate": 7.63395485770363e-05,
-      "loss": 0.076,
+      "learning_rate": 0.00016709912536443146,
+      "loss": 0.047,
       "step": 15700
     },
     {
       "epoch": 22.56,
-      "eval_loss": 0.14777785539627075,
-      "eval_runtime": 40.7818,
-      "eval_samples_per_second": 11.206,
-      "eval_steps_per_second": 1.422,
-      "eval_wer": 0.1596145158861617,
+      "eval_loss": 0.14910167455673218,
+      "eval_runtime": 39.949,
+      "eval_samples_per_second": 11.44,
+      "eval_steps_per_second": 1.452,
+      "eval_wer": 0.1453094413491944,
       "step": 15700
     },
     {
       "epoch": 22.57,
-      "learning_rate": 7.619234543670264e-05,
-      "loss": 0.0626,
+      "learning_rate": 0.00016701166180758015,
+      "loss": 0.0468,
       "step": 15710
     },
     {
       "epoch": 22.59,
-      "learning_rate": 7.604514229636898e-05,
-      "loss": 0.069,
+      "learning_rate": 0.00016692419825072884,
+      "loss": 0.0461,
       "step": 15720
     },
     {
       "epoch": 22.6,
-      "learning_rate": 7.589793915603531e-05,
-      "loss": 0.0668,
+      "learning_rate": 0.00016683673469387753,
+      "loss": 0.0453,
       "step": 15730
     },
     {
       "epoch": 22.61,
-      "learning_rate": 7.575073601570166e-05,
-      "loss": 0.0886,
+      "learning_rate": 0.00016674927113702624,
+      "loss": 0.0559,
       "step": 15740
     },
     {
       "epoch": 22.63,
-      "learning_rate": 7.560353287536799e-05,
-      "loss": 0.0619,
+      "learning_rate": 0.00016666180758017493,
+      "loss": 0.0422,
       "step": 15750
     },
     {
       "epoch": 22.64,
-      "learning_rate": 7.547105004906771e-05,
-      "loss": 0.0481,
+      "learning_rate": 0.00016657434402332362,
+      "loss": 0.0367,
       "step": 15760
     },
     {
       "epoch": 22.66,
-      "learning_rate": 7.532384690873406e-05,
-      "loss": 0.0674,
+      "learning_rate": 0.00016648688046647228,
+      "loss": 0.0499,
       "step": 15770
     },
     {
       "epoch": 22.67,
-      "learning_rate": 7.517664376840039e-05,
-      "loss": 0.073,
+      "learning_rate": 0.00016639941690962097,
+      "loss": 0.0454,
       "step": 15780
     },
     {
       "epoch": 22.69,
-      "learning_rate": 7.502944062806674e-05,
-      "loss": 0.057,
+      "learning_rate": 0.00016631195335276965,
+      "loss": 0.043,
       "step": 15790
     },
     {
       "epoch": 22.7,
-      "learning_rate": 7.488223748773307e-05,
-      "loss": 0.0802,
+      "learning_rate": 0.00016622448979591834,
+      "loss": 0.0571,
       "step": 15800
     },
     {
       "epoch": 22.7,
-      "eval_loss": 0.1453210413455963,
-      "eval_runtime": 40.4045,
-      "eval_samples_per_second": 11.311,
-      "eval_steps_per_second": 1.435,
-      "eval_wer": 0.15464538473121517,
+      "eval_loss": 0.1495945006608963,
+      "eval_runtime": 40.0515,
+      "eval_samples_per_second": 11.41,
+      "eval_steps_per_second": 1.448,
+      "eval_wer": 0.15148321035988557,
       "step": 15800
     },
     {
       "epoch": 22.72,
-      "learning_rate": 7.47350343473994e-05,
-      "loss": 0.063,
+      "learning_rate": 0.00016613702623906703,
+      "loss": 0.0405,
       "step": 15810
     },
     {
       "epoch": 22.73,
-      "learning_rate": 7.458783120706575e-05,
-      "loss": 0.0889,
+      "learning_rate": 0.00016604956268221574,
+      "loss": 0.0594,
       "step": 15820
     },
     {
       "epoch": 22.74,
-      "learning_rate": 7.444062806673208e-05,
-      "loss": 0.0682,
+      "learning_rate": 0.00016596209912536443,
+      "loss": 0.049,
       "step": 15830
     },
     {
       "epoch": 22.76,
-      "learning_rate": 7.429342492639843e-05,
-      "loss": 0.0548,
+      "learning_rate": 0.00016587463556851312,
+      "loss": 0.031,
       "step": 15840
     },
     {
       "epoch": 22.77,
-      "learning_rate": 7.414622178606477e-05,
-      "loss": 0.0677,
+      "learning_rate": 0.0001657871720116618,
+      "loss": 0.0434,
       "step": 15850
     },
     {
       "epoch": 22.79,
-      "learning_rate": 7.399901864573111e-05,
-      "loss": 0.0705,
+      "learning_rate": 0.00016569970845481047,
+      "loss": 0.0462,
       "step": 15860
     },
     {
       "epoch": 22.8,
-      "learning_rate": 7.385181550539743e-05,
-      "loss": 0.0634,
+      "learning_rate": 0.00016561224489795916,
+      "loss": 0.0427,
       "step": 15870
     },
     {
       "epoch": 22.82,
-      "learning_rate": 7.370461236506378e-05,
-      "loss": 0.0574,
+      "learning_rate": 0.00016552478134110784,
+      "loss": 0.0335,
       "step": 15880
     },
     {
       "epoch": 22.83,
-      "learning_rate": 7.355740922473011e-05,
-      "loss": 0.0654,
+      "learning_rate": 0.00016543731778425653,
+      "loss": 0.0445,
       "step": 15890
     },
     {
       "epoch": 22.84,
-      "learning_rate": 7.341020608439646e-05,
-      "loss": 0.0593,
+      "learning_rate": 0.00016534985422740522,
+      "loss": 0.0415,
       "step": 15900
     },
     {
       "epoch": 22.84,
-      "eval_loss": 0.14600467681884766,
-      "eval_runtime": 40.4483,
-      "eval_samples_per_second": 11.298,
-      "eval_steps_per_second": 1.434,
-      "eval_wer": 0.1552477036590875,
+      "eval_loss": 0.14830708503723145,
+      "eval_runtime": 40.0588,
+      "eval_samples_per_second": 11.408,
+      "eval_steps_per_second": 1.448,
+      "eval_wer": 0.15268784821563017,
       "step": 15900
     },
     {
       "epoch": 22.86,
-      "learning_rate": 7.326300294406279e-05,
-      "loss": 0.0698,
+      "learning_rate": 0.00016526239067055394,
+      "loss": 0.0393,
       "step": 15910
     },
     {
       "epoch": 22.87,
-      "learning_rate": 7.311579980372914e-05,
-      "loss": 0.0741,
+      "learning_rate": 0.00016517492711370262,
+      "loss": 0.0525,
       "step": 15920
     },
     {
       "epoch": 22.89,
-      "learning_rate": 7.296859666339547e-05,
-      "loss": 0.0498,
+      "learning_rate": 0.0001650874635568513,
+      "loss": 0.0349,
       "step": 15930
     },
     {
       "epoch": 22.9,
-      "learning_rate": 7.282139352306182e-05,
-      "loss": 0.0633,
+      "learning_rate": 0.000165,
+      "loss": 0.0378,
       "step": 15940
     },
     {
       "epoch": 22.92,
-      "learning_rate": 7.267419038272815e-05,
-      "loss": 0.0705,
+      "learning_rate": 0.00016491253644314866,
+      "loss": 0.0474,
       "step": 15950
     },
     {
       "epoch": 22.93,
-      "learning_rate": 7.25269872423945e-05,
-      "loss": 0.0648,
+      "learning_rate": 0.00016482507288629735,
+      "loss": 0.0412,
       "step": 15960
     },
     {
       "epoch": 22.95,
-      "learning_rate": 7.237978410206083e-05,
-      "loss": 0.0609,
+      "learning_rate": 0.00016473760932944604,
+      "loss": 0.0456,
       "step": 15970
     },
     {
       "epoch": 22.96,
-      "learning_rate": 7.223258096172718e-05,
-      "loss": 0.06,
+      "learning_rate": 0.00016465014577259472,
+      "loss": 0.0391,
       "step": 15980
     },
     {
       "epoch": 22.97,
-      "learning_rate": 7.208537782139351e-05,
-      "loss": 0.0535,
+      "learning_rate": 0.0001645626822157434,
+      "loss": 0.0386,
       "step": 15990
     },
     {
       "epoch": 22.99,
-      "learning_rate": 7.193817468105986e-05,
-      "loss": 0.0609,
+      "learning_rate": 0.00016447521865889213,
+      "loss": 0.0429,
       "step": 16000
     },
     {
       "epoch": 22.99,
-      "eval_loss": 0.14838065207004547,
-      "eval_runtime": 40.4523,
-      "eval_samples_per_second": 11.297,
-      "eval_steps_per_second": 1.434,
-      "eval_wer": 0.15675350097876825,
+      "eval_loss": 0.152438685297966,
+      "eval_runtime": 39.9987,
+      "eval_samples_per_second": 11.425,
+      "eval_steps_per_second": 1.45,
+      "eval_wer": 0.14696581840084325,
       "step": 16000
     },
     {
       "epoch": 23.0,
-      "learning_rate": 7.17909715407262e-05,
-      "loss": 0.0718,
+      "learning_rate": 0.00016438775510204081,
+      "loss": 0.0516,
       "step": 16010
     },
     {
       "epoch": 23.02,
-      "learning_rate": 7.164376840039253e-05,
-      "loss": 0.061,
+      "learning_rate": 0.0001643002915451895,
+      "loss": 0.0391,
       "step": 16020
     },
     {
       "epoch": 23.03,
-      "learning_rate": 7.149656526005887e-05,
-      "loss": 0.0591,
+      "learning_rate": 0.0001642128279883382,
+      "loss": 0.0378,
       "step": 16030
     },
     {
       "epoch": 23.05,
-      "learning_rate": 7.134936211972521e-05,
-      "loss": 0.0588,
+      "learning_rate": 0.00016412536443148685,
+      "loss": 0.0431,
       "step": 16040
     },
     {
       "epoch": 23.06,
-      "learning_rate": 7.120215897939155e-05,
-      "loss": 0.0599,
+      "learning_rate": 0.00016403790087463554,
+      "loss": 0.0377,
       "step": 16050
     },
     {
       "epoch": 23.07,
-      "learning_rate": 7.105495583905789e-05,
-      "loss": 0.0639,
+      "learning_rate": 0.00016395043731778423,
+      "loss": 0.0399,
       "step": 16060
     },
     {
       "epoch": 23.09,
-      "learning_rate": 7.090775269872424e-05,
-      "loss": 0.0617,
+      "learning_rate": 0.00016386297376093291,
+      "loss": 0.0415,
       "step": 16070
     },
     {
       "epoch": 23.1,
-      "learning_rate": 7.076054955839057e-05,
-      "loss": 0.0563,
+      "learning_rate": 0.00016377551020408163,
+      "loss": 0.0387,
       "step": 16080
     },
     {
       "epoch": 23.12,
-      "learning_rate": 7.061334641805692e-05,
-      "loss": 0.0638,
+      "learning_rate": 0.00016368804664723032,
+      "loss": 0.0465,
       "step": 16090
     },
     {
       "epoch": 23.13,
-      "learning_rate": 7.046614327772325e-05,
-      "loss": 0.0587,
+      "learning_rate": 0.000163600583090379,
+      "loss": 0.0379,
       "step": 16100
     },
     {
       "epoch": 23.13,
-      "eval_loss": 0.1451614946126938,
-      "eval_runtime": 40.4862,
-      "eval_samples_per_second": 11.288,
-      "eval_steps_per_second": 1.433,
-      "eval_wer": 0.15359132660743863,
+      "eval_loss": 0.14876484870910645,
+      "eval_runtime": 39.9457,
+      "eval_samples_per_second": 11.441,
+      "eval_steps_per_second": 1.452,
+      "eval_wer": 0.1498268333082367,
       "step": 16100
     },
     {
       "epoch": 23.15,
-      "learning_rate": 7.03189401373896e-05,
-      "loss": 0.069,
+      "learning_rate": 0.0001635131195335277,
+      "loss": 0.0488,
       "step": 16110
     },
     {
       "epoch": 23.16,
-      "learning_rate": 7.017173699705593e-05,
-      "loss": 0.0651,
+      "learning_rate": 0.00016342565597667638,
+      "loss": 0.0456,
       "step": 16120
     },
     {
       "epoch": 23.18,
-      "learning_rate": 7.002453385672228e-05,
-      "loss": 0.0551,
+      "learning_rate": 0.00016333819241982504,
+      "loss": 0.0353,
       "step": 16130
     },
     {
       "epoch": 23.19,
-      "learning_rate": 6.987733071638861e-05,
-      "loss": 0.064,
+      "learning_rate": 0.00016325072886297373,
+      "loss": 0.0409,
       "step": 16140
     },
     {
       "epoch": 23.2,
-      "learning_rate": 6.973012757605496e-05,
-      "loss": 0.0604,
+      "learning_rate": 0.00016316326530612242,
+      "loss": 0.0443,
       "step": 16150
     },
     {
       "epoch": 23.22,
-      "learning_rate": 6.958292443572129e-05,
-      "loss": 0.0697,
+      "learning_rate": 0.0001630758017492711,
+      "loss": 0.0501,
       "step": 16160
     },
     {
       "epoch": 23.23,
-      "learning_rate": 6.943572129538762e-05,
-      "loss": 0.0564,
+      "learning_rate": 0.00016298833819241982,
+      "loss": 0.0419,
       "step": 16170
     },
     {
       "epoch": 23.25,
-      "learning_rate": 6.928851815505397e-05,
-      "loss": 0.0583,
+      "learning_rate": 0.0001629008746355685,
+      "loss": 0.0354,
       "step": 16180
     },
     {
       "epoch": 23.26,
-      "learning_rate": 6.91413150147203e-05,
-      "loss": 0.0566,
+      "learning_rate": 0.0001628134110787172,
+      "loss": 0.0392,
       "step": 16190
     },
     {
       "epoch": 23.28,
-      "learning_rate": 6.899411187438665e-05,
-      "loss": 0.0636,
+      "learning_rate": 0.00016272594752186588,
+      "loss": 0.0433,
       "step": 16200
     },
     {
       "epoch": 23.28,
-      "eval_loss": 0.14748641848564148,
-      "eval_runtime": 40.5091,
-      "eval_samples_per_second": 11.281,
-      "eval_steps_per_second": 1.432,
-      "eval_wer": 0.15283842794759825,
+      "eval_loss": 0.1418367326259613,
+      "eval_runtime": 39.9209,
+      "eval_samples_per_second": 11.448,
+      "eval_steps_per_second": 1.453,
+      "eval_wer": 0.14561060081313054,
       "step": 16200
     },
     {
       "epoch": 23.29,
-      "learning_rate": 6.884690873405298e-05,
-      "loss": 0.0613,
+      "learning_rate": 0.00016263848396501457,
+      "loss": 0.0432,
       "step": 16210
     },
     {
       "epoch": 23.3,
-      "learning_rate": 6.869970559371933e-05,
-      "loss": 0.0664,
+      "learning_rate": 0.00016255102040816323,
+      "loss": 0.05,
       "step": 16220
     },
     {
       "epoch": 23.32,
-      "learning_rate": 6.855250245338566e-05,
-      "loss": 0.0562,
+      "learning_rate": 0.00016246355685131192,
+      "loss": 0.0386,
       "step": 16230
     },
     {
       "epoch": 23.33,
-      "learning_rate": 6.840529931305201e-05,
-      "loss": 0.0588,
+      "learning_rate": 0.0001623760932944606,
+      "loss": 0.0439,
       "step": 16240
     },
     {
       "epoch": 23.35,
-      "learning_rate": 6.825809617271834e-05,
-      "loss": 0.0571,
+      "learning_rate": 0.0001622886297376093,
+      "loss": 0.0391,
       "step": 16250
     },
     {
       "epoch": 23.36,
-      "learning_rate": 6.811089303238469e-05,
-      "loss": 0.0562,
+      "learning_rate": 0.000162201166180758,
+      "loss": 0.0307,
       "step": 16260
     },
     {
       "epoch": 23.38,
-      "learning_rate": 6.796368989205102e-05,
-      "loss": 0.0643,
+      "learning_rate": 0.0001621137026239067,
+      "loss": 0.0466,
       "step": 16270
     },
     {
       "epoch": 23.39,
-      "learning_rate": 6.781648675171737e-05,
-      "loss": 0.0578,
+      "learning_rate": 0.0001620262390670554,
+      "loss": 0.0401,
       "step": 16280
     },
     {
       "epoch": 23.41,
-      "learning_rate": 6.76692836113837e-05,
-      "loss": 0.0546,
+      "learning_rate": 0.00016193877551020408,
+      "loss": 0.0395,
       "step": 16290
     },
     {
       "epoch": 23.42,
-      "learning_rate": 6.752208047105005e-05,
-      "loss": 0.0506,
+      "learning_rate": 0.00016185131195335276,
+      "loss": 0.0336,
       "step": 16300
     },
     {
       "epoch": 23.42,
-      "eval_loss": 0.1494196057319641,
-      "eval_runtime": 40.2779,
-      "eval_samples_per_second": 11.346,
-      "eval_steps_per_second": 1.44,
-      "eval_wer": 0.15449480499924712,
+      "eval_loss": 0.14791372418403625,
+      "eval_runtime": 40.2525,
+      "eval_samples_per_second": 11.353,
+      "eval_steps_per_second": 1.441,
+      "eval_wer": 0.15088089143201325,
       "step": 16300
     },
     {
       "epoch": 23.43,
-      "learning_rate": 6.737487733071637e-05,
-      "loss": 0.0556,
+      "learning_rate": 0.00016176384839650142,
+      "loss": 0.0371,
       "step": 16310
     },
     {
       "epoch": 23.45,
-      "learning_rate": 6.722767419038272e-05,
-      "loss": 0.0671,
+      "learning_rate": 0.0001616763848396501,
+      "loss": 0.0456,
       "step": 16320
     },
     {
       "epoch": 23.46,
-      "learning_rate": 6.708047105004905e-05,
-      "loss": 0.0531,
+      "learning_rate": 0.0001615889212827988,
+      "loss": 0.0356,
       "step": 16330
     },
     {
       "epoch": 23.48,
-      "learning_rate": 6.69332679097154e-05,
-      "loss": 0.0664,
+      "learning_rate": 0.00016150145772594751,
+      "loss": 0.0459,
       "step": 16340
     },
     {
       "epoch": 23.49,
-      "learning_rate": 6.678606476938173e-05,
-      "loss": 0.0583,
+      "learning_rate": 0.0001614139941690962,
+      "loss": 0.038,
       "step": 16350
     },
     {
       "epoch": 23.51,
-      "learning_rate": 6.663886162904808e-05,
-      "loss": 0.0605,
+      "learning_rate": 0.0001613265306122449,
+      "loss": 0.0395,
       "step": 16360
     },
     {
       "epoch": 23.52,
-      "learning_rate": 6.649165848871441e-05,
-      "loss": 0.0523,
+      "learning_rate": 0.00016123906705539358,
+      "loss": 0.0395,
       "step": 16370
     },
     {
       "epoch": 23.53,
-      "learning_rate": 6.634445534838076e-05,
-      "loss": 0.0456,
+      "learning_rate": 0.00016115160349854227,
+      "loss": 0.0286,
       "step": 16380
     },
     {
       "epoch": 23.55,
-      "learning_rate": 6.61972522080471e-05,
-      "loss": 0.0773,
+      "learning_rate": 0.00016106413994169095,
+      "loss": 0.0518,
       "step": 16390
     },
     {
       "epoch": 23.56,
-      "learning_rate": 6.605004906771344e-05,
-      "loss": 0.0585,
+      "learning_rate": 0.00016097667638483962,
+      "loss": 0.037,
       "step": 16400
     },
     {
       "epoch": 23.56,
-      "eval_loss": 0.15359418094158173,
-      "eval_runtime": 40.8191,
-      "eval_samples_per_second": 11.196,
-      "eval_steps_per_second": 1.421,
-      "eval_wer": 0.16021683481403404,
+      "eval_loss": 0.14310529828071594,
+      "eval_runtime": 40.01,
+      "eval_samples_per_second": 11.422,
+      "eval_steps_per_second": 1.45,
+      "eval_wer": 0.14621291974100287,
       "step": 16400
     },
     {
       "epoch": 23.58,
-      "learning_rate": 6.590284592737977e-05,
-      "loss": 0.0512,
+      "learning_rate": 0.0001608892128279883,
+      "loss": 0.0287,
       "step": 16410
     },
     {
       "epoch": 23.59,
-      "learning_rate": 6.575564278704612e-05,
-      "loss": 0.0639,
+      "learning_rate": 0.000160801749271137,
+      "loss": 0.0442,
       "step": 16420
     },
     {
       "epoch": 23.61,
-      "learning_rate": 6.560843964671245e-05,
-      "loss": 0.0606,
+      "learning_rate": 0.0001607142857142857,
+      "loss": 0.0421,
       "step": 16430
     },
     {
       "epoch": 23.62,
-      "learning_rate": 6.54612365063788e-05,
-      "loss": 0.0591,
+      "learning_rate": 0.0001606268221574344,
+      "loss": 0.0384,
       "step": 16440
     },
     {
       "epoch": 23.64,
-      "learning_rate": 6.531403336604513e-05,
-      "loss": 0.0458,
+      "learning_rate": 0.00016053935860058308,
+      "loss": 0.0328,
       "step": 16450
     },
     {
       "epoch": 23.65,
-      "learning_rate": 6.516683022571147e-05,
-      "loss": 0.0609,
+      "learning_rate": 0.00016045189504373177,
+      "loss": 0.0388,
       "step": 16460
     },
     {
       "epoch": 23.66,
-      "learning_rate": 6.501962708537781e-05,
-      "loss": 0.0537,
+      "learning_rate": 0.00016036443148688046,
+      "loss": 0.0414,
       "step": 16470
     },
     {
       "epoch": 23.68,
-      "learning_rate": 6.487242394504415e-05,
-      "loss": 0.0546,
+      "learning_rate": 0.00016027696793002915,
+      "loss": 0.039,
       "step": 16480
     },
     {
       "epoch": 23.69,
-      "learning_rate": 6.47252208047105e-05,
-      "loss": 0.0694,
+      "learning_rate": 0.0001601895043731778,
+      "loss": 0.0477,
       "step": 16490
     },
     {
       "epoch": 23.71,
-      "learning_rate": 6.457801766437683e-05,
-      "loss": 0.0585,
+      "learning_rate": 0.0001601020408163265,
+      "loss": 0.0379,
       "step": 16500
     },
     {
       "epoch": 23.71,
-      "eval_loss": 0.14631050825119019,
-      "eval_runtime": 40.2888,
-      "eval_samples_per_second": 11.343,
-      "eval_steps_per_second": 1.44,
-      "eval_wer": 0.15735581990664058,
+      "eval_loss": 0.14307264983654022,
+      "eval_runtime": 39.9826,
+      "eval_samples_per_second": 11.43,
+      "eval_steps_per_second": 1.451,
+      "eval_wer": 0.15208552928775787,
       "step": 16500
     },
     {
       "epoch": 23.72,
-      "learning_rate": 6.443081452404318e-05,
-      "loss": 0.0717,
+      "learning_rate": 0.0001600145772594752,
+      "loss": 0.0455,
       "step": 16510
     },
     {
       "epoch": 23.74,
-      "learning_rate": 6.428361138370951e-05,
-      "loss": 0.0656,
+      "learning_rate": 0.0001599271137026239,
+      "loss": 0.0479,
       "step": 16520
     },
     {
       "epoch": 23.75,
-      "learning_rate": 6.413640824337586e-05,
-      "loss": 0.0461,
+      "learning_rate": 0.00015983965014577258,
+      "loss": 0.0348,
       "step": 16530
     },
     {
       "epoch": 23.76,
-      "learning_rate": 6.398920510304219e-05,
-      "loss": 0.0743,
+      "learning_rate": 0.00015975218658892127,
+      "loss": 0.0486,
       "step": 16540
     },
     {
       "epoch": 23.78,
-      "learning_rate": 6.384200196270854e-05,
-      "loss": 0.0533,
+      "learning_rate": 0.00015966472303206996,
+      "loss": 0.0372,
       "step": 16550
     },
     {
       "epoch": 23.79,
-      "learning_rate": 6.369479882237487e-05,
-      "loss": 0.0672,
+      "learning_rate": 0.00015957725947521865,
+      "loss": 0.0494,
       "step": 16560
     },
     {
       "epoch": 23.81,
-      "learning_rate": 6.354759568204122e-05,
-      "loss": 0.0631,
+      "learning_rate": 0.00015948979591836734,
+      "loss": 0.0484,
       "step": 16570
     },
     {
       "epoch": 23.82,
-      "learning_rate": 6.340039254170755e-05,
-      "loss": 0.0507,
+      "learning_rate": 0.000159402332361516,
+      "loss": 0.0376,
       "step": 16580
     },
     {
       "epoch": 23.84,
-      "learning_rate": 6.32531894013739e-05,
-      "loss": 0.071,
+      "learning_rate": 0.00015931486880466469,
+      "loss": 0.0504,
       "step": 16590
     },
     {
       "epoch": 23.85,
-      "learning_rate": 6.310598626104023e-05,
-      "loss": 0.0546,
+      "learning_rate": 0.0001592274052478134,
+      "loss": 0.0368,
       "step": 16600
     },
     {
       "epoch": 23.85,
-      "eval_loss": 0.14672046899795532,
-      "eval_runtime": 40.2017,
-      "eval_samples_per_second": 11.368,
-      "eval_steps_per_second": 1.443,
-      "eval_wer": 0.1550971239271194,
+      "eval_loss": 0.1425299048423767,
+      "eval_runtime": 40.0088,
+      "eval_samples_per_second": 11.422,
+      "eval_steps_per_second": 1.45,
+      "eval_wer": 0.14636349947297095,
       "step": 16600
     },
     {
       "epoch": 23.86,
-      "learning_rate": 6.295878312070656e-05,
-      "loss": 0.049,
+      "learning_rate": 0.0001591399416909621,
+      "loss": 0.0306,
       "step": 16610
     },
     {
       "epoch": 23.88,
-      "learning_rate": 6.281157998037291e-05,
-      "loss": 0.0709,
+      "learning_rate": 0.00015905247813411078,
+      "loss": 0.052,
       "step": 16620
     },
     {
       "epoch": 23.89,
-      "learning_rate": 6.266437684003924e-05,
-      "loss": 0.0539,
+      "learning_rate": 0.00015896501457725946,
+      "loss": 0.0363,
       "step": 16630
     },
     {
       "epoch": 23.91,
-      "learning_rate": 6.251717369970559e-05,
-      "loss": 0.0669,
+      "learning_rate": 0.00015887755102040815,
+      "loss": 0.0448,
       "step": 16640
     },
     {
       "epoch": 23.92,
-      "learning_rate": 6.236997055937192e-05,
-      "loss": 0.0532,
+      "learning_rate": 0.00015879008746355684,
+      "loss": 0.0367,
       "step": 16650
     },
     {
       "epoch": 23.94,
-      "learning_rate": 6.222276741903827e-05,
-      "loss": 0.047,
+      "learning_rate": 0.00015870262390670553,
+      "loss": 0.0334,
       "step": 16660
     },
     {
       "epoch": 23.95,
-      "learning_rate": 6.20755642787046e-05,
-      "loss": 0.0563,
+      "learning_rate": 0.0001586151603498542,
+      "loss": 0.0391,
       "step": 16670
     },
     {
       "epoch": 23.97,
-      "learning_rate": 6.192836113837095e-05,
-      "loss": 0.0506,
+      "learning_rate": 0.00015852769679300288,
+      "loss": 0.0355,
       "step": 16680
     },
     {
       "epoch": 23.98,
-      "learning_rate": 6.178115799803728e-05,
-      "loss": 0.0708,
+      "learning_rate": 0.0001584402332361516,
+      "loss": 0.0497,
       "step": 16690
     },
     {
       "epoch": 23.99,
-      "learning_rate": 6.163395485770363e-05,
-      "loss": 0.0816,
+      "learning_rate": 0.00015835276967930028,
+      "loss": 0.063,
       "step": 16700
     },
     {
       "epoch": 23.99,
-      "eval_loss": 0.15094982087612152,
-      "eval_runtime": 40.1931,
-      "eval_samples_per_second": 11.37,
-      "eval_steps_per_second": 1.443,
-      "eval_wer": 0.15464538473121517,
+      "eval_loss": 0.14010843634605408,
+      "eval_runtime": 40.037,
+      "eval_samples_per_second": 11.414,
+      "eval_steps_per_second": 1.449,
+      "eval_wer": 0.14305074536967324,
       "step": 16700
     },
     {
       "epoch": 24.01,
-      "learning_rate": 6.148675171736997e-05,
-      "loss": 0.0557,
+      "learning_rate": 0.00015826530612244897,
+      "loss": 0.0439,
       "step": 16710
     },
     {
       "epoch": 24.02,
-      "learning_rate": 6.133954857703631e-05,
-      "loss": 0.0544,
+      "learning_rate": 0.00015817784256559766,
+      "loss": 0.0418,
       "step": 16720
     },
     {
       "epoch": 24.04,
-      "learning_rate": 6.119234543670265e-05,
-      "loss": 0.0681,
+      "learning_rate": 0.00015809037900874634,
+      "loss": 0.043,
       "step": 16730
     },
     {
       "epoch": 24.05,
-      "learning_rate": 6.104514229636899e-05,
-      "loss": 0.0576,
+      "learning_rate": 0.00015800291545189503,
+      "loss": 0.0454,
       "step": 16740
     },
     {
       "epoch": 24.07,
-      "learning_rate": 6.0897939156035326e-05,
-      "loss": 0.0663,
+      "learning_rate": 0.00015791545189504372,
+      "loss": 0.0428,
       "step": 16750
     },
     {
       "epoch": 24.08,
-      "learning_rate": 6.075073601570166e-05,
-      "loss": 0.0617,
+      "learning_rate": 0.00015782798833819238,
+      "loss": 0.0446,
       "step": 16760
     },
     {
       "epoch": 24.09,
-      "learning_rate": 6.0603532875368e-05,
-      "loss": 0.05,
+      "learning_rate": 0.0001577405247813411,
+      "loss": 0.0352,
       "step": 16770
     },
     {
       "epoch": 24.11,
-      "learning_rate": 6.045632973503434e-05,
-      "loss": 0.0622,
+      "learning_rate": 0.00015765306122448978,
+      "loss": 0.0412,
       "step": 16780
     },
     {
       "epoch": 24.12,
-      "learning_rate": 6.030912659470068e-05,
-      "loss": 0.0681,
+      "learning_rate": 0.00015756559766763847,
+      "loss": 0.0537,
       "step": 16790
     },
     {
       "epoch": 24.14,
-      "learning_rate": 6.016192345436702e-05,
-      "loss": 0.0556,
+      "learning_rate": 0.00015747813411078716,
+      "loss": 0.0392,
       "step": 16800
     },
     {
       "epoch": 24.14,
-      "eval_loss": 0.15066751837730408,
-      "eval_runtime": 40.0599,
-      "eval_samples_per_second": 11.408,
-      "eval_steps_per_second": 1.448,
-      "eval_wer": 0.15600060231892787,
+      "eval_loss": 0.14533428847789764,
+      "eval_runtime": 40.1832,
+      "eval_samples_per_second": 11.373,
+      "eval_steps_per_second": 1.443,
+      "eval_wer": 0.15268784821563017,
       "step": 16800
     },
     {
       "epoch": 24.15,
-      "learning_rate": 6.001472031403336e-05,
-      "loss": 0.0585,
+      "learning_rate": 0.00015739067055393585,
+      "loss": 0.0407,
       "step": 16810
     },
     {
       "epoch": 24.17,
-      "learning_rate": 5.98675171736997e-05,
-      "loss": 0.0546,
+      "learning_rate": 0.00015730320699708453,
+      "loss": 0.0379,
       "step": 16820
     },
     {
       "epoch": 24.18,
-      "learning_rate": 5.972031403336604e-05,
-      "loss": 0.0532,
+      "learning_rate": 0.00015721574344023322,
+      "loss": 0.0382,
       "step": 16830
     },
     {
       "epoch": 24.2,
-      "learning_rate": 5.957311089303238e-05,
-      "loss": 0.0664,
+      "learning_rate": 0.00015712827988338194,
+      "loss": 0.0577,
       "step": 16840
     },
     {
       "epoch": 24.21,
-      "learning_rate": 5.942590775269872e-05,
-      "loss": 0.0492,
+      "learning_rate": 0.00015704081632653057,
+      "loss": 0.0334,
       "step": 16850
     },
     {
       "epoch": 24.22,
-      "learning_rate": 5.927870461236506e-05,
-      "loss": 0.0616,
+      "learning_rate": 0.00015695335276967929,
+      "loss": 0.0414,
       "step": 16860
     },
     {
       "epoch": 24.24,
-      "learning_rate": 5.91315014720314e-05,
-      "loss": 0.0582,
+      "learning_rate": 0.00015686588921282797,
+      "loss": 0.0406,
       "step": 16870
     },
     {
       "epoch": 24.25,
-      "learning_rate": 5.898429833169774e-05,
-      "loss": 0.0613,
+      "learning_rate": 0.0001567871720116618,
+      "loss": 0.0475,
       "step": 16880
     },
     {
       "epoch": 24.27,
-      "learning_rate": 5.883709519136408e-05,
-      "loss": 0.0545,
+      "learning_rate": 0.0001566997084548105,
+      "loss": 0.0404,
       "step": 16890
     },
     {
       "epoch": 24.28,
-      "learning_rate": 5.8689892051030415e-05,
-      "loss": 0.056,
+      "learning_rate": 0.00015661224489795918,
+      "loss": 0.0415,
       "step": 16900
     },
     {
       "epoch": 24.28,
-      "eval_loss": 0.15041780471801758,
-      "eval_runtime": 40.1424,
-      "eval_samples_per_second": 11.384,
-      "eval_steps_per_second": 1.445,
-      "eval_wer": 0.16051799427797017,
+      "eval_loss": 0.14745640754699707,
+      "eval_runtime": 40.0707,
+      "eval_samples_per_second": 11.405,
+      "eval_steps_per_second": 1.447,
+      "eval_wer": 0.1499774130402048,
       "step": 16900
     },
     {
       "epoch": 24.3,
-      "learning_rate": 5.8542688910696755e-05,
-      "loss": 0.0638,
+      "learning_rate": 0.00015652478134110787,
+      "loss": 0.0534,
       "step": 16910
     },
     {
       "epoch": 24.31,
-      "learning_rate": 5.8395485770363095e-05,
-      "loss": 0.0512,
+      "learning_rate": 0.00015643731778425653,
+      "loss": 0.0391,
       "step": 16920
     },
     {
       "epoch": 24.32,
-      "learning_rate": 5.8248282630029435e-05,
-      "loss": 0.0619,
+      "learning_rate": 0.00015634985422740522,
+      "loss": 0.043,
       "step": 16930
     },
     {
       "epoch": 24.34,
-      "learning_rate": 5.8101079489695775e-05,
-      "loss": 0.0641,
+      "learning_rate": 0.0001562623906705539,
+      "loss": 0.0499,
       "step": 16940
     },
     {
       "epoch": 24.35,
-      "learning_rate": 5.7953876349362116e-05,
-      "loss": 0.0507,
+      "learning_rate": 0.00015617492711370262,
+      "loss": 0.0401,
       "step": 16950
     },
     {
       "epoch": 24.37,
-      "learning_rate": 5.7806673209028456e-05,
-      "loss": 0.0543,
+      "learning_rate": 0.0001560874635568513,
+      "loss": 0.0477,
       "step": 16960
     },
     {
       "epoch": 24.38,
-      "learning_rate": 5.7659470068694796e-05,
-      "loss": 0.0493,
+      "learning_rate": 0.000156,
+      "loss": 0.0407,
       "step": 16970
     },
     {
       "epoch": 24.4,
-      "learning_rate": 5.7512266928361136e-05,
-      "loss": 0.0567,
+      "learning_rate": 0.00015591253644314869,
+      "loss": 0.0388,
       "step": 16980
     },
     {
       "epoch": 24.41,
-      "learning_rate": 5.7365063788027476e-05,
-      "loss": 0.0776,
+      "learning_rate": 0.00015582507288629737,
+      "loss": 0.0602,
       "step": 16990
     },
     {
       "epoch": 24.43,
-      "learning_rate": 5.7217860647693816e-05,
-      "loss": 0.0621,
+      "learning_rate": 0.00015573760932944606,
+      "loss": 0.0437,
       "step": 17000
     },
     {
       "epoch": 24.43,
-      "eval_loss": 0.14634354412555695,
-      "eval_runtime": 40.0304,
-      "eval_samples_per_second": 11.416,
-      "eval_steps_per_second": 1.449,
-      "eval_wer": 0.15825929829844904,
+      "eval_loss": 0.1374850571155548,
+      "eval_runtime": 40.0686,
+      "eval_samples_per_second": 11.405,
+      "eval_steps_per_second": 1.448,
+      "eval_wer": 0.1433519048336094,
       "step": 17000
     },
     {
       "epoch": 24.44,
-      "learning_rate": 5.7070657507360157e-05,
-      "loss": 0.0731,
+      "learning_rate": 0.00015565014577259472,
+      "loss": 0.0511,
       "step": 17010
     },
     {
       "epoch": 24.45,
-      "learning_rate": 5.69234543670265e-05,
-      "loss": 0.0588,
+      "learning_rate": 0.0001555626822157434,
+      "loss": 0.0409,
       "step": 17020
     },
     {
       "epoch": 24.47,
-      "learning_rate": 5.677625122669284e-05,
-      "loss": 0.0555,
+      "learning_rate": 0.0001554752186588921,
+      "loss": 0.0416,
       "step": 17030
     },
     {
       "epoch": 24.48,
-      "learning_rate": 5.662904808635918e-05,
-      "loss": 0.0633,
+      "learning_rate": 0.0001553877551020408,
+      "loss": 0.0463,
       "step": 17040
     },
     {
       "epoch": 24.5,
-      "learning_rate": 5.6481844946025504e-05,
-      "loss": 0.0477,
+      "learning_rate": 0.0001553002915451895,
+      "loss": 0.0344,
       "step": 17050
     },
     {
       "epoch": 24.51,
-      "learning_rate": 5.6334641805691844e-05,
-      "loss": 0.0584,
+      "learning_rate": 0.0001552128279883382,
+      "loss": 0.0474,
       "step": 17060
     },
     {
       "epoch": 24.53,
-      "learning_rate": 5.6187438665358184e-05,
-      "loss": 0.0594,
+      "learning_rate": 0.00015512536443148688,
+      "loss": 0.0418,
       "step": 17070
     },
     {
       "epoch": 24.54,
-      "learning_rate": 5.6040235525024524e-05,
-      "loss": 0.0613,
+      "learning_rate": 0.00015503790087463556,
+      "loss": 0.0429,
       "step": 17080
     },
     {
       "epoch": 24.55,
-      "learning_rate": 5.5893032384690864e-05,
-      "loss": 0.0608,
+      "learning_rate": 0.00015495043731778425,
+      "loss": 0.0468,
       "step": 17090
     },
     {
       "epoch": 24.57,
-      "learning_rate": 5.5745829244357205e-05,
-      "loss": 0.0514,
+      "learning_rate": 0.0001548629737609329,
+      "loss": 0.0409,
       "step": 17100
     },
     {
       "epoch": 24.57,
-      "eval_loss": 0.14699435234069824,
-      "eval_runtime": 40.2333,
-      "eval_samples_per_second": 11.359,
-      "eval_steps_per_second": 1.442,
-      "eval_wer": 0.15856045776238517,
+      "eval_loss": 0.13606612384319305,
+      "eval_runtime": 39.9263,
+      "eval_samples_per_second": 11.446,
+      "eval_steps_per_second": 1.453,
+      "eval_wer": 0.14711639813281133,
       "step": 17100
     },
     {
       "epoch": 24.58,
-      "learning_rate": 5.5598626104023545e-05,
-      "loss": 0.0718,
+      "learning_rate": 0.0001547755102040816,
+      "loss": 0.0513,
       "step": 17110
     },
     {
       "epoch": 24.6,
-      "learning_rate": 5.5451422963689885e-05,
-      "loss": 0.0498,
+      "learning_rate": 0.0001546880466472303,
+      "loss": 0.0425,
       "step": 17120
     },
     {
       "epoch": 24.61,
-      "learning_rate": 5.5304219823356225e-05,
-      "loss": 0.0534,
+      "learning_rate": 0.000154600583090379,
+      "loss": 0.0397,
       "step": 17130
     },
     {
       "epoch": 24.63,
-      "learning_rate": 5.5157016683022565e-05,
-      "loss": 0.0606,
+      "learning_rate": 0.0001545131195335277,
+      "loss": 0.0447,
       "step": 17140
     },
     {
       "epoch": 24.64,
-      "learning_rate": 5.5009813542688905e-05,
-      "loss": 0.0532,
+      "learning_rate": 0.00015442565597667638,
+      "loss": 0.0375,
       "step": 17150
     },
     {
       "epoch": 24.66,
-      "learning_rate": 5.4862610402355246e-05,
-      "loss": 0.062,
+      "learning_rate": 0.00015433819241982507,
+      "loss": 0.0448,
       "step": 17160
     },
     {
       "epoch": 24.67,
-      "learning_rate": 5.4715407262021586e-05,
-      "loss": 0.0483,
+      "learning_rate": 0.00015425072886297376,
+      "loss": 0.0401,
       "step": 17170
     },
     {
       "epoch": 24.68,
-      "learning_rate": 5.4568204121687926e-05,
-      "loss": 0.0428,
+      "learning_rate": 0.00015416326530612244,
+      "loss": 0.0365,
       "step": 17180
     },
     {
       "epoch": 24.7,
-      "learning_rate": 5.4421000981354266e-05,
-      "loss": 0.0725,
+      "learning_rate": 0.0001540758017492711,
+      "loss": 0.0542,
       "step": 17190
     },
     {
       "epoch": 24.71,
-      "learning_rate": 5.42737978410206e-05,
-      "loss": 0.0487,
+      "learning_rate": 0.0001539883381924198,
+      "loss": 0.0345,
       "step": 17200
     },
     {
       "epoch": 24.71,
-      "eval_loss": 0.1460464745759964,
-      "eval_runtime": 40.1151,
-      "eval_samples_per_second": 11.392,
-      "eval_steps_per_second": 1.446,
-      "eval_wer": 0.1556994428549917,
+      "eval_loss": 0.147772416472435,
+      "eval_runtime": 39.9882,
+      "eval_samples_per_second": 11.428,
+      "eval_steps_per_second": 1.45,
+      "eval_wer": 0.14711639813281133,
       "step": 17200
     },
     {
       "epoch": 24.73,
-      "learning_rate": 5.412659470068694e-05,
-      "loss": 0.0566,
+      "learning_rate": 0.0001539008746355685,
+      "loss": 0.0431,
       "step": 17210
     },
     {
       "epoch": 24.74,
-      "learning_rate": 5.397939156035328e-05,
-      "loss": 0.056,
+      "learning_rate": 0.0001538134110787172,
+      "loss": 0.0456,
       "step": 17220
     },
     {
       "epoch": 24.76,
-      "learning_rate": 5.383218842001962e-05,
-      "loss": 0.0526,
+      "learning_rate": 0.00015372594752186588,
+      "loss": 0.0338,
       "step": 17230
     },
     {
       "epoch": 24.77,
-      "learning_rate": 5.368498527968596e-05,
-      "loss": 0.0623,
+      "learning_rate": 0.00015363848396501457,
+      "loss": 0.0447,
       "step": 17240
     },
     {
       "epoch": 24.78,
-      "learning_rate": 5.35377821393523e-05,
-      "loss": 0.0532,
+      "learning_rate": 0.00015355102040816326,
+      "loss": 0.0334,
       "step": 17250
     },
     {
       "epoch": 24.8,
-      "learning_rate": 5.339057899901864e-05,
-      "loss": 0.0671,
+      "learning_rate": 0.00015346355685131195,
+      "loss": 0.0523,
       "step": 17260
     },
     {
       "epoch": 24.81,
-      "learning_rate": 5.324337585868498e-05,
-      "loss": 0.0525,
+      "learning_rate": 0.0001533760932944606,
+      "loss": 0.0383,
       "step": 17270
     },
     {
       "epoch": 24.83,
-      "learning_rate": 5.309617271835132e-05,
-      "loss": 0.0492,
+      "learning_rate": 0.0001532886297376093,
+      "loss": 0.0372,
       "step": 17280
     },
     {
       "epoch": 24.84,
-      "learning_rate": 5.294896957801766e-05,
-      "loss": 0.0518,
+      "learning_rate": 0.00015320116618075798,
+      "loss": 0.0465,
       "step": 17290
     },
     {
       "epoch": 24.86,
-      "learning_rate": 5.2801766437684e-05,
-      "loss": 0.0474,
+      "learning_rate": 0.0001531137026239067,
+      "loss": 0.0344,
       "step": 17300
     },
     {
       "epoch": 24.86,
-      "eval_loss": 0.1450837403535843,
-      "eval_runtime": 40.0465,
-      "eval_samples_per_second": 11.412,
-      "eval_steps_per_second": 1.448,
-      "eval_wer": 0.15238668875169403,
+      "eval_loss": 0.1439347267150879,
+      "eval_runtime": 40.2695,
+      "eval_samples_per_second": 11.349,
+      "eval_steps_per_second": 1.44,
+      "eval_wer": 0.14756813732871554,
       "step": 17300
     },
     {
       "epoch": 24.87,
-      "learning_rate": 5.265456329735034e-05,
-      "loss": 0.0572,
+      "learning_rate": 0.00015302623906705539,
+      "loss": 0.0427,
       "step": 17310
     },
     {
       "epoch": 24.89,
-      "learning_rate": 5.250736015701668e-05,
-      "loss": 0.0502,
+      "learning_rate": 0.00015293877551020407,
+      "loss": 0.0384,
       "step": 17320
     },
     {
       "epoch": 24.9,
-      "learning_rate": 5.236015701668302e-05,
-      "loss": 0.0592,
+      "learning_rate": 0.00015285131195335276,
+      "loss": 0.0364,
       "step": 17330
     },
     {
       "epoch": 24.91,
-      "learning_rate": 5.221295387634936e-05,
-      "loss": 0.0609,
+      "learning_rate": 0.00015276384839650145,
+      "loss": 0.04,
       "step": 17340
     },
     {
       "epoch": 24.93,
-      "learning_rate": 5.2065750736015695e-05,
-      "loss": 0.0617,
+      "learning_rate": 0.00015267638483965014,
+      "loss": 0.0469,
       "step": 17350
     },
     {
       "epoch": 24.94,
-      "learning_rate": 5.1918547595682035e-05,
-      "loss": 0.0525,
+      "learning_rate": 0.0001525889212827988,
+      "loss": 0.0425,
       "step": 17360
     },
     {
       "epoch": 24.96,
-      "learning_rate": 5.1771344455348375e-05,
-      "loss": 0.0649,
+      "learning_rate": 0.00015250145772594749,
+      "loss": 0.0441,
       "step": 17370
     },
     {
       "epoch": 24.97,
-      "learning_rate": 5.1624141315014716e-05,
-      "loss": 0.0487,
+      "learning_rate": 0.0001524139941690962,
+      "loss": 0.034,
       "step": 17380
     },
     {
       "epoch": 24.99,
-      "learning_rate": 5.1476938174681056e-05,
-      "loss": 0.062,
+      "learning_rate": 0.0001523265306122449,
+      "loss": 0.0459,
       "step": 17390
     },
     {
       "epoch": 25.0,
-      "learning_rate": 5.1329735034347396e-05,
-      "loss": 0.0402,
+      "learning_rate": 0.00015223906705539358,
+      "loss": 0.0328,
       "step": 17400
     },
     {
       "epoch": 25.0,
-      "eval_loss": 0.14534904062747955,
-      "eval_runtime": 40.0375,
-      "eval_samples_per_second": 11.414,
-      "eval_steps_per_second": 1.449,
-      "eval_wer": 0.1513326306279175,
+      "eval_loss": 0.14581048488616943,
+      "eval_runtime": 40.0126,
+      "eval_samples_per_second": 11.421,
+      "eval_steps_per_second": 1.45,
+      "eval_wer": 0.15057973196807709,
       "step": 17400
     },
     {
       "epoch": 25.01,
-      "learning_rate": 5.1182531894013736e-05,
-      "loss": 0.0566,
+      "learning_rate": 0.00015215160349854226,
+      "loss": 0.043,
       "step": 17410
     },
     {
       "epoch": 25.03,
-      "learning_rate": 5.1035328753680076e-05,
-      "loss": 0.0545,
+      "learning_rate": 0.00015206413994169095,
+      "loss": 0.0414,
       "step": 17420
     },
     {
       "epoch": 25.04,
-      "learning_rate": 5.0888125613346416e-05,
-      "loss": 0.0524,
+      "learning_rate": 0.00015197667638483964,
+      "loss": 0.0402,
       "step": 17430
     },
     {
       "epoch": 25.06,
-      "learning_rate": 5.074092247301276e-05,
-      "loss": 0.0482,
+      "learning_rate": 0.00015188921282798833,
+      "loss": 0.0372,
       "step": 17440
     },
     {
       "epoch": 25.07,
-      "learning_rate": 5.05937193326791e-05,
-      "loss": 0.0479,
+      "learning_rate": 0.000151801749271137,
+      "loss": 0.0334,
       "step": 17450
     },
     {
       "epoch": 25.09,
-      "learning_rate": 5.044651619234544e-05,
-      "loss": 0.0574,
+      "learning_rate": 0.00015171428571428568,
+      "loss": 0.0458,
       "step": 17460
     },
     {
       "epoch": 25.1,
-      "learning_rate": 5.029931305201178e-05,
-      "loss": 0.0516,
+      "learning_rate": 0.0001516268221574344,
+      "loss": 0.0364,
       "step": 17470
     },
     {
       "epoch": 25.11,
-      "learning_rate": 5.015210991167812e-05,
-      "loss": 0.0564,
+      "learning_rate": 0.00015153935860058308,
+      "loss": 0.0401,
       "step": 17480
     },
     {
       "epoch": 25.13,
-      "learning_rate": 5.000490677134446e-05,
-      "loss": 0.0478,
+      "learning_rate": 0.00015145189504373177,
+      "loss": 0.0359,
       "step": 17490
     },
     {
       "epoch": 25.14,
-      "learning_rate": 4.9857703631010784e-05,
-      "loss": 0.0754,
+      "learning_rate": 0.00015136443148688046,
+      "loss": 0.0548,
       "step": 17500
     },
     {
       "epoch": 25.14,
-      "eval_loss": 0.14766934514045715,
-      "eval_runtime": 39.973,
-      "eval_samples_per_second": 11.433,
-      "eval_steps_per_second": 1.451,
-      "eval_wer": 0.15148321035988557,
+      "eval_loss": 0.14898838102817535,
+      "eval_runtime": 39.994,
+      "eval_samples_per_second": 11.427,
+      "eval_steps_per_second": 1.45,
+      "eval_wer": 0.14410480349344978,
       "step": 17500
     },
     {
       "epoch": 25.16,
-      "learning_rate": 4.9710500490677124e-05,
-      "loss": 0.0675,
+      "learning_rate": 0.00015127696793002914,
+      "loss": 0.05,
       "step": 17510
     },
     {
       "epoch": 25.17,
-      "learning_rate": 4.9563297350343464e-05,
-      "loss": 0.0486,
+      "learning_rate": 0.00015118950437317783,
+      "loss": 0.0342,
       "step": 17520
     },
     {
       "epoch": 25.19,
-      "learning_rate": 4.9416094210009805e-05,
-      "loss": 0.0603,
+      "learning_rate": 0.00015110204081632652,
+      "loss": 0.0462,
       "step": 17530
     },
     {
       "epoch": 25.2,
-      "learning_rate": 4.9268891069676145e-05,
-      "loss": 0.0655,
+      "learning_rate": 0.00015101457725947518,
+      "loss": 0.0478,
       "step": 17540
     },
     {
       "epoch": 25.22,
-      "learning_rate": 4.9121687929342485e-05,
-      "loss": 0.0523,
+      "learning_rate": 0.00015092711370262387,
+      "loss": 0.0361,
       "step": 17550
     },
     {
       "epoch": 25.23,
-      "learning_rate": 4.8974484789008825e-05,
-      "loss": 0.0565,
+      "learning_rate": 0.00015083965014577258,
+      "loss": 0.0399,
       "step": 17560
     },
     {
       "epoch": 25.24,
-      "learning_rate": 4.8827281648675165e-05,
-      "loss": 0.051,
+      "learning_rate": 0.00015075218658892127,
+      "loss": 0.0403,
       "step": 17570
     },
     {
       "epoch": 25.26,
-      "learning_rate": 4.8680078508341505e-05,
-      "loss": 0.0615,
+      "learning_rate": 0.00015066472303206996,
+      "loss": 0.0442,
       "step": 17580
     },
     {
       "epoch": 25.27,
-      "learning_rate": 4.8532875368007846e-05,
-      "loss": 0.0576,
+      "learning_rate": 0.00015057725947521865,
+      "loss": 0.0401,
       "step": 17590
     },
     {
       "epoch": 25.29,
-      "learning_rate": 4.8385672227674186e-05,
-      "loss": 0.0409,
+      "learning_rate": 0.00015048979591836733,
+      "loss": 0.0335,
       "step": 17600
     },
     {
       "epoch": 25.29,
-      "eval_loss": 0.14516086876392365,
-      "eval_runtime": 40.1057,
-      "eval_samples_per_second": 11.395,
-      "eval_steps_per_second": 1.446,
-      "eval_wer": 0.14907393464839633,
+      "eval_loss": 0.14889946579933167,
+      "eval_runtime": 40.0346,
+      "eval_samples_per_second": 11.415,
+      "eval_steps_per_second": 1.449,
+      "eval_wer": 0.14636349947297095,
       "step": 17600
     },
     {
       "epoch": 25.3,
-      "learning_rate": 4.8238469087340526e-05,
-      "loss": 0.0673,
+      "learning_rate": 0.00015040233236151602,
+      "loss": 0.0489,
       "step": 17610
     },
     {
       "epoch": 25.32,
-      "learning_rate": 4.8091265947006866e-05,
-      "loss": 0.0686,
+      "learning_rate": 0.0001503148688046647,
+      "loss": 0.0495,
       "step": 17620
     },
     {
       "epoch": 25.33,
-      "learning_rate": 4.7944062806673206e-05,
-      "loss": 0.0512,
+      "learning_rate": 0.00015022740524781337,
+      "loss": 0.0428,
       "step": 17630
     },
     {
       "epoch": 25.34,
-      "learning_rate": 4.779685966633954e-05,
-      "loss": 0.0556,
+      "learning_rate": 0.00015013994169096209,
+      "loss": 0.0444,
       "step": 17640
     },
     {
       "epoch": 25.36,
-      "learning_rate": 4.764965652600588e-05,
-      "loss": 0.0619,
+      "learning_rate": 0.00015005247813411077,
+      "loss": 0.0428,
       "step": 17650
     },
     {
       "epoch": 25.37,
-      "learning_rate": 4.750245338567222e-05,
-      "loss": 0.0516,
+      "learning_rate": 0.00014996501457725946,
+      "loss": 0.0403,
       "step": 17660
     },
     {
       "epoch": 25.39,
-      "learning_rate": 4.735525024533856e-05,
-      "loss": 0.0491,
+      "learning_rate": 0.00014987755102040815,
+      "loss": 0.0379,
       "step": 17670
     },
     {
       "epoch": 25.4,
-      "learning_rate": 4.72080471050049e-05,
-      "loss": 0.0561,
+      "learning_rate": 0.00014979008746355684,
+      "loss": 0.0447,
       "step": 17680
     },
     {
       "epoch": 25.42,
-      "learning_rate": 4.706084396467124e-05,
-      "loss": 0.0609,
+      "learning_rate": 0.00014970262390670553,
+      "loss": 0.0449,
       "step": 17690
     },
     {
       "epoch": 25.43,
-      "learning_rate": 4.691364082433758e-05,
-      "loss": 0.0414,
+      "learning_rate": 0.00014961516034985421,
+      "loss": 0.0312,
       "step": 17700
     },
     {
       "epoch": 25.43,
-      "eval_loss": 0.1499926894903183,
-      "eval_runtime": 39.8722,
-      "eval_samples_per_second": 11.462,
-      "eval_steps_per_second": 1.455,
-      "eval_wer": 0.1511820508959494,
+      "eval_loss": 0.1425950974225998,
+      "eval_runtime": 40.0419,
+      "eval_samples_per_second": 11.413,
+      "eval_steps_per_second": 1.448,
+      "eval_wer": 0.13762987501882246,
       "step": 17700
     },
     {
       "epoch": 25.45,
-      "learning_rate": 4.676643768400392e-05,
-      "loss": 0.0555,
+      "learning_rate": 0.0001495276967930029,
+      "loss": 0.0418,
       "step": 17710
     },
     {
       "epoch": 25.46,
-      "learning_rate": 4.661923454367026e-05,
-      "loss": 0.0568,
+      "learning_rate": 0.0001494402332361516,
+      "loss": 0.0387,
       "step": 17720
     },
     {
       "epoch": 25.47,
-      "learning_rate": 4.64720314033366e-05,
-      "loss": 0.0551,
+      "learning_rate": 0.00014935276967930028,
+      "loss": 0.0375,
       "step": 17730
     },
     {
       "epoch": 25.49,
-      "learning_rate": 4.632482826300294e-05,
-      "loss": 0.0441,
+      "learning_rate": 0.00014926530612244897,
+      "loss": 0.0329,
       "step": 17740
     },
     {
       "epoch": 25.5,
-      "learning_rate": 4.617762512266928e-05,
-      "loss": 0.0469,
+      "learning_rate": 0.00014917784256559765,
+      "loss": 0.0336,
       "step": 17750
     },
     {
       "epoch": 25.52,
-      "learning_rate": 4.603042198233562e-05,
-      "loss": 0.0562,
+      "learning_rate": 0.00014909037900874634,
+      "loss": 0.0401,
       "step": 17760
     },
     {
       "epoch": 25.53,
-      "learning_rate": 4.588321884200196e-05,
-      "loss": 0.0511,
+      "learning_rate": 0.00014900291545189503,
+      "loss": 0.0386,
       "step": 17770
     },
     {
       "epoch": 25.55,
-      "learning_rate": 4.57360157016683e-05,
-      "loss": 0.0456,
+      "learning_rate": 0.00014891545189504372,
+      "loss": 0.0308,
       "step": 17780
     },
     {
       "epoch": 25.56,
-      "learning_rate": 4.5588812561334635e-05,
-      "loss": 0.0419,
+      "learning_rate": 0.0001488279883381924,
+      "loss": 0.0355,
       "step": 17790
     },
     {
       "epoch": 25.57,
-      "learning_rate": 4.5441609421000976e-05,
-      "loss": 0.0596,
+      "learning_rate": 0.0001487405247813411,
+      "loss": 0.039,
       "step": 17800
     },
     {
       "epoch": 25.57,
-      "eval_loss": 0.14854370057582855,
-      "eval_runtime": 40.2407,
-      "eval_samples_per_second": 11.357,
-      "eval_steps_per_second": 1.441,
-      "eval_wer": 0.1538924860713748,
+      "eval_loss": 0.14126811921596527,
+      "eval_runtime": 39.9334,
+      "eval_samples_per_second": 11.444,
+      "eval_steps_per_second": 1.452,
+      "eval_wer": 0.14094262912212016,
       "step": 17800
     },
     {
       "epoch": 25.59,
-      "learning_rate": 4.5294406280667316e-05,
-      "loss": 0.0575,
+      "learning_rate": 0.00014865306122448978,
+      "loss": 0.0416,
       "step": 17810
     },
     {
       "epoch": 25.6,
-      "learning_rate": 4.5147203140333656e-05,
-      "loss": 0.0565,
+      "learning_rate": 0.00014856559766763847,
+      "loss": 0.0464,
       "step": 17820
     },
     {
       "epoch": 25.62,
-      "learning_rate": 4.4999999999999996e-05,
-      "loss": 0.0598,
+      "learning_rate": 0.00014847813411078716,
+      "loss": 0.0431,
       "step": 17830
     },
     {
       "epoch": 25.63,
-      "learning_rate": 4.4852796859666336e-05,
-      "loss": 0.0482,
+      "learning_rate": 0.00014839067055393584,
+      "loss": 0.0362,
       "step": 17840
     },
     {
       "epoch": 25.65,
-      "learning_rate": 4.4705593719332676e-05,
-      "loss": 0.0505,
+      "learning_rate": 0.00014830320699708453,
+      "loss": 0.0383,
       "step": 17850
     },
     {
       "epoch": 25.66,
-      "learning_rate": 4.4558390578999017e-05,
-      "loss": 0.0588,
+      "learning_rate": 0.00014821574344023322,
+      "loss": 0.0416,
       "step": 17860
     },
     {
       "epoch": 25.68,
-      "learning_rate": 4.441118743866536e-05,
-      "loss": 0.0481,
+      "learning_rate": 0.0001481282798833819,
+      "loss": 0.0374,
       "step": 17870
     },
     {
       "epoch": 25.69,
-      "learning_rate": 4.42639842983317e-05,
-      "loss": 0.05,
+      "learning_rate": 0.0001480408163265306,
+      "loss": 0.0384,
       "step": 17880
     },
     {
       "epoch": 25.7,
-      "learning_rate": 4.411678115799804e-05,
-      "loss": 0.0587,
+      "learning_rate": 0.00014795335276967928,
+      "loss": 0.0486,
       "step": 17890
     },
     {
       "epoch": 25.72,
-      "learning_rate": 4.396957801766438e-05,
-      "loss": 0.0441,
+      "learning_rate": 0.00014786588921282797,
+      "loss": 0.0316,
       "step": 17900
     },
     {
       "epoch": 25.72,
-      "eval_loss": 0.14327119290828705,
-      "eval_runtime": 39.9632,
-      "eval_samples_per_second": 11.436,
-      "eval_steps_per_second": 1.451,
-      "eval_wer": 0.15344074687547057,
+      "eval_loss": 0.1351412832736969,
+      "eval_runtime": 40.4468,
+      "eval_samples_per_second": 11.299,
+      "eval_steps_per_second": 1.434,
+      "eval_wer": 0.13853335341063092,
       "step": 17900
     },
     {
       "epoch": 25.73,
-      "learning_rate": 4.382237487733072e-05,
-      "loss": 0.0595,
+      "learning_rate": 0.00014777842565597666,
+      "loss": 0.0483,
       "step": 17910
     },
     {
       "epoch": 25.75,
-      "learning_rate": 4.367517173699706e-05,
-      "loss": 0.0519,
+      "learning_rate": 0.00014769096209912535,
+      "loss": 0.0343,
       "step": 17920
     },
     {
       "epoch": 25.76,
-      "learning_rate": 4.35279685966634e-05,
-      "loss": 0.052,
+      "learning_rate": 0.00014760349854227404,
+      "loss": 0.0353,
       "step": 17930
     },
     {
       "epoch": 25.78,
-      "learning_rate": 4.3380765456329724e-05,
-      "loss": 0.0412,
+      "learning_rate": 0.00014751603498542272,
+      "loss": 0.0314,
       "step": 17940
     },
     {
       "epoch": 25.79,
-      "learning_rate": 4.3233562315996065e-05,
-      "loss": 0.0462,
+      "learning_rate": 0.0001474285714285714,
+      "loss": 0.0325,
       "step": 17950
     },
     {
       "epoch": 25.8,
-      "learning_rate": 4.3086359175662405e-05,
-      "loss": 0.0646,
+      "learning_rate": 0.0001473411078717201,
+      "loss": 0.0502,
       "step": 17960
     },
     {
       "epoch": 25.82,
-      "learning_rate": 4.2939156035328745e-05,
-      "loss": 0.0572,
+      "learning_rate": 0.0001472536443148688,
+      "loss": 0.0355,
       "step": 17970
     },
     {
       "epoch": 25.83,
-      "learning_rate": 4.2791952894995085e-05,
-      "loss": 0.062,
+      "learning_rate": 0.00014716618075801747,
+      "loss": 0.0504,
       "step": 17980
     },
     {
       "epoch": 25.85,
-      "learning_rate": 4.2644749754661425e-05,
-      "loss": 0.0485,
+      "learning_rate": 0.00014707871720116616,
+      "loss": 0.0347,
       "step": 17990
     },
     {
       "epoch": 25.86,
-      "learning_rate": 4.2497546614327765e-05,
-      "loss": 0.0517,
+      "learning_rate": 0.00014699125364431485,
+      "loss": 0.0414,
       "step": 18000
     },
     {
       "epoch": 25.86,
-      "eval_loss": 0.1446276158094406,
-      "eval_runtime": 40.1161,
-      "eval_samples_per_second": 11.392,
-      "eval_steps_per_second": 1.446,
-      "eval_wer": 0.15238668875169403,
+      "eval_loss": 0.14004768431186676,
+      "eval_runtime": 40.0192,
+      "eval_samples_per_second": 11.42,
+      "eval_steps_per_second": 1.449,
+      "eval_wer": 0.14410480349344978,
       "step": 18000
     },
     {
       "epoch": 25.88,
-      "learning_rate": 4.2350343473994106e-05,
-      "loss": 0.0556,
+      "learning_rate": 0.00014690379008746357,
+      "loss": 0.0441,
       "step": 18010
     },
     {
       "epoch": 25.89,
-      "learning_rate": 4.2203140333660446e-05,
-      "loss": 0.0542,
+      "learning_rate": 0.00014681632653061223,
+      "loss": 0.0397,
       "step": 18020
     },
     {
       "epoch": 25.91,
-      "learning_rate": 4.2055937193326786e-05,
-      "loss": 0.0517,
+      "learning_rate": 0.00014672886297376091,
+      "loss": 0.0396,
       "step": 18030
     },
     {
       "epoch": 25.92,
-      "learning_rate": 4.1908734052993126e-05,
-      "loss": 0.0483,
+      "learning_rate": 0.0001466413994169096,
+      "loss": 0.033,
       "step": 18040
     },
     {
       "epoch": 25.93,
-      "learning_rate": 4.1761530912659466e-05,
-      "loss": 0.0454,
+      "learning_rate": 0.0001465539358600583,
+      "loss": 0.037,
       "step": 18050
     },
     {
       "epoch": 25.95,
-      "learning_rate": 4.1614327772325806e-05,
-      "loss": 0.0649,
+      "learning_rate": 0.00014646647230320698,
+      "loss": 0.047,
       "step": 18060
     },
     {
       "epoch": 25.96,
-      "learning_rate": 4.1467124631992147e-05,
-      "loss": 0.0449,
+      "learning_rate": 0.00014637900874635567,
+      "loss": 0.0302,
       "step": 18070
     },
     {
       "epoch": 25.98,
-      "learning_rate": 4.131992149165849e-05,
-      "loss": 0.0604,
+      "learning_rate": 0.00014629154518950435,
+      "loss": 0.0434,
       "step": 18080
     },
     {
       "epoch": 25.99,
-      "learning_rate": 4.117271835132482e-05,
-      "loss": 0.0486,
+      "learning_rate": 0.00014620408163265304,
+      "loss": 0.0376,
       "step": 18090
     },
     {
       "epoch": 26.01,
-      "learning_rate": 4.102551521099116e-05,
-      "loss": 0.0752,
+      "learning_rate": 0.00014611661807580176,
+      "loss": 0.053,
       "step": 18100
     },
     {
       "epoch": 26.01,
-      "eval_loss": 0.14016593992710114,
-      "eval_runtime": 40.0408,
-      "eval_samples_per_second": 11.413,
-      "eval_steps_per_second": 1.449,
-      "eval_wer": 0.14967625357626863,
+      "eval_loss": 0.1369037628173828,
+      "eval_runtime": 40.0603,
+      "eval_samples_per_second": 11.408,
+      "eval_steps_per_second": 1.448,
+      "eval_wer": 0.13657581689504592,
       "step": 18100
     },
     {
       "epoch": 26.02,
-      "learning_rate": 4.08783120706575e-05,
-      "loss": 0.0568,
+      "learning_rate": 0.00014602915451895042,
+      "loss": 0.039,
       "step": 18110
     },
     {
       "epoch": 26.03,
-      "learning_rate": 4.073110893032384e-05,
-      "loss": 0.0504,
+      "learning_rate": 0.0001459416909620991,
+      "loss": 0.0339,
       "step": 18120
     },
     {
       "epoch": 26.05,
-      "learning_rate": 4.058390578999018e-05,
-      "loss": 0.0536,
+      "learning_rate": 0.0001458542274052478,
+      "loss": 0.0373,
       "step": 18130
     },
     {
       "epoch": 26.06,
-      "learning_rate": 4.043670264965652e-05,
-      "loss": 0.0465,
+      "learning_rate": 0.0001457667638483965,
+      "loss": 0.0338,
       "step": 18140
     },
     {
       "epoch": 26.08,
-      "learning_rate": 4.028949950932286e-05,
-      "loss": 0.05,
+      "learning_rate": 0.00014567930029154517,
+      "loss": 0.0325,
       "step": 18150
     },
     {
       "epoch": 26.09,
-      "learning_rate": 4.01422963689892e-05,
-      "loss": 0.047,
+      "learning_rate": 0.00014559183673469386,
+      "loss": 0.0351,
       "step": 18160
     },
     {
       "epoch": 26.11,
-      "learning_rate": 3.999509322865554e-05,
-      "loss": 0.0545,
+      "learning_rate": 0.00014550437317784254,
+      "loss": 0.0372,
       "step": 18170
     },
     {
       "epoch": 26.12,
-      "learning_rate": 3.984789008832188e-05,
-      "loss": 0.0559,
+      "learning_rate": 0.00014541690962099123,
+      "loss": 0.0431,
       "step": 18180
     },
     {
       "epoch": 26.14,
-      "learning_rate": 3.970068694798822e-05,
-      "loss": 0.049,
+      "learning_rate": 0.00014532944606413995,
+      "loss": 0.033,
       "step": 18190
     },
     {
       "epoch": 26.15,
-      "learning_rate": 3.955348380765456e-05,
-      "loss": 0.0543,
+      "learning_rate": 0.0001452419825072886,
+      "loss": 0.0461,
       "step": 18200
     },
     {
       "epoch": 26.15,
-      "eval_loss": 0.14435161650180817,
-      "eval_runtime": 40.7685,
-      "eval_samples_per_second": 11.21,
-      "eval_steps_per_second": 1.423,
-      "eval_wer": 0.1519349495557898,
+      "eval_loss": 0.14241348206996918,
+      "eval_runtime": 40.4143,
+      "eval_samples_per_second": 11.308,
+      "eval_steps_per_second": 1.435,
+      "eval_wer": 0.140641469658184,
       "step": 18200
     },
     {
       "epoch": 26.16,
-      "learning_rate": 3.94062806673209e-05,
-      "loss": 0.058,
+      "learning_rate": 0.0001451545189504373,
+      "loss": 0.0428,
       "step": 18210
     },
     {
       "epoch": 26.18,
-      "learning_rate": 3.925907752698724e-05,
-      "loss": 0.0459,
+      "learning_rate": 0.00014506705539358598,
+      "loss": 0.0327,
       "step": 18220
     },
     {
       "epoch": 26.19,
-      "learning_rate": 3.9111874386653576e-05,
-      "loss": 0.064,
+      "learning_rate": 0.0001449795918367347,
+      "loss": 0.0456,
       "step": 18230
     },
     {
       "epoch": 26.21,
-      "learning_rate": 3.8964671246319916e-05,
-      "loss": 0.0452,
+      "learning_rate": 0.00014489212827988336,
+      "loss": 0.0344,
       "step": 18240
     },
     {
       "epoch": 26.22,
-      "learning_rate": 3.8817468105986256e-05,
-      "loss": 0.0497,
+      "learning_rate": 0.00014480466472303205,
+      "loss": 0.0359,
       "step": 18250
     },
     {
       "epoch": 26.24,
-      "learning_rate": 3.8670264965652596e-05,
-      "loss": 0.0505,
+      "learning_rate": 0.00014471720116618074,
+      "loss": 0.0399,
       "step": 18260
     },
     {
       "epoch": 26.25,
-      "learning_rate": 3.8523061825318936e-05,
-      "loss": 0.0551,
+      "learning_rate": 0.00014462973760932945,
+      "loss": 0.0359,
       "step": 18270
     },
     {
       "epoch": 26.26,
-      "learning_rate": 3.8375858684985276e-05,
-      "loss": 0.0565,
+      "learning_rate": 0.00014454227405247814,
+      "loss": 0.0427,
       "step": 18280
     },
     {
       "epoch": 26.28,
-      "learning_rate": 3.822865554465162e-05,
-      "loss": 0.0501,
+      "learning_rate": 0.0001444548104956268,
+      "loss": 0.0387,
       "step": 18290
     },
     {
       "epoch": 26.29,
-      "learning_rate": 3.808145240431796e-05,
-      "loss": 0.0556,
+      "learning_rate": 0.0001443673469387755,
+      "loss": 0.0442,
       "step": 18300
     },
     {
       "epoch": 26.29,
-      "eval_loss": 0.1450069099664688,
-      "eval_runtime": 40.1396,
-      "eval_samples_per_second": 11.385,
-      "eval_steps_per_second": 1.445,
-      "eval_wer": 0.14967625357626863,
+      "eval_loss": 0.14577917754650116,
+      "eval_runtime": 40.0268,
+      "eval_samples_per_second": 11.417,
+      "eval_steps_per_second": 1.449,
+      "eval_wer": 0.142599006173769,
       "step": 18300
     },
     {
       "epoch": 26.31,
-      "learning_rate": 3.79342492639843e-05,
-      "loss": 0.0543,
+      "learning_rate": 0.00014427988338192418,
+      "loss": 0.043,
       "step": 18310
     },
     {
       "epoch": 26.32,
-      "learning_rate": 3.778704612365064e-05,
-      "loss": 0.0547,
+      "learning_rate": 0.0001441924198250729,
+      "loss": 0.0449,
       "step": 18320
     },
     {
       "epoch": 26.34,
-      "learning_rate": 3.763984298331698e-05,
-      "loss": 0.0608,
+      "learning_rate": 0.00014410495626822155,
+      "loss": 0.0458,
       "step": 18330
     },
     {
       "epoch": 26.35,
-      "learning_rate": 3.749263984298331e-05,
-      "loss": 0.0458,
+      "learning_rate": 0.00014401749271137024,
+      "loss": 0.0335,
       "step": 18340
     },
     {
       "epoch": 26.36,
-      "learning_rate": 3.734543670264965e-05,
-      "loss": 0.0481,
+      "learning_rate": 0.00014393002915451893,
+      "loss": 0.0378,
       "step": 18350
     },
     {
       "epoch": 26.38,
-      "learning_rate": 3.719823356231599e-05,
-      "loss": 0.0601,
+      "learning_rate": 0.00014384256559766764,
+      "loss": 0.0458,
       "step": 18360
     },
     {
       "epoch": 26.39,
-      "learning_rate": 3.705103042198233e-05,
-      "loss": 0.0451,
+      "learning_rate": 0.0001437551020408163,
+      "loss": 0.0329,
       "step": 18370
     },
     {
       "epoch": 26.41,
-      "learning_rate": 3.690382728164867e-05,
-      "loss": 0.0623,
+      "learning_rate": 0.000143667638483965,
+      "loss": 0.0468,
       "step": 18380
     },
     {
       "epoch": 26.42,
-      "learning_rate": 3.675662414131501e-05,
-      "loss": 0.0396,
+      "learning_rate": 0.00014358017492711368,
+      "loss": 0.0324,
       "step": 18390
     },
     {
       "epoch": 26.44,
-      "learning_rate": 3.660942100098135e-05,
-      "loss": 0.0514,
+      "learning_rate": 0.0001434927113702624,
+      "loss": 0.0407,
       "step": 18400
     },
     {
       "epoch": 26.44,
-      "eval_loss": 0.14546437561511993,
-      "eval_runtime": 40.3801,
-      "eval_samples_per_second": 11.317,
-      "eval_steps_per_second": 1.436,
-      "eval_wer": 0.15042915223610903,
+      "eval_loss": 0.14220750331878662,
+      "eval_runtime": 40.1386,
+      "eval_samples_per_second": 11.386,
+      "eval_steps_per_second": 1.445,
+      "eval_wer": 0.13988857099834362,
       "step": 18400
     },
     {
       "epoch": 26.45,
-      "learning_rate": 3.646221786064769e-05,
-      "loss": 0.065,
+      "learning_rate": 0.00014340524781341108,
+      "loss": 0.0483,
       "step": 18410
     },
     {
       "epoch": 26.47,
-      "learning_rate": 3.6315014720314025e-05,
-      "loss": 0.0446,
+      "learning_rate": 0.00014331778425655974,
+      "loss": 0.0292,
       "step": 18420
     },
     {
       "epoch": 26.48,
-      "learning_rate": 3.6167811579980365e-05,
-      "loss": 0.0568,
+      "learning_rate": 0.00014323032069970843,
+      "loss": 0.0394,
       "step": 18430
     },
     {
       "epoch": 26.49,
-      "learning_rate": 3.6020608439646706e-05,
-      "loss": 0.0467,
+      "learning_rate": 0.00014314285714285712,
+      "loss": 0.0318,
       "step": 18440
     },
     {
       "epoch": 26.51,
-      "learning_rate": 3.5873405299313046e-05,
-      "loss": 0.0515,
+      "learning_rate": 0.00014305539358600583,
+      "loss": 0.0401,
       "step": 18450
     },
     {
       "epoch": 26.52,
-      "learning_rate": 3.5726202158979386e-05,
-      "loss": 0.0528,
+      "learning_rate": 0.0001429679300291545,
+      "loss": 0.0361,
       "step": 18460
     },
     {
       "epoch": 26.54,
-      "learning_rate": 3.5578999018645726e-05,
-      "loss": 0.0518,
+      "learning_rate": 0.00014288046647230318,
+      "loss": 0.0359,
       "step": 18470
     },
     {
       "epoch": 26.55,
-      "learning_rate": 3.5431795878312066e-05,
-      "loss": 0.0506,
+      "learning_rate": 0.00014279300291545187,
+      "loss": 0.0395,
       "step": 18480
     },
     {
       "epoch": 26.57,
-      "learning_rate": 3.5284592737978406e-05,
-      "loss": 0.0501,
+      "learning_rate": 0.00014270553935860058,
+      "loss": 0.0381,
       "step": 18490
     },
     {
       "epoch": 26.58,
-      "learning_rate": 3.5137389597644747e-05,
-      "loss": 0.0424,
+      "learning_rate": 0.00014261807580174927,
+      "loss": 0.0346,
       "step": 18500
     },
     {
       "epoch": 26.58,
-      "eval_loss": 0.14528827369213104,
-      "eval_runtime": 40.49,
-      "eval_samples_per_second": 11.287,
-      "eval_steps_per_second": 1.432,
-      "eval_wer": 0.15057973196807709,
+      "eval_loss": 0.1381981372833252,
+      "eval_runtime": 40.4312,
+      "eval_samples_per_second": 11.303,
+      "eval_steps_per_second": 1.435,
+      "eval_wer": 0.13627465743110978,
       "step": 18500
     },
     {
       "epoch": 26.59,
-      "learning_rate": 3.499018645731109e-05,
-      "loss": 0.0556,
+      "learning_rate": 0.00014253061224489793,
+      "loss": 0.0421,
       "step": 18510
     },
     {
       "epoch": 26.61,
-      "learning_rate": 3.484298331697743e-05,
-      "loss": 0.0624,
+      "learning_rate": 0.00014244314868804662,
+      "loss": 0.0403,
       "step": 18520
     },
     {
       "epoch": 26.62,
-      "learning_rate": 3.469578017664377e-05,
-      "loss": 0.0551,
+      "learning_rate": 0.00014235568513119534,
+      "loss": 0.0424,
       "step": 18530
     },
     {
       "epoch": 26.64,
-      "learning_rate": 3.454857703631011e-05,
-      "loss": 0.0548,
+      "learning_rate": 0.00014226822157434402,
+      "loss": 0.036,
       "step": 18540
     },
     {
       "epoch": 26.65,
-      "learning_rate": 3.440137389597645e-05,
-      "loss": 0.0501,
+      "learning_rate": 0.00014218075801749268,
+      "loss": 0.0376,
       "step": 18550
     },
     {
       "epoch": 26.67,
-      "learning_rate": 3.425417075564279e-05,
-      "loss": 0.0622,
+      "learning_rate": 0.00014209329446064137,
+      "loss": 0.0429,
       "step": 18560
     },
     {
       "epoch": 26.68,
-      "learning_rate": 3.410696761530912e-05,
-      "loss": 0.0475,
+      "learning_rate": 0.0001420058309037901,
+      "loss": 0.0338,
       "step": 18570
     },
     {
       "epoch": 26.7,
-      "learning_rate": 3.395976447497546e-05,
-      "loss": 0.0534,
+      "learning_rate": 0.00014191836734693878,
+      "loss": 0.0379,
       "step": 18580
     },
     {
       "epoch": 26.71,
-      "learning_rate": 3.38125613346418e-05,
-      "loss": 0.0405,
+      "learning_rate": 0.00014183090379008746,
+      "loss": 0.0286,
       "step": 18590
     },
     {
       "epoch": 26.72,
-      "learning_rate": 3.366535819430814e-05,
-      "loss": 0.0689,
+      "learning_rate": 0.00014174344023323612,
+      "loss": 0.0483,
       "step": 18600
     },
     {
       "epoch": 26.72,
-      "eval_loss": 0.1431579440832138,
-      "eval_runtime": 39.9989,
-      "eval_samples_per_second": 11.425,
-      "eval_steps_per_second": 1.45,
-      "eval_wer": 0.15088089143201325,
+      "eval_loss": 0.13980348408222198,
+      "eval_runtime": 40.3916,
+      "eval_samples_per_second": 11.314,
+      "eval_steps_per_second": 1.436,
+      "eval_wer": 0.14018973046227978,
       "step": 18600
     },
     {
       "epoch": 26.74,
-      "learning_rate": 3.351815505397448e-05,
-      "loss": 0.0541,
+      "learning_rate": 0.0001416559766763848,
+      "loss": 0.0362,
       "step": 18610
     },
     {
       "epoch": 26.75,
-      "learning_rate": 3.337095191364082e-05,
-      "loss": 0.0604,
+      "learning_rate": 0.00014156851311953353,
+      "loss": 0.0378,
       "step": 18620
     },
     {
       "epoch": 26.77,
-      "learning_rate": 3.322374877330716e-05,
-      "loss": 0.0546,
+      "learning_rate": 0.00014148104956268222,
+      "loss": 0.0406,
       "step": 18630
     },
     {
       "epoch": 26.78,
-      "learning_rate": 3.30765456329735e-05,
-      "loss": 0.0476,
+      "learning_rate": 0.00014139358600583088,
+      "loss": 0.0332,
       "step": 18640
     },
     {
       "epoch": 26.8,
-      "learning_rate": 3.2929342492639836e-05,
-      "loss": 0.0532,
+      "learning_rate": 0.00014130612244897956,
+      "loss": 0.0361,
       "step": 18650
     },
     {
       "epoch": 26.81,
-      "learning_rate": 3.2782139352306176e-05,
-      "loss": 0.0361,
+      "learning_rate": 0.00014121865889212828,
+      "loss": 0.0279,
       "step": 18660
     },
     {
       "epoch": 26.82,
-      "learning_rate": 3.2634936211972516e-05,
-      "loss": 0.0366,
+      "learning_rate": 0.00014113119533527697,
+      "loss": 0.025,
       "step": 18670
     },
     {
       "epoch": 26.84,
-      "learning_rate": 3.2487733071638856e-05,
-      "loss": 0.0511,
+      "learning_rate": 0.00014104373177842565,
+      "loss": 0.0405,
       "step": 18680
     },
     {
       "epoch": 26.85,
-      "learning_rate": 3.2340529931305196e-05,
-      "loss": 0.0484,
+      "learning_rate": 0.00014095626822157432,
+      "loss": 0.04,
       "step": 18690
     },
     {
       "epoch": 26.87,
-      "learning_rate": 3.2193326790971536e-05,
-      "loss": 0.0467,
+      "learning_rate": 0.00014086880466472303,
+      "loss": 0.0377,
       "step": 18700
     },
     {
       "epoch": 26.87,
-      "eval_loss": 0.1416984647512436,
-      "eval_runtime": 40.7855,
-      "eval_samples_per_second": 11.205,
-      "eval_steps_per_second": 1.422,
-      "eval_wer": 0.15088089143201325,
+      "eval_loss": 0.13871943950653076,
+      "eval_runtime": 40.3957,
+      "eval_samples_per_second": 11.313,
+      "eval_steps_per_second": 1.436,
+      "eval_wer": 0.14034031019424786,
       "step": 18700
     },
     {
       "epoch": 26.88,
-      "learning_rate": 3.2046123650637877e-05,
-      "loss": 0.0566,
+      "learning_rate": 0.00014078134110787172,
+      "loss": 0.0424,
       "step": 18710
     },
     {
       "epoch": 26.9,
-      "learning_rate": 3.189892051030422e-05,
-      "loss": 0.0387,
+      "learning_rate": 0.0001406938775510204,
+      "loss": 0.0277,
       "step": 18720
     },
     {
       "epoch": 26.91,
-      "learning_rate": 3.175171736997056e-05,
-      "loss": 0.0504,
+      "learning_rate": 0.00014060641399416907,
+      "loss": 0.0386,
       "step": 18730
     },
     {
       "epoch": 26.93,
-      "learning_rate": 3.16045142296369e-05,
-      "loss": 0.0447,
+      "learning_rate": 0.00014051895043731775,
+      "loss": 0.0378,
       "step": 18740
     },
     {
       "epoch": 26.94,
-      "learning_rate": 3.145731108930324e-05,
-      "loss": 0.0454,
+      "learning_rate": 0.00014043148688046647,
+      "loss": 0.0342,
       "step": 18750
     },
     {
       "epoch": 26.95,
-      "learning_rate": 3.131010794896958e-05,
-      "loss": 0.0518,
+      "learning_rate": 0.00014034402332361516,
+      "loss": 0.0389,
       "step": 18760
     },
     {
       "epoch": 26.97,
-      "learning_rate": 3.116290480863592e-05,
-      "loss": 0.0476,
+      "learning_rate": 0.00014025655976676385,
+      "loss": 0.0339,
       "step": 18770
     },
     {
       "epoch": 26.98,
-      "learning_rate": 3.101570166830226e-05,
-      "loss": 0.0463,
+      "learning_rate": 0.0001401690962099125,
+      "loss": 0.0336,
       "step": 18780
     },
     {
       "epoch": 27.0,
-      "learning_rate": 3.086849852796859e-05,
-      "loss": 0.0527,
+      "learning_rate": 0.00014008163265306122,
+      "loss": 0.0383,
       "step": 18790
     },
     {
       "epoch": 27.01,
-      "learning_rate": 3.072129538763493e-05,
-      "loss": 0.0409,
+      "learning_rate": 0.0001399941690962099,
+      "loss": 0.0322,
       "step": 18800
     },
     {
       "epoch": 27.01,
-      "eval_loss": 0.14151474833488464,
-      "eval_runtime": 40.0113,
-      "eval_samples_per_second": 11.422,
-      "eval_steps_per_second": 1.45,
-      "eval_wer": 0.15298900767956633,
+      "eval_loss": 0.1367408037185669,
+      "eval_runtime": 40.3646,
+      "eval_samples_per_second": 11.322,
+      "eval_steps_per_second": 1.437,
+      "eval_wer": 0.13371480198765245,
       "step": 18800
     },
     {
       "epoch": 27.03,
-      "learning_rate": 3.057409224730127e-05,
-      "loss": 0.047,
+      "learning_rate": 0.0001399067055393586,
+      "loss": 0.0354,
       "step": 18810
     },
     {
       "epoch": 27.04,
-      "learning_rate": 3.042688910696761e-05,
-      "loss": 0.0531,
+      "learning_rate": 0.00013981924198250726,
+      "loss": 0.0372,
       "step": 18820
     },
     {
       "epoch": 27.05,
-      "learning_rate": 3.0279685966633952e-05,
-      "loss": 0.0498,
+      "learning_rate": 0.00013973177842565597,
+      "loss": 0.0382,
       "step": 18830
     },
     {
       "epoch": 27.07,
-      "learning_rate": 3.0132482826300292e-05,
-      "loss": 0.0562,
+      "learning_rate": 0.00013964431486880466,
+      "loss": 0.0437,
       "step": 18840
     },
     {
       "epoch": 27.08,
-      "learning_rate": 2.9985279685966632e-05,
-      "loss": 0.0657,
+      "learning_rate": 0.00013955685131195335,
+      "loss": 0.0413,
       "step": 18850
     },
     {
       "epoch": 27.1,
-      "learning_rate": 2.983807654563297e-05,
-      "loss": 0.0474,
+      "learning_rate": 0.00013946938775510204,
+      "loss": 0.0347,
       "step": 18860
     },
     {
       "epoch": 27.11,
-      "learning_rate": 2.969087340529931e-05,
-      "loss": 0.0656,
+      "learning_rate": 0.0001393819241982507,
+      "loss": 0.0434,
       "step": 18870
     },
     {
       "epoch": 27.13,
-      "learning_rate": 2.954367026496565e-05,
-      "loss": 0.0483,
+      "learning_rate": 0.0001392944606413994,
+      "loss": 0.037,
       "step": 18880
     },
     {
       "epoch": 27.14,
-      "learning_rate": 2.939646712463199e-05,
-      "loss": 0.0482,
+      "learning_rate": 0.0001392069970845481,
+      "loss": 0.0355,
       "step": 18890
     },
     {
       "epoch": 27.16,
-      "learning_rate": 2.924926398429833e-05,
-      "loss": 0.0612,
+      "learning_rate": 0.0001391195335276968,
+      "loss": 0.0487,
       "step": 18900
     },
     {
       "epoch": 27.16,
-      "eval_loss": 0.1420021653175354,
-      "eval_runtime": 40.3238,
-      "eval_samples_per_second": 11.333,
-      "eval_steps_per_second": 1.438,
-      "eval_wer": 0.14952567384430054,
+      "eval_loss": 0.13838794827461243,
+      "eval_runtime": 40.1371,
+      "eval_samples_per_second": 11.386,
+      "eval_steps_per_second": 1.445,
+      "eval_wer": 0.140641469658184,
       "step": 18900
     },
     {
       "epoch": 27.17,
-      "learning_rate": 2.910206084396467e-05,
-      "loss": 0.0621,
+      "learning_rate": 0.00013903206997084545,
+      "loss": 0.0444,
       "step": 18910
     },
     {
       "epoch": 27.18,
-      "learning_rate": 2.895485770363101e-05,
-      "loss": 0.0583,
+      "learning_rate": 0.00013894460641399416,
+      "loss": 0.0459,
       "step": 18920
     },
     {
       "epoch": 27.2,
-      "learning_rate": 2.880765456329735e-05,
-      "loss": 0.0569,
+      "learning_rate": 0.00013885714285714285,
+      "loss": 0.0453,
       "step": 18930
     },
     {
       "epoch": 27.21,
-      "learning_rate": 2.8660451422963687e-05,
-      "loss": 0.0529,
+      "learning_rate": 0.00013876967930029154,
+      "loss": 0.0403,
       "step": 18940
     },
     {
       "epoch": 27.23,
-      "learning_rate": 2.8513248282630027e-05,
-      "loss": 0.07,
+      "learning_rate": 0.00013869096209912535,
+      "loss": 0.0487,
       "step": 18950
     },
     {
       "epoch": 27.24,
-      "learning_rate": 2.8366045142296367e-05,
-      "loss": 0.0585,
+      "learning_rate": 0.00013860349854227403,
+      "loss": 0.0404,
       "step": 18960
     },
     {
       "epoch": 27.26,
-      "learning_rate": 2.8218842001962707e-05,
-      "loss": 0.054,
+      "learning_rate": 0.00013851603498542275,
+      "loss": 0.0355,
       "step": 18970
     },
     {
       "epoch": 27.27,
-      "learning_rate": 2.8071638861629047e-05,
-      "loss": 0.0523,
+      "learning_rate": 0.0001384285714285714,
+      "loss": 0.0371,
       "step": 18980
     },
     {
       "epoch": 27.28,
-      "learning_rate": 2.7924435721295388e-05,
-      "loss": 0.0416,
+      "learning_rate": 0.0001383411078717201,
+      "loss": 0.03,
       "step": 18990
     },
     {
       "epoch": 27.3,
-      "learning_rate": 2.7777232580961728e-05,
-      "loss": 0.0501,
+      "learning_rate": 0.00013825364431486878,
+      "loss": 0.0368,
       "step": 19000
     },
     {
       "epoch": 27.3,
-      "eval_loss": 0.14232105016708374,
-      "eval_runtime": 40.1869,
-      "eval_samples_per_second": 11.372,
-      "eval_steps_per_second": 1.443,
-      "eval_wer": 0.14892335491642825,
+      "eval_loss": 0.14037658274173737,
+      "eval_runtime": 40.5049,
+      "eval_samples_per_second": 11.283,
+      "eval_steps_per_second": 1.432,
+      "eval_wer": 0.14034031019424786,
       "step": 19000
     },
     {
       "epoch": 27.31,
-      "learning_rate": 2.763002944062806e-05,
-      "loss": 0.0531,
+      "learning_rate": 0.00013816618075801747,
+      "loss": 0.0412,
       "step": 19010
     },
     {
       "epoch": 27.33,
-      "learning_rate": 2.74828263002944e-05,
-      "loss": 0.0529,
+      "learning_rate": 0.00013807871720116616,
+      "loss": 0.0392,
       "step": 19020
     },
     {
       "epoch": 27.34,
-      "learning_rate": 2.733562315996074e-05,
-      "loss": 0.0519,
+      "learning_rate": 0.00013799125364431485,
+      "loss": 0.0418,
       "step": 19030
     },
     {
       "epoch": 27.36,
-      "learning_rate": 2.718842001962708e-05,
-      "loss": 0.0456,
+      "learning_rate": 0.00013790379008746354,
+      "loss": 0.0365,
       "step": 19040
     },
     {
       "epoch": 27.37,
-      "learning_rate": 2.7041216879293422e-05,
-      "loss": 0.0546,
+      "learning_rate": 0.00013781632653061222,
+      "loss": 0.0397,
       "step": 19050
     },
     {
       "epoch": 27.39,
-      "learning_rate": 2.6894013738959762e-05,
-      "loss": 0.0488,
+      "learning_rate": 0.00013772886297376094,
+      "loss": 0.039,
       "step": 19060
     },
     {
       "epoch": 27.4,
-      "learning_rate": 2.6746810598626102e-05,
-      "loss": 0.052,
+      "learning_rate": 0.0001376413994169096,
+      "loss": 0.0391,
       "step": 19070
     },
     {
       "epoch": 27.41,
-      "learning_rate": 2.659960745829244e-05,
-      "loss": 0.0537,
+      "learning_rate": 0.0001375539358600583,
+      "loss": 0.0402,
       "step": 19080
     },
     {
       "epoch": 27.43,
-      "learning_rate": 2.645240431795878e-05,
-      "loss": 0.0474,
+      "learning_rate": 0.00013746647230320698,
+      "loss": 0.0334,
       "step": 19090
     },
     {
       "epoch": 27.44,
-      "learning_rate": 2.630520117762512e-05,
-      "loss": 0.0579,
+      "learning_rate": 0.0001373790087463557,
+      "loss": 0.0418,
       "step": 19100
     },
     {
       "epoch": 27.44,
-      "eval_loss": 0.14417009055614471,
-      "eval_runtime": 40.3442,
-      "eval_samples_per_second": 11.328,
-      "eval_steps_per_second": 1.438,
-      "eval_wer": 0.15238668875169403,
+      "eval_loss": 0.13838353753089905,
+      "eval_runtime": 40.2187,
+      "eval_samples_per_second": 11.363,
+      "eval_steps_per_second": 1.442,
+      "eval_wer": 0.14079204939015208,
       "step": 19100
     },
     {
       "epoch": 27.46,
-      "learning_rate": 2.615799803729146e-05,
-      "loss": 0.0403,
+      "learning_rate": 0.00013729154518950435,
+      "loss": 0.0298,
       "step": 19110
     },
     {
       "epoch": 27.47,
-      "learning_rate": 2.60107948969578e-05,
-      "loss": 0.0537,
+      "learning_rate": 0.00013720408163265304,
+      "loss": 0.0382,
       "step": 19120
     },
     {
       "epoch": 27.49,
-      "learning_rate": 2.586359175662414e-05,
-      "loss": 0.0529,
+      "learning_rate": 0.00013711661807580173,
+      "loss": 0.0394,
       "step": 19130
     },
     {
       "epoch": 27.5,
-      "learning_rate": 2.571638861629048e-05,
-      "loss": 0.0466,
+      "learning_rate": 0.00013702915451895044,
+      "loss": 0.0341,
       "step": 19140
     },
     {
       "epoch": 27.51,
-      "learning_rate": 2.556918547595682e-05,
-      "loss": 0.0552,
+      "learning_rate": 0.00013694169096209913,
+      "loss": 0.0382,
       "step": 19150
     },
     {
       "epoch": 27.53,
-      "learning_rate": 2.5421982335623157e-05,
-      "loss": 0.0525,
+      "learning_rate": 0.0001368542274052478,
+      "loss": 0.0379,
       "step": 19160
     },
     {
       "epoch": 27.54,
-      "learning_rate": 2.5274779195289497e-05,
-      "loss": 0.0508,
+      "learning_rate": 0.00013676676384839648,
+      "loss": 0.0357,
       "step": 19170
     },
     {
       "epoch": 27.56,
-      "learning_rate": 2.5127576054955837e-05,
-      "loss": 0.047,
+      "learning_rate": 0.00013667930029154517,
+      "loss": 0.0358,
       "step": 19180
     },
     {
       "epoch": 27.57,
-      "learning_rate": 2.4980372914622177e-05,
-      "loss": 0.0459,
+      "learning_rate": 0.00013659183673469388,
+      "loss": 0.031,
       "step": 19190
     },
     {
       "epoch": 27.59,
-      "learning_rate": 2.4833169774288518e-05,
-      "loss": 0.0449,
+      "learning_rate": 0.00013650437317784254,
+      "loss": 0.0332,
       "step": 19200
     },
     {
       "epoch": 27.59,
-      "eval_loss": 0.142539843916893,
-      "eval_runtime": 40.2027,
-      "eval_samples_per_second": 11.367,
-      "eval_steps_per_second": 1.443,
-      "eval_wer": 0.1519349495557898,
+      "eval_loss": 0.14226850867271423,
+      "eval_runtime": 40.214,
+      "eval_samples_per_second": 11.364,
+      "eval_steps_per_second": 1.442,
+      "eval_wer": 0.14154494804999246,
       "step": 19200
     },
     {
       "epoch": 27.6,
-      "learning_rate": 2.4685966633954858e-05,
-      "loss": 0.0526,
+      "learning_rate": 0.00013641690962099123,
+      "loss": 0.0347,
       "step": 19210
     },
     {
       "epoch": 27.61,
-      "learning_rate": 2.4538763493621198e-05,
-      "loss": 0.0453,
+      "learning_rate": 0.00013632944606413992,
+      "loss": 0.0392,
       "step": 19220
     },
     {
       "epoch": 27.63,
-      "learning_rate": 2.439156035328753e-05,
-      "loss": 0.0568,
+      "learning_rate": 0.00013624198250728863,
+      "loss": 0.0424,
       "step": 19230
     },
     {
       "epoch": 27.64,
-      "learning_rate": 2.424435721295387e-05,
-      "loss": 0.0441,
+      "learning_rate": 0.00013615451895043732,
+      "loss": 0.0322,
       "step": 19240
     },
     {
       "epoch": 27.66,
-      "learning_rate": 2.409715407262021e-05,
-      "loss": 0.0516,
+      "learning_rate": 0.00013606705539358598,
+      "loss": 0.0395,
       "step": 19250
     },
     {
       "epoch": 27.67,
-      "learning_rate": 2.3949950932286552e-05,
-      "loss": 0.0448,
+      "learning_rate": 0.00013597959183673467,
+      "loss": 0.0338,
       "step": 19260
     },
     {
       "epoch": 27.69,
-      "learning_rate": 2.3802747791952892e-05,
-      "loss": 0.0411,
+      "learning_rate": 0.00013589212827988339,
+      "loss": 0.0334,
       "step": 19270
     },
     {
       "epoch": 27.7,
-      "learning_rate": 2.3655544651619232e-05,
-      "loss": 0.0495,
+      "learning_rate": 0.00013580466472303207,
+      "loss": 0.0409,
       "step": 19280
     },
     {
       "epoch": 27.72,
-      "learning_rate": 2.3508341511285572e-05,
-      "loss": 0.0483,
+      "learning_rate": 0.00013571720116618073,
+      "loss": 0.0358,
       "step": 19290
     },
     {
       "epoch": 27.73,
-      "learning_rate": 2.336113837095191e-05,
-      "loss": 0.0615,
+      "learning_rate": 0.00013562973760932942,
+      "loss": 0.046,
       "step": 19300
     },
     {
       "epoch": 27.73,
-      "eval_loss": 0.14353878796100616,
-      "eval_runtime": 40.2924,
-      "eval_samples_per_second": 11.342,
-      "eval_steps_per_second": 1.439,
-      "eval_wer": 0.15148321035988557,
+      "eval_loss": 0.13767357170581818,
+      "eval_runtime": 40.1195,
+      "eval_samples_per_second": 11.391,
+      "eval_steps_per_second": 1.446,
+      "eval_wer": 0.13657581689504592,
       "step": 19300
     },
     {
       "epoch": 27.74,
-      "learning_rate": 2.321393523061825e-05,
-      "loss": 0.0478,
+      "learning_rate": 0.0001355422740524781,
+      "loss": 0.039,
       "step": 19310
     },
     {
       "epoch": 27.76,
-      "learning_rate": 2.306673209028459e-05,
-      "loss": 0.0515,
+      "learning_rate": 0.00013545481049562682,
+      "loss": 0.034,
       "step": 19320
     },
     {
       "epoch": 27.77,
-      "learning_rate": 2.291952894995093e-05,
-      "loss": 0.0488,
+      "learning_rate": 0.0001353673469387755,
+      "loss": 0.0393,
       "step": 19330
     },
     {
       "epoch": 27.79,
-      "learning_rate": 2.277232580961727e-05,
-      "loss": 0.0482,
+      "learning_rate": 0.00013527988338192417,
+      "loss": 0.0351,
       "step": 19340
     },
     {
       "epoch": 27.8,
-      "learning_rate": 2.262512266928361e-05,
-      "loss": 0.0503,
+      "learning_rate": 0.00013519241982507286,
+      "loss": 0.0412,
       "step": 19350
     },
     {
       "epoch": 27.82,
-      "learning_rate": 2.247791952894995e-05,
-      "loss": 0.0388,
+      "learning_rate": 0.00013510495626822158,
+      "loss": 0.03,
       "step": 19360
     },
     {
       "epoch": 27.83,
-      "learning_rate": 2.233071638861629e-05,
-      "loss": 0.0654,
+      "learning_rate": 0.00013501749271137026,
+      "loss": 0.0443,
       "step": 19370
     },
     {
       "epoch": 27.84,
-      "learning_rate": 2.2183513248282627e-05,
-      "loss": 0.0439,
+      "learning_rate": 0.00013493002915451892,
+      "loss": 0.0352,
       "step": 19380
     },
     {
       "epoch": 27.86,
-      "learning_rate": 2.2036310107948967e-05,
-      "loss": 0.0409,
+      "learning_rate": 0.0001348425655976676,
+      "loss": 0.0307,
       "step": 19390
     },
     {
       "epoch": 27.87,
-      "learning_rate": 2.1889106967615307e-05,
-      "loss": 0.063,
+      "learning_rate": 0.00013475510204081633,
+      "loss": 0.0525,
       "step": 19400
     },
     {
       "epoch": 27.87,
-      "eval_loss": 0.14183540642261505,
-      "eval_runtime": 40.2851,
-      "eval_samples_per_second": 11.344,
-      "eval_steps_per_second": 1.44,
-      "eval_wer": 0.1511820508959494,
+      "eval_loss": 0.13637852668762207,
+      "eval_runtime": 40.0608,
+      "eval_samples_per_second": 11.408,
+      "eval_steps_per_second": 1.448,
+      "eval_wer": 0.13657581689504592,
       "step": 19400
     },
     {
       "epoch": 27.89,
-      "learning_rate": 2.1741903827281648e-05,
-      "loss": 0.0444,
+      "learning_rate": 0.00013467638483965013,
+      "loss": 0.0328,
       "step": 19410
     },
     {
       "epoch": 27.9,
-      "learning_rate": 2.1594700686947988e-05,
-      "loss": 0.0494,
+      "learning_rate": 0.00013458892128279882,
+      "loss": 0.0347,
       "step": 19420
     },
     {
       "epoch": 27.92,
-      "learning_rate": 2.1447497546614328e-05,
-      "loss": 0.0464,
+      "learning_rate": 0.0001345014577259475,
+      "loss": 0.0349,
       "step": 19430
     },
     {
       "epoch": 27.93,
-      "learning_rate": 2.1300294406280668e-05,
-      "loss": 0.0425,
+      "learning_rate": 0.0001344139941690962,
+      "loss": 0.0373,
       "step": 19440
     },
     {
       "epoch": 27.95,
-      "learning_rate": 2.1153091265947e-05,
-      "loss": 0.0519,
+      "learning_rate": 0.00013432653061224488,
+      "loss": 0.0393,
       "step": 19450
     },
     {
       "epoch": 27.96,
-      "learning_rate": 2.100588812561334e-05,
-      "loss": 0.0466,
+      "learning_rate": 0.00013423906705539357,
+      "loss": 0.0351,
       "step": 19460
     },
     {
       "epoch": 27.97,
-      "learning_rate": 2.0858684985279682e-05,
-      "loss": 0.0479,
+      "learning_rate": 0.00013415160349854226,
+      "loss": 0.0375,
       "step": 19470
     },
     {
       "epoch": 27.99,
-      "learning_rate": 2.0711481844946022e-05,
-      "loss": 0.0433,
+      "learning_rate": 0.00013406413994169095,
+      "loss": 0.0385,
       "step": 19480
     },
     {
       "epoch": 28.0,
-      "learning_rate": 2.0564278704612362e-05,
-      "loss": 0.0472,
+      "learning_rate": 0.00013397667638483964,
+      "loss": 0.0372,
       "step": 19490
     },
     {
       "epoch": 28.02,
-      "learning_rate": 2.0417075564278702e-05,
-      "loss": 0.0422,
+      "learning_rate": 0.00013388921282798832,
+      "loss": 0.0349,
       "step": 19500
     },
     {
       "epoch": 28.02,
-      "eval_loss": 0.14156275987625122,
-      "eval_runtime": 40.1542,
-      "eval_samples_per_second": 11.381,
-      "eval_steps_per_second": 1.444,
-      "eval_wer": 0.14817045625658787,
+      "eval_loss": 0.1400323212146759,
+      "eval_runtime": 40.3762,
+      "eval_samples_per_second": 11.319,
+      "eval_steps_per_second": 1.436,
+      "eval_wer": 0.13627465743110978,
       "step": 19500
     },
     {
       "epoch": 28.03,
-      "learning_rate": 2.0269872423945042e-05,
-      "loss": 0.0457,
+      "learning_rate": 0.000133801749271137,
+      "loss": 0.0333,
       "step": 19510
     },
     {
       "epoch": 28.05,
-      "learning_rate": 2.0122669283611383e-05,
-      "loss": 0.051,
+      "learning_rate": 0.0001337142857142857,
+      "loss": 0.0414,
       "step": 19520
     },
     {
       "epoch": 28.06,
-      "learning_rate": 1.997546614327772e-05,
-      "loss": 0.0448,
+      "learning_rate": 0.0001336268221574344,
+      "loss": 0.037,
       "step": 19530
     },
     {
       "epoch": 28.07,
-      "learning_rate": 1.982826300294406e-05,
-      "loss": 0.0419,
+      "learning_rate": 0.00013353935860058308,
+      "loss": 0.0319,
       "step": 19540
     },
     {
       "epoch": 28.09,
-      "learning_rate": 1.96810598626104e-05,
-      "loss": 0.0488,
+      "learning_rate": 0.00013345189504373176,
+      "loss": 0.0366,
       "step": 19550
     },
     {
       "epoch": 28.1,
-      "learning_rate": 1.953385672227674e-05,
-      "loss": 0.0524,
+      "learning_rate": 0.00013336443148688045,
+      "loss": 0.0377,
       "step": 19560
     },
     {
       "epoch": 28.12,
-      "learning_rate": 1.938665358194308e-05,
-      "loss": 0.0404,
+      "learning_rate": 0.00013327696793002914,
+      "loss": 0.0317,
       "step": 19570
     },
     {
       "epoch": 28.13,
-      "learning_rate": 1.923945044160942e-05,
-      "loss": 0.0431,
+      "learning_rate": 0.00013318950437317783,
+      "loss": 0.0347,
       "step": 19580
     },
     {
       "epoch": 28.15,
-      "learning_rate": 1.909224730127576e-05,
-      "loss": 0.0428,
+      "learning_rate": 0.00013310204081632652,
+      "loss": 0.0322,
       "step": 19590
     },
     {
       "epoch": 28.16,
-      "learning_rate": 1.8945044160942097e-05,
-      "loss": 0.0468,
+      "learning_rate": 0.0001330145772594752,
+      "loss": 0.0375,
       "step": 19600
     },
     {
       "epoch": 28.16,
-      "eval_loss": 0.14033865928649902,
-      "eval_runtime": 40.1669,
-      "eval_samples_per_second": 11.378,
-      "eval_steps_per_second": 1.444,
-      "eval_wer": 0.14952567384430054,
+      "eval_loss": 0.14294388890266418,
+      "eval_runtime": 40.0691,
+      "eval_samples_per_second": 11.405,
+      "eval_steps_per_second": 1.447,
+      "eval_wer": 0.13793103448275862,
       "step": 19600
     },
     {
       "epoch": 28.18,
-      "learning_rate": 1.8797841020608437e-05,
-      "loss": 0.0554,
+      "learning_rate": 0.0001329271137026239,
+      "loss": 0.0404,
       "step": 19610
     },
     {
       "epoch": 28.19,
-      "learning_rate": 1.8650637880274777e-05,
-      "loss": 0.057,
+      "learning_rate": 0.00013283965014577258,
+      "loss": 0.0446,
       "step": 19620
     },
     {
       "epoch": 28.2,
-      "learning_rate": 1.8503434739941118e-05,
-      "loss": 0.0387,
+      "learning_rate": 0.00013275218658892127,
+      "loss": 0.0307,
       "step": 19630
     },
     {
       "epoch": 28.22,
-      "learning_rate": 1.8356231599607458e-05,
-      "loss": 0.0496,
+      "learning_rate": 0.00013266472303206996,
+      "loss": 0.0413,
       "step": 19640
     },
     {
       "epoch": 28.23,
-      "learning_rate": 1.8209028459273798e-05,
-      "loss": 0.0547,
+      "learning_rate": 0.00013257725947521864,
+      "loss": 0.043,
       "step": 19650
     },
     {
       "epoch": 28.25,
-      "learning_rate": 1.8061825318940135e-05,
-      "loss": 0.0383,
+      "learning_rate": 0.00013248979591836733,
+      "loss": 0.0303,
       "step": 19660
     },
     {
       "epoch": 28.26,
-      "learning_rate": 1.7914622178606475e-05,
-      "loss": 0.0445,
+      "learning_rate": 0.00013240233236151602,
+      "loss": 0.0339,
       "step": 19670
     },
     {
       "epoch": 28.28,
-      "learning_rate": 1.7767419038272815e-05,
-      "loss": 0.0391,
+      "learning_rate": 0.0001323148688046647,
+      "loss": 0.032,
       "step": 19680
     },
     {
       "epoch": 28.29,
-      "learning_rate": 1.7620215897939155e-05,
-      "loss": 0.052,
+      "learning_rate": 0.0001322274052478134,
+      "loss": 0.0403,
       "step": 19690
     },
     {
       "epoch": 28.3,
-      "learning_rate": 1.7473012757605495e-05,
-      "loss": 0.0421,
+      "learning_rate": 0.00013213994169096208,
+      "loss": 0.0315,
       "step": 19700
     },
     {
       "epoch": 28.3,
-      "eval_loss": 0.14019441604614258,
-      "eval_runtime": 40.1804,
-      "eval_samples_per_second": 11.374,
-      "eval_steps_per_second": 1.443,
-      "eval_wer": 0.1511820508959494,
+      "eval_loss": 0.134770467877388,
+      "eval_runtime": 40.3128,
+      "eval_samples_per_second": 11.336,
+      "eval_steps_per_second": 1.439,
+      "eval_wer": 0.13401596145158862,
       "step": 19700
     },
     {
       "epoch": 28.32,
-      "learning_rate": 1.7325809617271832e-05,
-      "loss": 0.0461,
+      "learning_rate": 0.00013205247813411077,
+      "loss": 0.0327,
       "step": 19710
     },
     {
       "epoch": 28.33,
-      "learning_rate": 1.7178606476938172e-05,
-      "loss": 0.0495,
+      "learning_rate": 0.00013196501457725946,
+      "loss": 0.0407,
       "step": 19720
     },
     {
       "epoch": 28.35,
-      "learning_rate": 1.7031403336604513e-05,
-      "loss": 0.0406,
+      "learning_rate": 0.00013187755102040815,
+      "loss": 0.0304,
       "step": 19730
     },
     {
       "epoch": 28.36,
-      "learning_rate": 1.6884200196270853e-05,
-      "loss": 0.0612,
+      "learning_rate": 0.00013179008746355683,
+      "loss": 0.0445,
       "step": 19740
     },
     {
       "epoch": 28.38,
-      "learning_rate": 1.6736997055937193e-05,
-      "loss": 0.044,
+      "learning_rate": 0.00013170262390670552,
+      "loss": 0.0361,
       "step": 19750
     },
     {
       "epoch": 28.39,
-      "learning_rate": 1.6589793915603533e-05,
-      "loss": 0.0439,
+      "learning_rate": 0.0001316151603498542,
+      "loss": 0.0341,
       "step": 19760
     },
     {
       "epoch": 28.41,
-      "learning_rate": 1.644259077526987e-05,
-      "loss": 0.0503,
+      "learning_rate": 0.0001315276967930029,
+      "loss": 0.0342,
       "step": 19770
     },
     {
       "epoch": 28.42,
-      "learning_rate": 1.629538763493621e-05,
-      "loss": 0.04,
+      "learning_rate": 0.00013144023323615159,
+      "loss": 0.0297,
       "step": 19780
     },
     {
       "epoch": 28.43,
-      "learning_rate": 1.614818449460255e-05,
-      "loss": 0.035,
+      "learning_rate": 0.00013135276967930027,
+      "loss": 0.0251,
       "step": 19790
     },
     {
       "epoch": 28.45,
-      "learning_rate": 1.600098135426889e-05,
-      "loss": 0.0446,
+      "learning_rate": 0.00013126530612244896,
+      "loss": 0.0329,
       "step": 19800
     },
     {
       "epoch": 28.45,
-      "eval_loss": 0.1425383985042572,
-      "eval_runtime": 40.1802,
-      "eval_samples_per_second": 11.374,
-      "eval_steps_per_second": 1.443,
-      "eval_wer": 0.1493750941123325,
+      "eval_loss": 0.13576379418373108,
+      "eval_runtime": 40.3207,
+      "eval_samples_per_second": 11.334,
+      "eval_steps_per_second": 1.438,
+      "eval_wer": 0.13296190332781208,
       "step": 19800
     },
     {
       "epoch": 28.46,
-      "learning_rate": 1.585377821393523e-05,
-      "loss": 0.0487,
+      "learning_rate": 0.00013117784256559765,
+      "loss": 0.0331,
       "step": 19810
     },
     {
       "epoch": 28.48,
-      "learning_rate": 1.570657507360157e-05,
-      "loss": 0.066,
+      "learning_rate": 0.00013109037900874634,
+      "loss": 0.0456,
       "step": 19820
     },
     {
       "epoch": 28.49,
-      "learning_rate": 1.555937193326791e-05,
-      "loss": 0.0541,
+      "learning_rate": 0.00013100291545189503,
+      "loss": 0.0336,
       "step": 19830
     },
     {
       "epoch": 28.51,
-      "learning_rate": 1.5412168792934248e-05,
-      "loss": 0.047,
+      "learning_rate": 0.00013091545189504374,
+      "loss": 0.0333,
       "step": 19840
     },
     {
       "epoch": 28.52,
-      "learning_rate": 1.5264965652600588e-05,
-      "loss": 0.0415,
+      "learning_rate": 0.0001308279883381924,
+      "loss": 0.0321,
       "step": 19850
     },
     {
       "epoch": 28.53,
-      "learning_rate": 1.5117762512266928e-05,
-      "loss": 0.0401,
+      "learning_rate": 0.0001307405247813411,
+      "loss": 0.0279,
       "step": 19860
     },
     {
       "epoch": 28.55,
-      "learning_rate": 1.4970559371933268e-05,
-      "loss": 0.0457,
+      "learning_rate": 0.00013065306122448978,
+      "loss": 0.0321,
       "step": 19870
     },
     {
       "epoch": 28.56,
-      "learning_rate": 1.4823356231599605e-05,
-      "loss": 0.0445,
+      "learning_rate": 0.00013056559766763846,
+      "loss": 0.0329,
       "step": 19880
     },
     {
       "epoch": 28.58,
-      "learning_rate": 1.4676153091265945e-05,
-      "loss": 0.0419,
+      "learning_rate": 0.00013047813411078715,
+      "loss": 0.0339,
       "step": 19890
     },
     {
       "epoch": 28.59,
-      "learning_rate": 1.4528949950932285e-05,
-      "loss": 0.048,
+      "learning_rate": 0.00013039067055393584,
+      "loss": 0.0388,
       "step": 19900
     },
     {
       "epoch": 28.59,
-      "eval_loss": 0.14261671900749207,
-      "eval_runtime": 39.9856,
-      "eval_samples_per_second": 11.429,
-      "eval_steps_per_second": 1.451,
-      "eval_wer": 0.1474175575967475,
+      "eval_loss": 0.13492873311042786,
+      "eval_runtime": 40.1943,
+      "eval_samples_per_second": 11.37,
+      "eval_steps_per_second": 1.443,
+      "eval_wer": 0.13883451287456708,
       "step": 19900
     },
     {
       "epoch": 28.61,
-      "learning_rate": 1.4381746810598624e-05,
-      "loss": 0.043,
+      "learning_rate": 0.00013030320699708453,
+      "loss": 0.029,
       "step": 19910
     },
     {
       "epoch": 28.62,
-      "learning_rate": 1.4234543670264964e-05,
-      "loss": 0.0526,
+      "learning_rate": 0.00013021574344023322,
+      "loss": 0.0425,
       "step": 19920
     },
     {
       "epoch": 28.64,
-      "learning_rate": 1.4087340529931304e-05,
-      "loss": 0.0433,
+      "learning_rate": 0.00013012827988338193,
+      "loss": 0.0335,
       "step": 19930
     },
     {
       "epoch": 28.65,
-      "learning_rate": 1.3940137389597644e-05,
-      "loss": 0.0523,
+      "learning_rate": 0.0001300408163265306,
+      "loss": 0.0396,
       "step": 19940
     },
     {
       "epoch": 28.66,
-      "learning_rate": 1.3792934249263983e-05,
-      "loss": 0.0477,
+      "learning_rate": 0.00012995335276967928,
+      "loss": 0.0384,
       "step": 19950
     },
     {
       "epoch": 28.68,
-      "learning_rate": 1.3645731108930323e-05,
-      "loss": 0.0438,
+      "learning_rate": 0.00012986588921282797,
+      "loss": 0.0332,
       "step": 19960
     },
     {
       "epoch": 28.69,
-      "learning_rate": 1.3498527968596663e-05,
-      "loss": 0.0634,
+      "learning_rate": 0.00012977842565597668,
+      "loss": 0.0479,
       "step": 19970
     },
     {
       "epoch": 28.71,
-      "learning_rate": 1.3351324828263003e-05,
-      "loss": 0.0488,
+      "learning_rate": 0.00012969096209912534,
+      "loss": 0.0391,
       "step": 19980
     },
     {
       "epoch": 28.72,
-      "learning_rate": 1.320412168792934e-05,
-      "loss": 0.0561,
+      "learning_rate": 0.00012960349854227403,
+      "loss": 0.0411,
       "step": 19990
     },
     {
       "epoch": 28.74,
-      "learning_rate": 1.305691854759568e-05,
-      "loss": 0.0429,
+      "learning_rate": 0.00012951603498542272,
+      "loss": 0.0337,
       "step": 20000
     },
     {
       "epoch": 28.74,
-      "eval_loss": 0.14055244624614716,
-      "eval_runtime": 40.1463,
-      "eval_samples_per_second": 11.383,
-      "eval_steps_per_second": 1.445,
-      "eval_wer": 0.146514079204939,
+      "eval_loss": 0.1368357241153717,
+      "eval_runtime": 40.167,
+      "eval_samples_per_second": 11.378,
+      "eval_steps_per_second": 1.444,
+      "eval_wer": 0.136726396627014,
       "step": 20000
     },
     {
       "epoch": 28.75,
-      "learning_rate": 1.290971540726202e-05,
-      "loss": 0.0455,
+      "learning_rate": 0.0001294285714285714,
+      "loss": 0.0295,
       "step": 20010
     },
     {
       "epoch": 28.76,
-      "learning_rate": 1.2762512266928359e-05,
-      "loss": 0.0559,
+      "learning_rate": 0.00012934110787172012,
+      "loss": 0.0394,
       "step": 20020
     },
     {
       "epoch": 28.78,
-      "learning_rate": 1.2615309126594699e-05,
-      "loss": 0.0384,
+      "learning_rate": 0.00012925364431486878,
+      "loss": 0.0262,
       "step": 20030
     },
     {
       "epoch": 28.79,
-      "learning_rate": 1.2468105986261039e-05,
-      "loss": 0.0537,
+      "learning_rate": 0.00012916618075801747,
+      "loss": 0.0351,
       "step": 20040
     },
     {
       "epoch": 28.81,
-      "learning_rate": 1.232090284592738e-05,
-      "loss": 0.0477,
+      "learning_rate": 0.00012907871720116616,
+      "loss": 0.0354,
       "step": 20050
     },
     {
       "epoch": 28.82,
-      "learning_rate": 1.2173699705593718e-05,
-      "loss": 0.0438,
+      "learning_rate": 0.00012899125364431487,
+      "loss": 0.0332,
       "step": 20060
     },
     {
       "epoch": 28.84,
-      "learning_rate": 1.2026496565260058e-05,
-      "loss": 0.0424,
+      "learning_rate": 0.00012890379008746353,
+      "loss": 0.0349,
       "step": 20070
     },
     {
       "epoch": 28.85,
-      "learning_rate": 1.1879293424926398e-05,
-      "loss": 0.0658,
+      "learning_rate": 0.00012881632653061222,
+      "loss": 0.0518,
       "step": 20080
     },
     {
       "epoch": 28.86,
-      "learning_rate": 1.1732090284592738e-05,
-      "loss": 0.0496,
+      "learning_rate": 0.0001287288629737609,
+      "loss": 0.0388,
       "step": 20090
     },
     {
       "epoch": 28.88,
-      "learning_rate": 1.1584887144259077e-05,
-      "loss": 0.0457,
+      "learning_rate": 0.00012864139941690963,
+      "loss": 0.0363,
       "step": 20100
     },
     {
       "epoch": 28.88,
-      "eval_loss": 0.14235013723373413,
-      "eval_runtime": 40.2402,
-      "eval_samples_per_second": 11.357,
-      "eval_steps_per_second": 1.441,
-      "eval_wer": 0.14606234000903479,
+      "eval_loss": 0.13563397526741028,
+      "eval_runtime": 40.5164,
+      "eval_samples_per_second": 11.279,
+      "eval_steps_per_second": 1.432,
+      "eval_wer": 0.13717813582291824,
       "step": 20100
     },
     {
       "epoch": 28.89,
-      "learning_rate": 1.1437684003925415e-05,
-      "loss": 0.0438,
+      "learning_rate": 0.0001285539358600583,
+      "loss": 0.0339,
       "step": 20110
     },
     {
       "epoch": 28.91,
-      "learning_rate": 1.1290480863591755e-05,
-      "loss": 0.0575,
+      "learning_rate": 0.00012846647230320697,
+      "loss": 0.0446,
       "step": 20120
     },
     {
       "epoch": 28.92,
-      "learning_rate": 1.1143277723258095e-05,
-      "loss": 0.048,
+      "learning_rate": 0.00012837900874635566,
+      "loss": 0.0373,
       "step": 20130
     },
     {
       "epoch": 28.94,
-      "learning_rate": 1.0996074582924434e-05,
-      "loss": 0.0448,
+      "learning_rate": 0.00012829154518950435,
+      "loss": 0.0337,
       "step": 20140
     },
     {
       "epoch": 28.95,
-      "learning_rate": 1.0848871442590774e-05,
-      "loss": 0.0463,
+      "learning_rate": 0.00012820408163265306,
+      "loss": 0.0361,
       "step": 20150
     },
     {
       "epoch": 28.97,
-      "learning_rate": 1.0701668302257114e-05,
-      "loss": 0.0513,
+      "learning_rate": 0.00012811661807580173,
+      "loss": 0.0359,
       "step": 20160
     },
     {
       "epoch": 28.98,
-      "learning_rate": 1.0554465161923453e-05,
-      "loss": 0.0438,
+      "learning_rate": 0.0001280291545189504,
+      "loss": 0.0392,
       "step": 20170
     },
     {
       "epoch": 28.99,
-      "learning_rate": 1.0407262021589793e-05,
-      "loss": 0.0452,
+      "learning_rate": 0.0001279416909620991,
+      "loss": 0.0352,
       "step": 20180
     },
     {
       "epoch": 29.01,
-      "learning_rate": 1.0260058881256133e-05,
-      "loss": 0.0487,
+      "learning_rate": 0.00012785422740524782,
+      "loss": 0.0446,
       "step": 20190
     },
     {
       "epoch": 29.02,
-      "learning_rate": 1.0112855740922473e-05,
-      "loss": 0.0469,
+      "learning_rate": 0.0001277667638483965,
+      "loss": 0.0353,
       "step": 20200
     },
     {
       "epoch": 29.02,
-      "eval_loss": 0.14129261672496796,
-      "eval_runtime": 40.3493,
-      "eval_samples_per_second": 11.326,
-      "eval_steps_per_second": 1.437,
-      "eval_wer": 0.14561060081313054,
+      "eval_loss": 0.13083459436893463,
+      "eval_runtime": 40.1834,
+      "eval_samples_per_second": 11.373,
+      "eval_steps_per_second": 1.443,
+      "eval_wer": 0.13386538171962054,
       "step": 20200
     },
     {
       "epoch": 29.04,
-      "learning_rate": 9.965652600588812e-06,
-      "loss": 0.0488,
+      "learning_rate": 0.00012767930029154517,
+      "loss": 0.0378,
       "step": 20210
     },
     {
       "epoch": 29.05,
-      "learning_rate": 9.818449460255152e-06,
-      "loss": 0.0448,
+      "learning_rate": 0.00012759183673469385,
+      "loss": 0.0341,
       "step": 20220
     },
     {
       "epoch": 29.07,
-      "learning_rate": 9.671246319921492e-06,
-      "loss": 0.0451,
+      "learning_rate": 0.00012750437317784257,
+      "loss": 0.0314,
       "step": 20230
     },
     {
       "epoch": 29.08,
-      "learning_rate": 9.52404317958783e-06,
-      "loss": 0.0551,
+      "learning_rate": 0.00012741690962099126,
+      "loss": 0.0382,
       "step": 20240
     },
     {
       "epoch": 29.09,
-      "learning_rate": 9.376840039254169e-06,
-      "loss": 0.0475,
+      "learning_rate": 0.00012732944606413992,
+      "loss": 0.0338,
       "step": 20250
     },
     {
       "epoch": 29.11,
-      "learning_rate": 9.22963689892051e-06,
-      "loss": 0.0424,
+      "learning_rate": 0.0001272419825072886,
+      "loss": 0.0278,
       "step": 20260
     },
     {
       "epoch": 29.12,
-      "learning_rate": 9.08243375858685e-06,
-      "loss": 0.0427,
+      "learning_rate": 0.00012715451895043732,
+      "loss": 0.032,
       "step": 20270
     },
     {
       "epoch": 29.14,
-      "learning_rate": 8.93523061825319e-06,
-      "loss": 0.0487,
+      "learning_rate": 0.000127067055393586,
+      "loss": 0.0379,
       "step": 20280
     },
     {
       "epoch": 29.15,
-      "learning_rate": 8.788027477919528e-06,
-      "loss": 0.0553,
+      "learning_rate": 0.0001269795918367347,
+      "loss": 0.0429,
       "step": 20290
     },
     {
       "epoch": 29.17,
-      "learning_rate": 8.640824337585868e-06,
-      "loss": 0.0543,
+      "learning_rate": 0.00012689212827988336,
+      "loss": 0.0457,
       "step": 20300
     },
     {
       "epoch": 29.17,
-      "eval_loss": 0.14143255352973938,
-      "eval_runtime": 40.1054,
-      "eval_samples_per_second": 11.395,
-      "eval_steps_per_second": 1.446,
-      "eval_wer": 0.146514079204939,
+      "eval_loss": 0.13166461884975433,
+      "eval_runtime": 40.2261,
+      "eval_samples_per_second": 11.361,
+      "eval_steps_per_second": 1.442,
+      "eval_wer": 0.13537117903930132,
       "step": 20300
     },
     {
       "epoch": 29.18,
-      "learning_rate": 8.493621197252207e-06,
-      "loss": 0.0515,
+      "learning_rate": 0.00012680466472303204,
+      "loss": 0.0394,
       "step": 20310
     },
     {
       "epoch": 29.2,
-      "learning_rate": 8.346418056918547e-06,
-      "loss": 0.0423,
+      "learning_rate": 0.00012671720116618076,
+      "loss": 0.0358,
       "step": 20320
     },
     {
       "epoch": 29.21,
-      "learning_rate": 8.199214916584887e-06,
-      "loss": 0.0382,
+      "learning_rate": 0.00012662973760932945,
+      "loss": 0.029,
       "step": 20330
     },
     {
       "epoch": 29.22,
-      "learning_rate": 8.052011776251227e-06,
-      "loss": 0.0436,
+      "learning_rate": 0.0001265422740524781,
+      "loss": 0.0356,
       "step": 20340
     },
     {
       "epoch": 29.24,
-      "learning_rate": 7.904808635917566e-06,
-      "loss": 0.0447,
+      "learning_rate": 0.0001264548104956268,
+      "loss": 0.0298,
       "step": 20350
     },
     {
       "epoch": 29.25,
-      "learning_rate": 7.757605495583906e-06,
-      "loss": 0.0386,
+      "learning_rate": 0.0001263673469387755,
+      "loss": 0.0255,
       "step": 20360
     },
     {
       "epoch": 29.27,
-      "learning_rate": 7.610402355250245e-06,
-      "loss": 0.0537,
+      "learning_rate": 0.0001262798833819242,
+      "loss": 0.0427,
       "step": 20370
     },
     {
       "epoch": 29.28,
-      "learning_rate": 7.4631992149165835e-06,
-      "loss": 0.0408,
+      "learning_rate": 0.00012619241982507289,
+      "loss": 0.0296,
       "step": 20380
     },
     {
       "epoch": 29.3,
-      "learning_rate": 7.315996074582924e-06,
-      "loss": 0.041,
+      "learning_rate": 0.00012610495626822155,
+      "loss": 0.0307,
       "step": 20390
     },
     {
       "epoch": 29.31,
-      "learning_rate": 7.168792934249263e-06,
-      "loss": 0.0416,
+      "learning_rate": 0.00012601749271137026,
+      "loss": 0.0338,
       "step": 20400
     },
     {
       "epoch": 29.31,
-      "eval_loss": 0.14171102643013,
-      "eval_runtime": 40.0654,
-      "eval_samples_per_second": 11.406,
-      "eval_steps_per_second": 1.448,
-      "eval_wer": 0.14681523866887516,
+      "eval_loss": 0.12933945655822754,
+      "eval_runtime": 40.0847,
+      "eval_samples_per_second": 11.401,
+      "eval_steps_per_second": 1.447,
+      "eval_wer": 0.13552175877126937,
       "step": 20400
     },
     {
       "epoch": 29.32,
-      "learning_rate": 7.021589793915603e-06,
-      "loss": 0.0445,
+      "learning_rate": 0.00012593002915451895,
+      "loss": 0.0313,
       "step": 20410
     },
     {
       "epoch": 29.34,
-      "learning_rate": 6.8743866535819425e-06,
-      "loss": 0.0501,
+      "learning_rate": 0.00012584256559766764,
+      "loss": 0.04,
       "step": 20420
     },
     {
       "epoch": 29.35,
-      "learning_rate": 6.741903827281648e-06,
-      "loss": 0.0454,
+      "learning_rate": 0.0001257551020408163,
+      "loss": 0.0357,
       "step": 20430
     },
     {
       "epoch": 29.37,
-      "learning_rate": 6.594700686947987e-06,
-      "loss": 0.0383,
+      "learning_rate": 0.000125667638483965,
+      "loss": 0.0279,
       "step": 20440
     },
     {
       "epoch": 29.38,
-      "learning_rate": 6.4474975466143275e-06,
-      "loss": 0.0403,
+      "learning_rate": 0.0001255801749271137,
+      "loss": 0.0315,
       "step": 20450
     },
     {
       "epoch": 29.4,
-      "learning_rate": 6.300294406280667e-06,
-      "loss": 0.0387,
+      "learning_rate": 0.0001254927113702624,
+      "loss": 0.0286,
       "step": 20460
     },
     {
       "epoch": 29.41,
-      "learning_rate": 6.153091265947007e-06,
-      "loss": 0.055,
+      "learning_rate": 0.00012540524781341108,
+      "loss": 0.0405,
       "step": 20470
     },
     {
       "epoch": 29.43,
-      "learning_rate": 6.0058881256133454e-06,
-      "loss": 0.0377,
+      "learning_rate": 0.00012531778425655974,
+      "loss": 0.0317,
       "step": 20480
     },
     {
       "epoch": 29.44,
-      "learning_rate": 5.858684985279686e-06,
-      "loss": 0.0539,
+      "learning_rate": 0.00012523032069970845,
+      "loss": 0.0394,
       "step": 20490
     },
     {
       "epoch": 29.45,
-      "learning_rate": 5.711481844946025e-06,
-      "loss": 0.0491,
+      "learning_rate": 0.00012514285714285714,
+      "loss": 0.0355,
       "step": 20500
     },
     {
       "epoch": 29.45,
-      "eval_loss": 0.1411919742822647,
-      "eval_runtime": 40.3501,
-      "eval_samples_per_second": 11.326,
-      "eval_steps_per_second": 1.437,
-      "eval_wer": 0.14711639813281133,
+      "eval_loss": 0.13684523105621338,
+      "eval_runtime": 40.2263,
+      "eval_samples_per_second": 11.361,
+      "eval_steps_per_second": 1.442,
+      "eval_wer": 0.12979972895648245,
       "step": 20500
     },
     {
       "epoch": 29.47,
-      "learning_rate": 5.564278704612364e-06,
-      "loss": 0.0521,
+      "learning_rate": 0.00012505539358600583,
+      "loss": 0.0354,
       "step": 20510
     },
     {
       "epoch": 29.48,
-      "learning_rate": 5.417075564278704e-06,
-      "loss": 0.0384,
+      "learning_rate": 0.0001249679300291545,
+      "loss": 0.0322,
       "step": 20520
     },
     {
       "epoch": 29.5,
-      "learning_rate": 5.269872423945043e-06,
-      "loss": 0.0496,
+      "learning_rate": 0.0001248804664723032,
+      "loss": 0.0346,
       "step": 20530
     },
     {
       "epoch": 29.51,
-      "learning_rate": 5.122669283611383e-06,
-      "loss": 0.0565,
+      "learning_rate": 0.0001247930029154519,
+      "loss": 0.0439,
       "step": 20540
     },
     {
       "epoch": 29.53,
-      "learning_rate": 4.975466143277722e-06,
-      "loss": 0.0496,
+      "learning_rate": 0.00012470553935860058,
+      "loss": 0.0402,
       "step": 20550
     },
     {
       "epoch": 29.54,
-      "learning_rate": 4.8282630029440625e-06,
-      "loss": 0.0423,
+      "learning_rate": 0.00012461807580174927,
+      "loss": 0.0325,
       "step": 20560
     },
     {
       "epoch": 29.55,
-      "learning_rate": 4.681059862610402e-06,
-      "loss": 0.0462,
+      "learning_rate": 0.00012453061224489793,
+      "loss": 0.0358,
       "step": 20570
     },
     {
       "epoch": 29.57,
-      "learning_rate": 4.533856722276741e-06,
-      "loss": 0.0471,
+      "learning_rate": 0.00012444314868804664,
+      "loss": 0.0349,
       "step": 20580
     },
     {
       "epoch": 29.58,
-      "learning_rate": 4.386653581943081e-06,
-      "loss": 0.0482,
+      "learning_rate": 0.00012435568513119533,
+      "loss": 0.0391,
       "step": 20590
     },
     {
       "epoch": 29.6,
-      "learning_rate": 4.239450441609421e-06,
-      "loss": 0.0441,
+      "learning_rate": 0.00012426822157434402,
+      "loss": 0.0339,
       "step": 20600
     },
     {
       "epoch": 29.6,
-      "eval_loss": 0.14090383052825928,
-      "eval_runtime": 40.3963,
-      "eval_samples_per_second": 11.313,
-      "eval_steps_per_second": 1.436,
-      "eval_wer": 0.14711639813281133,
+      "eval_loss": 0.13588006794452667,
+      "eval_runtime": 40.1534,
+      "eval_samples_per_second": 11.381,
+      "eval_steps_per_second": 1.444,
+      "eval_wer": 0.13311248305978016,
       "step": 20600
     },
     {
       "epoch": 29.61,
-      "learning_rate": 4.09224730127576e-06,
-      "loss": 0.05,
+      "learning_rate": 0.00012418075801749268,
+      "loss": 0.0418,
       "step": 20610
     },
     {
       "epoch": 29.63,
-      "learning_rate": 3.9450441609421e-06,
-      "loss": 0.0457,
+      "learning_rate": 0.0001240932944606414,
+      "loss": 0.0374,
       "step": 20620
     },
     {
       "epoch": 29.64,
-      "learning_rate": 3.7978410206084395e-06,
-      "loss": 0.0535,
+      "learning_rate": 0.00012400583090379008,
+      "loss": 0.0358,
       "step": 20630
     },
     {
       "epoch": 29.66,
-      "learning_rate": 3.6506378802747788e-06,
-      "loss": 0.0479,
+      "learning_rate": 0.00012391836734693877,
+      "loss": 0.0438,
       "step": 20640
     },
     {
       "epoch": 29.67,
-      "learning_rate": 3.5034347399411185e-06,
-      "loss": 0.0386,
+      "learning_rate": 0.00012383090379008746,
+      "loss": 0.0319,
       "step": 20650
     },
     {
       "epoch": 29.68,
-      "learning_rate": 3.3562315996074583e-06,
-      "loss": 0.0382,
+      "learning_rate": 0.00012374344023323615,
+      "loss": 0.0266,
       "step": 20660
     },
     {
       "epoch": 29.7,
-      "learning_rate": 3.2090284592737976e-06,
-      "loss": 0.0424,
+      "learning_rate": 0.00012365597667638484,
+      "loss": 0.035,
       "step": 20670
     },
     {
       "epoch": 29.71,
-      "learning_rate": 3.0618253189401373e-06,
-      "loss": 0.0628,
+      "learning_rate": 0.00012356851311953352,
+      "loss": 0.045,
       "step": 20680
     },
     {
       "epoch": 29.73,
-      "learning_rate": 2.9146221786064766e-06,
-      "loss": 0.0585,
+      "learning_rate": 0.0001234810495626822,
+      "loss": 0.0471,
       "step": 20690
     },
     {
       "epoch": 29.74,
-      "learning_rate": 2.767419038272816e-06,
-      "loss": 0.0542,
+      "learning_rate": 0.0001233935860058309,
+      "loss": 0.0374,
       "step": 20700
     },
     {
       "epoch": 29.74,
-      "eval_loss": 0.14069251716136932,
-      "eval_runtime": 40.3961,
-      "eval_samples_per_second": 11.313,
-      "eval_steps_per_second": 1.436,
-      "eval_wer": 0.14696581840084325,
+      "eval_loss": 0.13162796199321747,
+      "eval_runtime": 40.4509,
+      "eval_samples_per_second": 11.298,
+      "eval_steps_per_second": 1.434,
+      "eval_wer": 0.12995030868845053,
       "step": 20700
     },
     {
       "epoch": 29.76,
-      "learning_rate": 2.6202158979391557e-06,
-      "loss": 0.0418,
+      "learning_rate": 0.0001233061224489796,
+      "loss": 0.0285,
       "step": 20710
     },
     {
       "epoch": 29.77,
-      "learning_rate": 2.4730127576054954e-06,
-      "loss": 0.0415,
+      "learning_rate": 0.00012321865889212827,
+      "loss": 0.0359,
       "step": 20720
     },
     {
       "epoch": 29.78,
-      "learning_rate": 2.3258096172718348e-06,
-      "loss": 0.0373,
+      "learning_rate": 0.00012313119533527696,
+      "loss": 0.0284,
       "step": 20730
     },
     {
       "epoch": 29.8,
-      "learning_rate": 2.1786064769381745e-06,
-      "loss": 0.0381,
+      "learning_rate": 0.00012304373177842565,
+      "loss": 0.0243,
       "step": 20740
     },
     {
       "epoch": 29.81,
-      "learning_rate": 2.0314033366045143e-06,
-      "loss": 0.0445,
+      "learning_rate": 0.00012295626822157434,
+      "loss": 0.0306,
       "step": 20750
     },
     {
       "epoch": 29.83,
-      "learning_rate": 1.8842001962708538e-06,
-      "loss": 0.0482,
+      "learning_rate": 0.00012286880466472303,
+      "loss": 0.038,
       "step": 20760
     },
     {
       "epoch": 29.84,
-      "learning_rate": 1.7369970559371933e-06,
-      "loss": 0.0463,
+      "learning_rate": 0.00012278134110787171,
+      "loss": 0.0338,
       "step": 20770
     },
     {
       "epoch": 29.86,
-      "learning_rate": 1.5897939156035326e-06,
-      "loss": 0.0402,
+      "learning_rate": 0.0001226938775510204,
+      "loss": 0.0304,
       "step": 20780
     },
     {
       "epoch": 29.87,
-      "learning_rate": 1.4425907752698722e-06,
-      "loss": 0.0503,
+      "learning_rate": 0.0001226064139941691,
+      "loss": 0.0361,
       "step": 20790
     },
     {
       "epoch": 29.89,
-      "learning_rate": 1.295387634936212e-06,
-      "loss": 0.0471,
+      "learning_rate": 0.00012251895043731778,
+      "loss": 0.0358,
       "step": 20800
     },
     {
       "epoch": 29.89,
-      "eval_loss": 0.1404000222682953,
-      "eval_runtime": 41.0194,
-      "eval_samples_per_second": 11.141,
-      "eval_steps_per_second": 1.414,
-      "eval_wer": 0.14711639813281133,
+      "eval_loss": 0.13410452008247375,
+      "eval_runtime": 40.2318,
+      "eval_samples_per_second": 11.359,
+      "eval_steps_per_second": 1.442,
+      "eval_wer": 0.13507001957536516,
       "step": 20800
+    },
+    {
+      "epoch": 29.9,
+      "learning_rate": 0.00012243148688046647,
+      "loss": 0.0341,
+      "step": 20810
+    },
+    {
+      "epoch": 29.91,
+      "learning_rate": 0.00012234402332361515,
+      "loss": 0.0434,
+      "step": 20820
+    },
+    {
+      "epoch": 29.93,
+      "learning_rate": 0.00012225655976676384,
+      "loss": 0.0317,
+      "step": 20830
+    },
+    {
+      "epoch": 29.94,
+      "learning_rate": 0.00012216909620991253,
+      "loss": 0.0415,
+      "step": 20840
+    },
+    {
+      "epoch": 29.96,
+      "learning_rate": 0.00012208163265306122,
+      "loss": 0.0355,
+      "step": 20850
+    },
+    {
+      "epoch": 29.97,
+      "learning_rate": 0.0001219941690962099,
+      "loss": 0.0376,
+      "step": 20860
+    },
+    {
+      "epoch": 29.99,
+      "learning_rate": 0.0001219067055393586,
+      "loss": 0.0373,
+      "step": 20870
+    },
+    {
+      "epoch": 30.0,
+      "learning_rate": 0.00012181924198250727,
+      "loss": 0.0283,
+      "step": 20880
+    },
+    {
+      "epoch": 30.01,
+      "learning_rate": 0.00012173177842565597,
+      "loss": 0.0411,
+      "step": 20890
+    },
+    {
+      "epoch": 30.03,
+      "learning_rate": 0.00012164431486880466,
+      "loss": 0.0336,
+      "step": 20900
+    },
+    {
+      "epoch": 30.03,
+      "eval_loss": 0.1340400129556656,
+      "eval_runtime": 40.2902,
+      "eval_samples_per_second": 11.343,
+      "eval_steps_per_second": 1.44,
+      "eval_wer": 0.13205842493600362,
+      "step": 20900
+    },
+    {
+      "epoch": 30.04,
+      "learning_rate": 0.00012155685131195334,
+      "loss": 0.0405,
+      "step": 20910
+    },
+    {
+      "epoch": 30.06,
+      "learning_rate": 0.00012146938775510203,
+      "loss": 0.0406,
+      "step": 20920
+    },
+    {
+      "epoch": 30.07,
+      "learning_rate": 0.00012138192419825071,
+      "loss": 0.0304,
+      "step": 20930
+    },
+    {
+      "epoch": 30.09,
+      "learning_rate": 0.00012129446064139941,
+      "loss": 0.0366,
+      "step": 20940
+    },
+    {
+      "epoch": 30.1,
+      "learning_rate": 0.0001212069970845481,
+      "loss": 0.0356,
+      "step": 20950
+    },
+    {
+      "epoch": 30.11,
+      "learning_rate": 0.00012111953352769678,
+      "loss": 0.0369,
+      "step": 20960
+    },
+    {
+      "epoch": 30.13,
+      "learning_rate": 0.00012103206997084546,
+      "loss": 0.034,
+      "step": 20970
+    },
+    {
+      "epoch": 30.14,
+      "learning_rate": 0.00012094460641399416,
+      "loss": 0.0273,
+      "step": 20980
+    },
+    {
+      "epoch": 30.16,
+      "learning_rate": 0.00012085714285714285,
+      "loss": 0.0409,
+      "step": 20990
+    },
+    {
+      "epoch": 30.17,
+      "learning_rate": 0.00012076967930029154,
+      "loss": 0.0393,
+      "step": 21000
+    },
+    {
+      "epoch": 30.17,
+      "eval_loss": 0.13167157769203186,
+      "eval_runtime": 40.4932,
+      "eval_samples_per_second": 11.286,
+      "eval_steps_per_second": 1.432,
+      "eval_wer": 0.13461828037946091,
+      "step": 21000
+    },
+    {
+      "epoch": 30.19,
+      "learning_rate": 0.00012068221574344022,
+      "loss": 0.0313,
+      "step": 21010
+    },
+    {
+      "epoch": 30.2,
+      "learning_rate": 0.00012059475218658891,
+      "loss": 0.0302,
+      "step": 21020
+    },
+    {
+      "epoch": 30.22,
+      "learning_rate": 0.0001205072886297376,
+      "loss": 0.0266,
+      "step": 21030
+    },
+    {
+      "epoch": 30.23,
+      "learning_rate": 0.00012041982507288629,
+      "loss": 0.0415,
+      "step": 21040
+    },
+    {
+      "epoch": 30.24,
+      "learning_rate": 0.00012033236151603498,
+      "loss": 0.0286,
+      "step": 21050
+    },
+    {
+      "epoch": 30.26,
+      "learning_rate": 0.00012024489795918365,
+      "loss": 0.0344,
+      "step": 21060
+    },
+    {
+      "epoch": 30.27,
+      "learning_rate": 0.00012015743440233235,
+      "loss": 0.0317,
+      "step": 21070
+    },
+    {
+      "epoch": 30.29,
+      "learning_rate": 0.00012006997084548104,
+      "loss": 0.0271,
+      "step": 21080
+    },
+    {
+      "epoch": 30.3,
+      "learning_rate": 0.00011998250728862973,
+      "loss": 0.0418,
+      "step": 21090
+    },
+    {
+      "epoch": 30.32,
+      "learning_rate": 0.00011989504373177843,
+      "loss": 0.0383,
+      "step": 21100
+    },
+    {
+      "epoch": 30.32,
+      "eval_loss": 0.12635038793087006,
+      "eval_runtime": 40.1747,
+      "eval_samples_per_second": 11.375,
+      "eval_steps_per_second": 1.444,
+      "eval_wer": 0.13627465743110978,
+      "step": 21100
+    },
+    {
+      "epoch": 30.33,
+      "learning_rate": 0.0001198075801749271,
+      "loss": 0.0352,
+      "step": 21110
+    },
+    {
+      "epoch": 30.34,
+      "learning_rate": 0.00011972011661807579,
+      "loss": 0.0324,
+      "step": 21120
+    },
+    {
+      "epoch": 30.36,
+      "learning_rate": 0.00011963265306122448,
+      "loss": 0.0326,
+      "step": 21130
+    },
+    {
+      "epoch": 30.37,
+      "learning_rate": 0.00011954518950437318,
+      "loss": 0.0368,
+      "step": 21140
+    },
+    {
+      "epoch": 30.39,
+      "learning_rate": 0.00011945772594752185,
+      "loss": 0.0299,
+      "step": 21150
+    },
+    {
+      "epoch": 30.4,
+      "learning_rate": 0.00011937026239067054,
+      "loss": 0.0316,
+      "step": 21160
+    },
+    {
+      "epoch": 30.42,
+      "learning_rate": 0.00011928279883381923,
+      "loss": 0.0304,
+      "step": 21170
+    },
+    {
+      "epoch": 30.43,
+      "learning_rate": 0.00011919533527696792,
+      "loss": 0.0307,
+      "step": 21180
+    },
+    {
+      "epoch": 30.45,
+      "learning_rate": 0.00011910787172011662,
+      "loss": 0.0369,
+      "step": 21190
+    },
+    {
+      "epoch": 30.46,
+      "learning_rate": 0.0001190204081632653,
+      "loss": 0.0353,
+      "step": 21200
+    },
+    {
+      "epoch": 30.46,
+      "eval_loss": 0.1339302957057953,
+      "eval_runtime": 40.1253,
+      "eval_samples_per_second": 11.389,
+      "eval_steps_per_second": 1.445,
+      "eval_wer": 0.13642523716307786,
+      "step": 21200
+    },
+    {
+      "epoch": 30.47,
+      "learning_rate": 0.00011893294460641398,
+      "loss": 0.039,
+      "step": 21210
+    },
+    {
+      "epoch": 30.49,
+      "learning_rate": 0.00011884548104956267,
+      "loss": 0.0304,
+      "step": 21220
+    },
+    {
+      "epoch": 30.5,
+      "learning_rate": 0.00011875801749271137,
+      "loss": 0.0294,
+      "step": 21230
+    },
+    {
+      "epoch": 30.52,
+      "learning_rate": 0.00011867055393586005,
+      "loss": 0.0529,
+      "step": 21240
+    },
+    {
+      "epoch": 30.53,
+      "learning_rate": 0.00011858309037900873,
+      "loss": 0.028,
+      "step": 21250
+    },
+    {
+      "epoch": 30.55,
+      "learning_rate": 0.00011849562682215742,
+      "loss": 0.0378,
+      "step": 21260
+    },
+    {
+      "epoch": 30.56,
+      "learning_rate": 0.00011840816326530612,
+      "loss": 0.0351,
+      "step": 21270
+    },
+    {
+      "epoch": 30.57,
+      "learning_rate": 0.00011832069970845481,
+      "loss": 0.0325,
+      "step": 21280
+    },
+    {
+      "epoch": 30.59,
+      "learning_rate": 0.00011823323615160348,
+      "loss": 0.0425,
+      "step": 21290
+    },
+    {
+      "epoch": 30.6,
+      "learning_rate": 0.00011814577259475217,
+      "loss": 0.0254,
+      "step": 21300
+    },
+    {
+      "epoch": 30.6,
+      "eval_loss": 0.13099804520606995,
+      "eval_runtime": 40.1721,
+      "eval_samples_per_second": 11.376,
+      "eval_steps_per_second": 1.444,
+      "eval_wer": 0.13702755609095016,
+      "step": 21300
+    },
+    {
+      "epoch": 30.62,
+      "learning_rate": 0.00011805830903790086,
+      "loss": 0.0342,
+      "step": 21310
+    },
+    {
+      "epoch": 30.63,
+      "learning_rate": 0.00011797084548104956,
+      "loss": 0.044,
+      "step": 21320
+    },
+    {
+      "epoch": 30.65,
+      "learning_rate": 0.00011788338192419824,
+      "loss": 0.0278,
+      "step": 21330
+    },
+    {
+      "epoch": 30.66,
+      "learning_rate": 0.00011779591836734692,
+      "loss": 0.0322,
+      "step": 21340
+    },
+    {
+      "epoch": 30.68,
+      "learning_rate": 0.00011770845481049561,
+      "loss": 0.0468,
+      "step": 21350
+    },
+    {
+      "epoch": 30.69,
+      "learning_rate": 0.00011762099125364431,
+      "loss": 0.0375,
+      "step": 21360
+    },
+    {
+      "epoch": 30.7,
+      "learning_rate": 0.000117533527696793,
+      "loss": 0.0311,
+      "step": 21370
+    },
+    {
+      "epoch": 30.72,
+      "learning_rate": 0.00011744606413994168,
+      "loss": 0.026,
+      "step": 21380
+    },
+    {
+      "epoch": 30.73,
+      "learning_rate": 0.00011735860058309036,
+      "loss": 0.0439,
+      "step": 21390
+    },
+    {
+      "epoch": 30.75,
+      "learning_rate": 0.00011727113702623907,
+      "loss": 0.0341,
+      "step": 21400
+    },
+    {
+      "epoch": 30.75,
+      "eval_loss": 0.1271417737007141,
+      "eval_runtime": 40.1531,
+      "eval_samples_per_second": 11.381,
+      "eval_steps_per_second": 1.444,
+      "eval_wer": 0.13883451287456708,
+      "step": 21400
+    },
+    {
+      "epoch": 30.76,
+      "learning_rate": 0.00011718367346938775,
+      "loss": 0.044,
+      "step": 21410
+    },
+    {
+      "epoch": 30.78,
+      "learning_rate": 0.00011709620991253643,
+      "loss": 0.0266,
+      "step": 21420
+    },
+    {
+      "epoch": 30.79,
+      "learning_rate": 0.00011700874635568512,
+      "loss": 0.0301,
+      "step": 21430
+    },
+    {
+      "epoch": 30.8,
+      "learning_rate": 0.00011693002915451893,
+      "loss": 0.0308,
+      "step": 21440
+    },
+    {
+      "epoch": 30.82,
+      "learning_rate": 0.00011684256559766762,
+      "loss": 0.0251,
+      "step": 21450
+    },
+    {
+      "epoch": 30.83,
+      "learning_rate": 0.00011675510204081632,
+      "loss": 0.0336,
+      "step": 21460
+    },
+    {
+      "epoch": 30.85,
+      "learning_rate": 0.00011666763848396501,
+      "loss": 0.0348,
+      "step": 21470
+    },
+    {
+      "epoch": 30.86,
+      "learning_rate": 0.00011658017492711369,
+      "loss": 0.0274,
+      "step": 21480
+    },
+    {
+      "epoch": 30.88,
+      "learning_rate": 0.00011649271137026237,
+      "loss": 0.0361,
+      "step": 21490
+    },
+    {
+      "epoch": 30.89,
+      "learning_rate": 0.00011640524781341106,
+      "loss": 0.0337,
+      "step": 21500
+    },
+    {
+      "epoch": 30.89,
+      "eval_loss": 0.1293652504682541,
+      "eval_runtime": 40.341,
+      "eval_samples_per_second": 11.328,
+      "eval_steps_per_second": 1.438,
+      "eval_wer": 0.1380816142147267,
+      "step": 21500
+    },
+    {
+      "epoch": 30.91,
+      "learning_rate": 0.00011631778425655976,
+      "loss": 0.0439,
+      "step": 21510
+    },
+    {
+      "epoch": 30.92,
+      "learning_rate": 0.00011623032069970845,
+      "loss": 0.0347,
+      "step": 21520
+    },
+    {
+      "epoch": 30.93,
+      "learning_rate": 0.00011614285714285713,
+      "loss": 0.0281,
+      "step": 21530
+    },
+    {
+      "epoch": 30.95,
+      "learning_rate": 0.00011605539358600581,
+      "loss": 0.0493,
+      "step": 21540
+    },
+    {
+      "epoch": 30.96,
+      "learning_rate": 0.00011596793002915452,
+      "loss": 0.0321,
+      "step": 21550
+    },
+    {
+      "epoch": 30.98,
+      "learning_rate": 0.0001158804664723032,
+      "loss": 0.0409,
+      "step": 21560
+    },
+    {
+      "epoch": 30.99,
+      "learning_rate": 0.00011579300291545188,
+      "loss": 0.048,
+      "step": 21570
+    },
+    {
+      "epoch": 31.01,
+      "learning_rate": 0.00011570553935860057,
+      "loss": 0.0271,
+      "step": 21580
+    },
+    {
+      "epoch": 31.02,
+      "learning_rate": 0.00011561807580174927,
+      "loss": 0.0318,
+      "step": 21590
+    },
+    {
+      "epoch": 31.03,
+      "learning_rate": 0.00011553061224489795,
+      "loss": 0.0261,
+      "step": 21600
+    },
+    {
+      "epoch": 31.03,
+      "eval_loss": 0.1307568997144699,
+      "eval_runtime": 40.326,
+      "eval_samples_per_second": 11.333,
+      "eval_steps_per_second": 1.438,
+      "eval_wer": 0.13552175877126937,
+      "step": 21600
+    },
+    {
+      "epoch": 31.05,
+      "learning_rate": 0.00011544314868804664,
+      "loss": 0.0325,
+      "step": 21610
+    },
+    {
+      "epoch": 31.06,
+      "learning_rate": 0.00011535568513119532,
+      "loss": 0.0298,
+      "step": 21620
+    },
+    {
+      "epoch": 31.08,
+      "learning_rate": 0.000115268221574344,
+      "loss": 0.0292,
+      "step": 21630
+    },
+    {
+      "epoch": 31.09,
+      "learning_rate": 0.0001151807580174927,
+      "loss": 0.0315,
+      "step": 21640
+    },
+    {
+      "epoch": 31.11,
+      "learning_rate": 0.0001150932944606414,
+      "loss": 0.0319,
+      "step": 21650
+    },
+    {
+      "epoch": 31.12,
+      "learning_rate": 0.00011500583090379007,
+      "loss": 0.0301,
+      "step": 21660
+    },
+    {
+      "epoch": 31.14,
+      "learning_rate": 0.00011491836734693876,
+      "loss": 0.0399,
+      "step": 21670
+    },
+    {
+      "epoch": 31.15,
+      "learning_rate": 0.00011483090379008746,
+      "loss": 0.033,
+      "step": 21680
+    },
+    {
+      "epoch": 31.16,
+      "learning_rate": 0.00011474344023323615,
+      "loss": 0.0348,
+      "step": 21690
+    },
+    {
+      "epoch": 31.18,
+      "learning_rate": 0.00011465597667638483,
+      "loss": 0.0276,
+      "step": 21700
+    },
+    {
+      "epoch": 31.18,
+      "eval_loss": 0.12548431754112244,
+      "eval_runtime": 40.1766,
+      "eval_samples_per_second": 11.375,
+      "eval_steps_per_second": 1.444,
+      "eval_wer": 0.13793103448275862,
+      "step": 21700
+    },
+    {
+      "epoch": 31.19,
+      "learning_rate": 0.00011456851311953351,
+      "loss": 0.0376,
+      "step": 21710
+    },
+    {
+      "epoch": 31.21,
+      "learning_rate": 0.00011448104956268221,
+      "loss": 0.0291,
+      "step": 21720
+    },
+    {
+      "epoch": 31.22,
+      "learning_rate": 0.0001143935860058309,
+      "loss": 0.0276,
+      "step": 21730
+    },
+    {
+      "epoch": 31.24,
+      "learning_rate": 0.00011430612244897959,
+      "loss": 0.0343,
+      "step": 21740
+    },
+    {
+      "epoch": 31.25,
+      "learning_rate": 0.00011421865889212826,
+      "loss": 0.0292,
+      "step": 21750
+    },
+    {
+      "epoch": 31.26,
+      "learning_rate": 0.00011413119533527695,
+      "loss": 0.0313,
+      "step": 21760
+    },
+    {
+      "epoch": 31.28,
+      "learning_rate": 0.00011404373177842565,
+      "loss": 0.0306,
+      "step": 21770
+    },
+    {
+      "epoch": 31.29,
+      "learning_rate": 0.00011395626822157434,
+      "loss": 0.0362,
+      "step": 21780
+    },
+    {
+      "epoch": 31.31,
+      "learning_rate": 0.00011386880466472302,
+      "loss": 0.04,
+      "step": 21790
+    },
+    {
+      "epoch": 31.32,
+      "learning_rate": 0.0001137813411078717,
+      "loss": 0.0316,
+      "step": 21800
+    },
+    {
+      "epoch": 31.32,
+      "eval_loss": 0.13308794796466827,
+      "eval_runtime": 40.3702,
+      "eval_samples_per_second": 11.32,
+      "eval_steps_per_second": 1.437,
+      "eval_wer": 0.13823219394669478,
+      "step": 21800
+    },
+    {
+      "epoch": 31.34,
+      "learning_rate": 0.0001136938775510204,
+      "loss": 0.0391,
+      "step": 21810
+    },
+    {
+      "epoch": 31.35,
+      "learning_rate": 0.00011360641399416909,
+      "loss": 0.032,
+      "step": 21820
+    },
+    {
+      "epoch": 31.36,
+      "learning_rate": 0.00011351895043731778,
+      "loss": 0.0261,
+      "step": 21830
+    },
+    {
+      "epoch": 31.38,
+      "learning_rate": 0.00011343148688046645,
+      "loss": 0.0276,
+      "step": 21840
+    },
+    {
+      "epoch": 31.39,
+      "learning_rate": 0.00011334402332361515,
+      "loss": 0.0348,
+      "step": 21850
+    },
+    {
+      "epoch": 31.41,
+      "learning_rate": 0.00011325655976676384,
+      "loss": 0.0352,
+      "step": 21860
+    },
+    {
+      "epoch": 31.42,
+      "learning_rate": 0.00011316909620991253,
+      "loss": 0.0249,
+      "step": 21870
+    },
+    {
+      "epoch": 31.44,
+      "learning_rate": 0.00011308163265306122,
+      "loss": 0.0339,
+      "step": 21880
+    },
+    {
+      "epoch": 31.45,
+      "learning_rate": 0.00011299416909620989,
+      "loss": 0.0358,
+      "step": 21890
+    },
+    {
+      "epoch": 31.47,
+      "learning_rate": 0.00011290670553935859,
+      "loss": 0.0274,
+      "step": 21900
+    },
+    {
+      "epoch": 31.47,
+      "eval_loss": 0.13432462513446808,
+      "eval_runtime": 40.5527,
+      "eval_samples_per_second": 11.269,
+      "eval_steps_per_second": 1.43,
+      "eval_wer": 0.13311248305978016,
+      "step": 21900
+    },
+    {
+      "epoch": 31.48,
+      "learning_rate": 0.00011281924198250728,
+      "loss": 0.0294,
+      "step": 21910
+    },
+    {
+      "epoch": 31.49,
+      "learning_rate": 0.00011273177842565597,
+      "loss": 0.0273,
+      "step": 21920
+    },
+    {
+      "epoch": 31.51,
+      "learning_rate": 0.00011264431486880464,
+      "loss": 0.0275,
+      "step": 21930
+    },
+    {
+      "epoch": 31.52,
+      "learning_rate": 0.00011255685131195334,
+      "loss": 0.0301,
+      "step": 21940
+    },
+    {
+      "epoch": 31.54,
+      "learning_rate": 0.00011246938775510203,
+      "loss": 0.0265,
+      "step": 21950
+    },
+    {
+      "epoch": 31.55,
+      "learning_rate": 0.00011238192419825072,
+      "loss": 0.0343,
+      "step": 21960
+    },
+    {
+      "epoch": 31.57,
+      "learning_rate": 0.00011229446064139942,
+      "loss": 0.0278,
+      "step": 21970
+    },
+    {
+      "epoch": 31.58,
+      "learning_rate": 0.0001122069970845481,
+      "loss": 0.0347,
+      "step": 21980
+    },
+    {
+      "epoch": 31.59,
+      "learning_rate": 0.00011211953352769678,
+      "loss": 0.0329,
+      "step": 21990
+    },
+    {
+      "epoch": 31.61,
+      "learning_rate": 0.00011203206997084547,
+      "loss": 0.0317,
+      "step": 22000
+    },
+    {
+      "epoch": 31.61,
+      "eval_loss": 0.129713773727417,
+      "eval_runtime": 40.2564,
+      "eval_samples_per_second": 11.352,
+      "eval_steps_per_second": 1.441,
+      "eval_wer": 0.13341364252371632,
+      "step": 22000
+    },
+    {
+      "epoch": 31.62,
+      "learning_rate": 0.00011194460641399416,
+      "loss": 0.0409,
+      "step": 22010
+    },
+    {
+      "epoch": 31.64,
+      "learning_rate": 0.00011185714285714285,
+      "loss": 0.0268,
+      "step": 22020
+    },
+    {
+      "epoch": 31.65,
+      "learning_rate": 0.00011176967930029153,
+      "loss": 0.0461,
+      "step": 22030
+    },
+    {
+      "epoch": 31.67,
+      "learning_rate": 0.00011168221574344022,
+      "loss": 0.0362,
+      "step": 22040
+    },
+    {
+      "epoch": 31.68,
+      "learning_rate": 0.00011159475218658891,
+      "loss": 0.0282,
+      "step": 22050
+    },
+    {
+      "epoch": 31.7,
+      "learning_rate": 0.00011150728862973761,
+      "loss": 0.0447,
+      "step": 22060
+    },
+    {
+      "epoch": 31.71,
+      "learning_rate": 0.00011141982507288629,
+      "loss": 0.0306,
+      "step": 22070
+    },
+    {
+      "epoch": 31.72,
+      "learning_rate": 0.00011133236151603497,
+      "loss": 0.0367,
+      "step": 22080
+    },
+    {
+      "epoch": 31.74,
+      "learning_rate": 0.00011124489795918366,
+      "loss": 0.0244,
+      "step": 22090
+    },
+    {
+      "epoch": 31.75,
+      "learning_rate": 0.00011115743440233236,
+      "loss": 0.029,
+      "step": 22100
+    },
+    {
+      "epoch": 31.75,
+      "eval_loss": 0.1280178278684616,
+      "eval_runtime": 40.1733,
+      "eval_samples_per_second": 11.376,
+      "eval_steps_per_second": 1.444,
+      "eval_wer": 0.13130552627616324,
+      "step": 22100
+    },
+    {
+      "epoch": 31.77,
+      "learning_rate": 0.00011106997084548104,
+      "loss": 0.0368,
+      "step": 22110
+    },
+    {
+      "epoch": 31.78,
+      "learning_rate": 0.00011098250728862973,
+      "loss": 0.0244,
+      "step": 22120
+    },
+    {
+      "epoch": 31.8,
+      "learning_rate": 0.00011089504373177841,
+      "loss": 0.0295,
+      "step": 22130
+    },
+    {
+      "epoch": 31.81,
+      "learning_rate": 0.0001108075801749271,
+      "loss": 0.0315,
+      "step": 22140
+    },
+    {
+      "epoch": 31.82,
+      "learning_rate": 0.0001107201166180758,
+      "loss": 0.0213,
+      "step": 22150
+    },
+    {
+      "epoch": 31.84,
+      "learning_rate": 0.00011063265306122448,
+      "loss": 0.0409,
+      "step": 22160
+    },
+    {
+      "epoch": 31.85,
+      "learning_rate": 0.00011054518950437316,
+      "loss": 0.0407,
+      "step": 22170
+    },
+    {
+      "epoch": 31.87,
+      "learning_rate": 0.00011045772594752185,
+      "loss": 0.0375,
+      "step": 22180
+    },
+    {
+      "epoch": 31.88,
+      "learning_rate": 0.00011037026239067055,
+      "loss": 0.0301,
+      "step": 22190
+    },
+    {
+      "epoch": 31.9,
+      "learning_rate": 0.00011028279883381923,
+      "loss": 0.0243,
+      "step": 22200
+    },
+    {
+      "epoch": 31.9,
+      "eval_loss": 0.13117167353630066,
+      "eval_runtime": 40.1562,
+      "eval_samples_per_second": 11.381,
+      "eval_steps_per_second": 1.444,
+      "eval_wer": 0.13130552627616324,
+      "step": 22200
+    },
+    {
+      "epoch": 31.91,
+      "learning_rate": 0.00011019533527696792,
+      "loss": 0.0305,
+      "step": 22210
+    },
+    {
+      "epoch": 31.93,
+      "learning_rate": 0.0001101078717201166,
+      "loss": 0.0302,
+      "step": 22220
+    },
+    {
+      "epoch": 31.94,
+      "learning_rate": 0.0001100204081632653,
+      "loss": 0.0364,
+      "step": 22230
+    },
+    {
+      "epoch": 31.95,
+      "learning_rate": 0.000109932944606414,
+      "loss": 0.0331,
+      "step": 22240
+    },
+    {
+      "epoch": 31.97,
+      "learning_rate": 0.00010984548104956267,
+      "loss": 0.0286,
+      "step": 22250
+    },
+    {
+      "epoch": 31.98,
+      "learning_rate": 0.00010975801749271136,
+      "loss": 0.0303,
+      "step": 22260
+    },
+    {
+      "epoch": 32.0,
+      "learning_rate": 0.00010967055393586006,
+      "loss": 0.0318,
+      "step": 22270
+    },
+    {
+      "epoch": 32.01,
+      "learning_rate": 0.00010958309037900875,
+      "loss": 0.0431,
+      "step": 22280
+    },
+    {
+      "epoch": 32.03,
+      "learning_rate": 0.00010949562682215742,
+      "loss": 0.0271,
+      "step": 22290
+    },
+    {
+      "epoch": 32.04,
+      "learning_rate": 0.00010940816326530611,
+      "loss": 0.0298,
+      "step": 22300
+    },
+    {
+      "epoch": 32.04,
+      "eval_loss": 0.1335451751947403,
+      "eval_runtime": 40.2607,
+      "eval_samples_per_second": 11.351,
+      "eval_steps_per_second": 1.441,
+      "eval_wer": 0.13251016413190783,
+      "step": 22300
+    },
+    {
+      "epoch": 32.05,
+      "learning_rate": 0.0001093206997084548,
+      "loss": 0.0277,
+      "step": 22310
+    },
+    {
+      "epoch": 32.07,
+      "learning_rate": 0.0001092332361516035,
+      "loss": 0.0312,
+      "step": 22320
+    },
+    {
+      "epoch": 32.08,
+      "learning_rate": 0.00010914577259475218,
+      "loss": 0.0444,
+      "step": 22330
+    },
+    {
+      "epoch": 32.1,
+      "learning_rate": 0.00010905830903790086,
+      "loss": 0.0323,
+      "step": 22340
+    },
+    {
+      "epoch": 32.11,
+      "learning_rate": 0.00010897084548104955,
+      "loss": 0.0309,
+      "step": 22350
+    },
+    {
+      "epoch": 32.13,
+      "learning_rate": 0.00010888338192419825,
+      "loss": 0.0302,
+      "step": 22360
+    },
+    {
+      "epoch": 32.14,
+      "learning_rate": 0.00010879591836734694,
+      "loss": 0.0257,
+      "step": 22370
+    },
+    {
+      "epoch": 32.16,
+      "learning_rate": 0.00010870845481049561,
+      "loss": 0.0451,
+      "step": 22380
+    },
+    {
+      "epoch": 32.17,
+      "learning_rate": 0.0001086209912536443,
+      "loss": 0.0345,
+      "step": 22390
+    },
+    {
+      "epoch": 32.18,
+      "learning_rate": 0.000108533527696793,
+      "loss": 0.0251,
+      "step": 22400
+    },
+    {
+      "epoch": 32.18,
+      "eval_loss": 0.133877694606781,
+      "eval_runtime": 40.4225,
+      "eval_samples_per_second": 11.306,
+      "eval_steps_per_second": 1.435,
+      "eval_wer": 0.134768860111429,
+      "step": 22400
+    },
+    {
+      "epoch": 32.2,
+      "learning_rate": 0.00010844606413994169,
+      "loss": 0.0355,
+      "step": 22410
+    },
+    {
+      "epoch": 32.21,
+      "learning_rate": 0.00010835860058309038,
+      "loss": 0.0202,
+      "step": 22420
+    },
+    {
+      "epoch": 32.23,
+      "learning_rate": 0.00010827113702623905,
+      "loss": 0.0383,
+      "step": 22430
+    },
+    {
+      "epoch": 32.24,
+      "learning_rate": 0.00010818367346938774,
+      "loss": 0.0391,
+      "step": 22440
+    },
+    {
+      "epoch": 32.26,
+      "learning_rate": 0.00010809620991253644,
+      "loss": 0.0356,
+      "step": 22450
+    },
+    {
+      "epoch": 32.27,
+      "learning_rate": 0.00010800874635568513,
+      "loss": 0.0307,
+      "step": 22460
+    },
+    {
+      "epoch": 32.28,
+      "learning_rate": 0.0001079212827988338,
+      "loss": 0.0267,
+      "step": 22470
+    },
+    {
+      "epoch": 32.3,
+      "learning_rate": 0.00010783381924198249,
+      "loss": 0.0341,
+      "step": 22480
+    },
+    {
+      "epoch": 32.31,
+      "learning_rate": 0.00010774635568513119,
+      "loss": 0.026,
+      "step": 22490
+    },
+    {
+      "epoch": 32.33,
+      "learning_rate": 0.00010765889212827988,
+      "loss": 0.0287,
+      "step": 22500
+    },
+    {
+      "epoch": 32.33,
+      "eval_loss": 0.14132900536060333,
+      "eval_runtime": 40.4253,
+      "eval_samples_per_second": 11.305,
+      "eval_steps_per_second": 1.435,
+      "eval_wer": 0.13371480198765245,
+      "step": 22500
+    },
+    {
+      "epoch": 32.34,
+      "learning_rate": 0.00010757142857142857,
+      "loss": 0.0334,
+      "step": 22510
+    },
+    {
+      "epoch": 32.36,
+      "learning_rate": 0.00010748396501457724,
+      "loss": 0.0309,
+      "step": 22520
+    },
+    {
+      "epoch": 32.37,
+      "learning_rate": 0.00010739650145772594,
+      "loss": 0.0347,
+      "step": 22530
+    },
+    {
+      "epoch": 32.39,
+      "learning_rate": 0.00010730903790087463,
+      "loss": 0.0392,
+      "step": 22540
+    },
+    {
+      "epoch": 32.4,
+      "learning_rate": 0.00010722157434402332,
+      "loss": 0.0311,
+      "step": 22550
+    },
+    {
+      "epoch": 32.41,
+      "learning_rate": 0.00010713411078717199,
+      "loss": 0.0363,
+      "step": 22560
+    },
+    {
+      "epoch": 32.43,
+      "learning_rate": 0.00010704664723032068,
+      "loss": 0.0336,
+      "step": 22570
+    },
+    {
+      "epoch": 32.44,
+      "learning_rate": 0.00010695918367346938,
+      "loss": 0.0425,
+      "step": 22580
+    },
+    {
+      "epoch": 32.46,
+      "learning_rate": 0.00010687172011661807,
+      "loss": 0.0257,
+      "step": 22590
+    },
+    {
+      "epoch": 32.47,
+      "learning_rate": 0.00010678425655976676,
+      "loss": 0.0322,
+      "step": 22600
+    },
+    {
+      "epoch": 32.47,
+      "eval_loss": 0.1421317458152771,
+      "eval_runtime": 40.1693,
+      "eval_samples_per_second": 11.377,
+      "eval_steps_per_second": 1.444,
+      "eval_wer": 0.13461828037946091,
+      "step": 22600
+    },
+    {
+      "epoch": 32.49,
+      "learning_rate": 0.00010669679300291543,
+      "loss": 0.0385,
+      "step": 22610
+    },
+    {
+      "epoch": 32.5,
+      "learning_rate": 0.00010660932944606413,
+      "loss": 0.0222,
+      "step": 22620
+    },
+    {
+      "epoch": 32.51,
+      "learning_rate": 0.00010652186588921282,
+      "loss": 0.0464,
+      "step": 22630
+    },
+    {
+      "epoch": 32.53,
+      "learning_rate": 0.00010643440233236151,
+      "loss": 0.0272,
+      "step": 22640
+    },
+    {
+      "epoch": 32.54,
+      "learning_rate": 0.00010634693877551018,
+      "loss": 0.0287,
+      "step": 22650
+    },
+    {
+      "epoch": 32.56,
+      "learning_rate": 0.00010625947521865889,
+      "loss": 0.0261,
+      "step": 22660
+    },
+    {
+      "epoch": 32.57,
+      "learning_rate": 0.00010617201166180757,
+      "loss": 0.0238,
+      "step": 22670
+    },
+    {
+      "epoch": 32.59,
+      "learning_rate": 0.00010608454810495626,
+      "loss": 0.0369,
+      "step": 22680
+    },
+    {
+      "epoch": 32.6,
+      "learning_rate": 0.00010599708454810495,
+      "loss": 0.0292,
+      "step": 22690
+    },
+    {
+      "epoch": 32.61,
+      "learning_rate": 0.00010590962099125362,
+      "loss": 0.0274,
+      "step": 22700
+    },
+    {
+      "epoch": 32.61,
+      "eval_loss": 0.1342337280511856,
+      "eval_runtime": 40.3469,
+      "eval_samples_per_second": 11.327,
+      "eval_steps_per_second": 1.438,
+      "eval_wer": 0.13145610600813132,
+      "step": 22700
+    },
+    {
+      "epoch": 32.63,
+      "learning_rate": 0.00010582215743440232,
+      "loss": 0.0285,
+      "step": 22710
+    },
+    {
+      "epoch": 32.64,
+      "learning_rate": 0.00010573469387755101,
+      "loss": 0.0391,
+      "step": 22720
+    },
+    {
+      "epoch": 32.66,
+      "learning_rate": 0.0001056472303206997,
+      "loss": 0.0394,
+      "step": 22730
+    },
+    {
+      "epoch": 32.67,
+      "learning_rate": 0.00010555976676384837,
+      "loss": 0.0302,
+      "step": 22740
+    },
+    {
+      "epoch": 32.69,
+      "learning_rate": 0.00010547230320699708,
+      "loss": 0.0268,
+      "step": 22750
+    },
+    {
+      "epoch": 32.7,
+      "learning_rate": 0.00010538483965014576,
+      "loss": 0.034,
+      "step": 22760
+    },
+    {
+      "epoch": 32.72,
+      "learning_rate": 0.00010529737609329445,
+      "loss": 0.0301,
+      "step": 22770
+    },
+    {
+      "epoch": 32.73,
+      "learning_rate": 0.00010520991253644315,
+      "loss": 0.0293,
+      "step": 22780
+    },
+    {
+      "epoch": 32.74,
+      "learning_rate": 0.00010512244897959183,
+      "loss": 0.0378,
+      "step": 22790
+    },
+    {
+      "epoch": 32.76,
+      "learning_rate": 0.00010503498542274052,
+      "loss": 0.0307,
+      "step": 22800
+    },
+    {
+      "epoch": 32.76,
+      "eval_loss": 0.1377181112766266,
+      "eval_runtime": 40.1267,
+      "eval_samples_per_second": 11.389,
+      "eval_steps_per_second": 1.445,
+      "eval_wer": 0.13160668574009937,
+      "step": 22800
+    },
+    {
+      "epoch": 32.77,
+      "learning_rate": 0.0001049475218658892,
+      "loss": 0.034,
+      "step": 22810
+    },
+    {
+      "epoch": 32.79,
+      "learning_rate": 0.00010486005830903789,
+      "loss": 0.0302,
+      "step": 22820
+    },
+    {
+      "epoch": 32.8,
+      "learning_rate": 0.00010477259475218658,
+      "loss": 0.0349,
+      "step": 22830
+    },
+    {
+      "epoch": 32.82,
+      "learning_rate": 0.00010468513119533527,
+      "loss": 0.0373,
+      "step": 22840
+    },
+    {
+      "epoch": 32.83,
+      "learning_rate": 0.00010459766763848396,
+      "loss": 0.0304,
+      "step": 22850
+    },
+    {
+      "epoch": 32.84,
+      "learning_rate": 0.00010451020408163264,
+      "loss": 0.0311,
+      "step": 22860
+    },
+    {
+      "epoch": 32.86,
+      "learning_rate": 0.00010442274052478134,
+      "loss": 0.0352,
+      "step": 22870
+    },
+    {
+      "epoch": 32.87,
+      "learning_rate": 0.00010433527696793002,
+      "loss": 0.0306,
+      "step": 22880
+    },
+    {
+      "epoch": 32.89,
+      "learning_rate": 0.0001042478134110787,
+      "loss": 0.0288,
+      "step": 22890
+    },
+    {
+      "epoch": 32.9,
+      "learning_rate": 0.0001041603498542274,
+      "loss": 0.0297,
+      "step": 22900
+    },
+    {
+      "epoch": 32.9,
+      "eval_loss": 0.13861913979053497,
+      "eval_runtime": 40.2824,
+      "eval_samples_per_second": 11.345,
+      "eval_steps_per_second": 1.44,
+      "eval_wer": 0.1322090046679717,
+      "step": 22900
+    },
+    {
+      "epoch": 32.92,
+      "learning_rate": 0.0001040728862973761,
+      "loss": 0.0379,
+      "step": 22910
+    },
+    {
+      "epoch": 32.93,
+      "learning_rate": 0.00010398542274052477,
+      "loss": 0.0299,
+      "step": 22920
+    },
+    {
+      "epoch": 32.95,
+      "learning_rate": 0.00010389795918367346,
+      "loss": 0.0483,
+      "step": 22930
+    },
+    {
+      "epoch": 32.96,
+      "learning_rate": 0.00010381049562682215,
+      "loss": 0.0323,
+      "step": 22940
+    },
+    {
+      "epoch": 32.97,
+      "learning_rate": 0.00010372303206997083,
+      "loss": 0.0406,
+      "step": 22950
+    },
+    {
+      "epoch": 32.99,
+      "learning_rate": 0.00010363556851311954,
+      "loss": 0.0332,
+      "step": 22960
+    },
+    {
+      "epoch": 33.0,
+      "learning_rate": 0.00010354810495626821,
+      "loss": 0.0298,
+      "step": 22970
+    },
+    {
+      "epoch": 33.02,
+      "learning_rate": 0.0001034606413994169,
+      "loss": 0.0409,
+      "step": 22980
+    },
+    {
+      "epoch": 33.03,
+      "learning_rate": 0.00010337317784256559,
+      "loss": 0.0228,
+      "step": 22990
+    },
+    {
+      "epoch": 33.05,
+      "learning_rate": 0.00010328571428571429,
+      "loss": 0.0311,
+      "step": 23000
+    },
+    {
+      "epoch": 33.05,
+      "eval_loss": 0.13526782393455505,
+      "eval_runtime": 40.1803,
+      "eval_samples_per_second": 11.374,
+      "eval_steps_per_second": 1.443,
+      "eval_wer": 0.13175726547206745,
+      "step": 23000
+    },
+    {
+      "epoch": 33.06,
+      "learning_rate": 0.00010319825072886296,
+      "loss": 0.0345,
+      "step": 23010
+    },
+    {
+      "epoch": 33.07,
+      "learning_rate": 0.00010311078717201165,
+      "loss": 0.0289,
+      "step": 23020
+    },
+    {
+      "epoch": 33.09,
+      "learning_rate": 0.00010302332361516034,
+      "loss": 0.0321,
+      "step": 23030
+    },
+    {
+      "epoch": 33.1,
+      "learning_rate": 0.00010293586005830904,
+      "loss": 0.0231,
+      "step": 23040
+    },
+    {
+      "epoch": 33.12,
+      "learning_rate": 0.00010284839650145773,
+      "loss": 0.0366,
+      "step": 23050
+    },
+    {
+      "epoch": 33.13,
+      "learning_rate": 0.0001027609329446064,
+      "loss": 0.0343,
+      "step": 23060
+    },
+    {
+      "epoch": 33.15,
+      "learning_rate": 0.00010267346938775509,
+      "loss": 0.0263,
+      "step": 23070
+    },
+    {
+      "epoch": 33.16,
+      "learning_rate": 0.00010258600583090379,
+      "loss": 0.0291,
+      "step": 23080
+    },
+    {
+      "epoch": 33.18,
+      "learning_rate": 0.00010249854227405248,
+      "loss": 0.0263,
+      "step": 23090
+    },
+    {
+      "epoch": 33.19,
+      "learning_rate": 0.00010241107871720115,
+      "loss": 0.0402,
+      "step": 23100
+    },
+    {
+      "epoch": 33.19,
+      "eval_loss": 0.1352507621049881,
+      "eval_runtime": 40.2484,
+      "eval_samples_per_second": 11.354,
+      "eval_steps_per_second": 1.441,
+      "eval_wer": 0.13130552627616324,
+      "step": 23100
+    },
+    {
+      "epoch": 33.2,
+      "learning_rate": 0.00010232361516034984,
+      "loss": 0.0336,
+      "step": 23110
+    },
+    {
+      "epoch": 33.22,
+      "learning_rate": 0.00010223615160349853,
+      "loss": 0.0223,
+      "step": 23120
+    },
+    {
+      "epoch": 33.23,
+      "learning_rate": 0.00010214868804664723,
+      "loss": 0.0308,
+      "step": 23130
+    },
+    {
+      "epoch": 33.25,
+      "learning_rate": 0.00010206122448979592,
+      "loss": 0.0265,
+      "step": 23140
+    },
+    {
+      "epoch": 33.26,
+      "learning_rate": 0.00010197376093294459,
+      "loss": 0.0335,
+      "step": 23150
+    },
+    {
+      "epoch": 33.28,
+      "learning_rate": 0.00010188629737609328,
+      "loss": 0.0282,
+      "step": 23160
+    },
+    {
+      "epoch": 33.29,
+      "learning_rate": 0.00010179883381924198,
+      "loss": 0.0322,
+      "step": 23170
+    },
+    {
+      "epoch": 33.3,
+      "learning_rate": 0.00010171137026239067,
+      "loss": 0.0352,
+      "step": 23180
+    },
+    {
+      "epoch": 33.32,
+      "learning_rate": 0.00010162390670553934,
+      "loss": 0.0235,
+      "step": 23190
+    },
+    {
+      "epoch": 33.33,
+      "learning_rate": 0.00010153644314868803,
+      "loss": 0.0327,
+      "step": 23200
+    },
+    {
+      "epoch": 33.33,
+      "eval_loss": 0.13268069922924042,
+      "eval_runtime": 40.328,
+      "eval_samples_per_second": 11.332,
+      "eval_steps_per_second": 1.438,
+      "eval_wer": 0.1341665411835567,
+      "step": 23200
+    },
+    {
+      "epoch": 33.35,
+      "learning_rate": 0.00010144897959183673,
+      "loss": 0.0308,
+      "step": 23210
+    },
+    {
+      "epoch": 33.36,
+      "learning_rate": 0.00010136151603498542,
+      "loss": 0.032,
+      "step": 23220
+    },
+    {
+      "epoch": 33.38,
+      "learning_rate": 0.00010127405247813411,
+      "loss": 0.0361,
+      "step": 23230
+    },
+    {
+      "epoch": 33.39,
+      "learning_rate": 0.00010118658892128278,
+      "loss": 0.0276,
+      "step": 23240
+    },
+    {
+      "epoch": 33.41,
+      "learning_rate": 0.00010109912536443147,
+      "loss": 0.0325,
+      "step": 23250
+    },
+    {
+      "epoch": 33.42,
+      "learning_rate": 0.00010101166180758017,
+      "loss": 0.0301,
+      "step": 23260
+    },
+    {
+      "epoch": 33.43,
+      "learning_rate": 0.00010092419825072886,
+      "loss": 0.0376,
+      "step": 23270
+    },
+    {
+      "epoch": 33.45,
+      "learning_rate": 0.00010083673469387753,
+      "loss": 0.0297,
+      "step": 23280
+    },
+    {
+      "epoch": 33.46,
+      "learning_rate": 0.00010074927113702622,
+      "loss": 0.028,
+      "step": 23290
+    },
+    {
+      "epoch": 33.48,
+      "learning_rate": 0.00010066180758017492,
+      "loss": 0.0249,
+      "step": 23300
+    },
+    {
+      "epoch": 33.48,
+      "eval_loss": 0.13825057446956635,
+      "eval_runtime": 40.2776,
+      "eval_samples_per_second": 11.346,
+      "eval_steps_per_second": 1.44,
+      "eval_wer": 0.13687697635898208,
+      "step": 23300
+    },
+    {
+      "epoch": 33.49,
+      "learning_rate": 0.00010057434402332361,
+      "loss": 0.0319,
+      "step": 23310
+    },
+    {
+      "epoch": 33.51,
+      "learning_rate": 0.0001004868804664723,
+      "loss": 0.0353,
+      "step": 23320
+    },
+    {
+      "epoch": 33.52,
+      "learning_rate": 0.00010039941690962097,
+      "loss": 0.0345,
+      "step": 23330
+    },
+    {
+      "epoch": 33.53,
+      "learning_rate": 0.00010031195335276968,
+      "loss": 0.0312,
+      "step": 23340
+    },
+    {
+      "epoch": 33.55,
+      "learning_rate": 0.00010022448979591836,
+      "loss": 0.0414,
+      "step": 23350
+    },
+    {
+      "epoch": 33.56,
+      "learning_rate": 0.00010013702623906705,
+      "loss": 0.027,
+      "step": 23360
+    },
+    {
+      "epoch": 33.58,
+      "learning_rate": 0.00010004956268221573,
+      "loss": 0.0313,
+      "step": 23370
+    },
+    {
+      "epoch": 33.59,
+      "learning_rate": 9.996209912536441e-05,
+      "loss": 0.0367,
+      "step": 23380
+    },
+    {
+      "epoch": 33.61,
+      "learning_rate": 9.987463556851312e-05,
+      "loss": 0.0298,
+      "step": 23390
+    },
+    {
+      "epoch": 33.62,
+      "learning_rate": 9.97871720116618e-05,
+      "loss": 0.0331,
+      "step": 23400
+    },
+    {
+      "epoch": 33.62,
+      "eval_loss": 0.13275952637195587,
+      "eval_runtime": 40.5257,
+      "eval_samples_per_second": 11.277,
+      "eval_steps_per_second": 1.431,
+      "eval_wer": 0.13356422225568437,
+      "step": 23400
+    },
+    {
+      "epoch": 33.64,
+      "learning_rate": 9.969970845481049e-05,
+      "loss": 0.0283,
+      "step": 23410
+    },
+    {
+      "epoch": 33.65,
+      "learning_rate": 9.961224489795917e-05,
+      "loss": 0.0227,
+      "step": 23420
+    },
+    {
+      "epoch": 33.66,
+      "learning_rate": 9.952478134110787e-05,
+      "loss": 0.0382,
+      "step": 23430
+    },
+    {
+      "epoch": 33.68,
+      "learning_rate": 9.943731778425655e-05,
+      "loss": 0.0297,
+      "step": 23440
+    },
+    {
+      "epoch": 33.69,
+      "learning_rate": 9.934985422740524e-05,
+      "loss": 0.0435,
+      "step": 23450
+    },
+    {
+      "epoch": 33.71,
+      "learning_rate": 9.926239067055392e-05,
+      "loss": 0.0277,
+      "step": 23460
+    },
+    {
+      "epoch": 33.72,
+      "learning_rate": 9.917492711370262e-05,
+      "loss": 0.0277,
+      "step": 23470
+    },
+    {
+      "epoch": 33.74,
+      "learning_rate": 9.90874635568513e-05,
+      "loss": 0.0335,
+      "step": 23480
+    },
+    {
+      "epoch": 33.75,
+      "learning_rate": 9.9e-05,
+      "loss": 0.0244,
+      "step": 23490
+    },
+    {
+      "epoch": 33.76,
+      "learning_rate": 9.891253644314867e-05,
+      "loss": 0.0249,
+      "step": 23500
+    },
+    {
+      "epoch": 33.76,
+      "eval_loss": 0.13773199915885925,
+      "eval_runtime": 40.1237,
+      "eval_samples_per_second": 11.39,
+      "eval_steps_per_second": 1.446,
+      "eval_wer": 0.13296190332781208,
+      "step": 23500
+    },
+    {
+      "epoch": 33.78,
+      "learning_rate": 9.882507288629736e-05,
+      "loss": 0.0326,
+      "step": 23510
+    },
+    {
+      "epoch": 33.79,
+      "learning_rate": 9.873760932944606e-05,
+      "loss": 0.0273,
+      "step": 23520
+    },
+    {
+      "epoch": 33.81,
+      "learning_rate": 9.865014577259475e-05,
+      "loss": 0.033,
+      "step": 23530
+    },
+    {
+      "epoch": 33.82,
+      "learning_rate": 9.856268221574343e-05,
+      "loss": 0.0273,
+      "step": 23540
+    },
+    {
+      "epoch": 33.84,
+      "learning_rate": 9.847521865889211e-05,
+      "loss": 0.0266,
+      "step": 23550
+    },
+    {
+      "epoch": 33.85,
+      "learning_rate": 9.838775510204081e-05,
+      "loss": 0.0297,
+      "step": 23560
+    },
+    {
+      "epoch": 33.86,
+      "learning_rate": 9.83002915451895e-05,
+      "loss": 0.033,
+      "step": 23570
+    },
+    {
+      "epoch": 33.88,
+      "learning_rate": 9.821282798833819e-05,
+      "loss": 0.0366,
+      "step": 23580
+    },
+    {
+      "epoch": 33.89,
+      "learning_rate": 9.812536443148686e-05,
+      "loss": 0.0279,
+      "step": 23590
+    },
+    {
+      "epoch": 33.91,
+      "learning_rate": 9.803790087463556e-05,
+      "loss": 0.0374,
+      "step": 23600
+    },
+    {
+      "epoch": 33.91,
+      "eval_loss": 0.13894473016262054,
+      "eval_runtime": 40.1032,
+      "eval_samples_per_second": 11.396,
+      "eval_steps_per_second": 1.446,
+      "eval_wer": 0.1287456708327059,
+      "step": 23600
+    },
+    {
+      "epoch": 33.92,
+      "learning_rate": 9.795043731778425e-05,
+      "loss": 0.0311,
+      "step": 23610
+    },
+    {
+      "epoch": 33.94,
+      "learning_rate": 9.786297376093294e-05,
+      "loss": 0.0306,
+      "step": 23620
+    },
+    {
+      "epoch": 33.95,
+      "learning_rate": 9.777551020408162e-05,
+      "loss": 0.0286,
+      "step": 23630
+    },
+    {
+      "epoch": 33.97,
+      "learning_rate": 9.768804664723031e-05,
+      "loss": 0.0365,
+      "step": 23640
+    },
+    {
+      "epoch": 33.98,
+      "learning_rate": 9.7600583090379e-05,
+      "loss": 0.0326,
+      "step": 23650
+    },
+    {
+      "epoch": 33.99,
+      "learning_rate": 9.751311953352769e-05,
+      "loss": 0.0268,
+      "step": 23660
+    },
+    {
+      "epoch": 34.01,
+      "learning_rate": 9.742565597667638e-05,
+      "loss": 0.0312,
+      "step": 23670
+    },
+    {
+      "epoch": 34.02,
+      "learning_rate": 9.733819241982505e-05,
+      "loss": 0.0326,
+      "step": 23680
+    },
+    {
+      "epoch": 34.04,
+      "learning_rate": 9.725072886297375e-05,
+      "loss": 0.0228,
+      "step": 23690
+    },
+    {
+      "epoch": 34.05,
+      "learning_rate": 9.716326530612244e-05,
+      "loss": 0.0356,
+      "step": 23700
+    },
+    {
+      "epoch": 34.05,
+      "eval_loss": 0.13883648812770844,
+      "eval_runtime": 40.284,
+      "eval_samples_per_second": 11.344,
+      "eval_steps_per_second": 1.44,
+      "eval_wer": 0.12964914922451437,
+      "step": 23700
+    },
+    {
+      "epoch": 34.07,
+      "learning_rate": 9.707580174927113e-05,
+      "loss": 0.0236,
+      "step": 23710
+    },
+    {
+      "epoch": 34.08,
+      "learning_rate": 9.698833819241983e-05,
+      "loss": 0.0282,
+      "step": 23720
+    },
+    {
+      "epoch": 34.09,
+      "learning_rate": 9.69008746355685e-05,
+      "loss": 0.0319,
+      "step": 23730
+    },
+    {
+      "epoch": 34.11,
+      "learning_rate": 9.681341107871719e-05,
+      "loss": 0.0324,
+      "step": 23740
+    },
+    {
+      "epoch": 34.12,
+      "learning_rate": 9.672594752186588e-05,
+      "loss": 0.0314,
+      "step": 23750
+    },
+    {
+      "epoch": 34.14,
+      "learning_rate": 9.663848396501457e-05,
+      "loss": 0.0232,
+      "step": 23760
+    },
+    {
+      "epoch": 34.15,
+      "learning_rate": 9.655102040816326e-05,
+      "loss": 0.028,
+      "step": 23770
+    },
+    {
+      "epoch": 34.17,
+      "learning_rate": 9.646355685131194e-05,
+      "loss": 0.0244,
+      "step": 23780
+    },
+    {
+      "epoch": 34.18,
+      "learning_rate": 9.637609329446063e-05,
+      "loss": 0.0243,
+      "step": 23790
+    },
+    {
+      "epoch": 34.2,
+      "learning_rate": 9.628862973760932e-05,
+      "loss": 0.0337,
+      "step": 23800
+    },
+    {
+      "epoch": 34.2,
+      "eval_loss": 0.13111823797225952,
+      "eval_runtime": 40.2412,
+      "eval_samples_per_second": 11.357,
+      "eval_steps_per_second": 1.441,
+      "eval_wer": 0.1287456708327059,
+      "step": 23800
+    },
+    {
+      "epoch": 34.21,
+      "learning_rate": 9.620116618075802e-05,
+      "loss": 0.0394,
+      "step": 23810
+    },
+    {
+      "epoch": 34.22,
+      "learning_rate": 9.61137026239067e-05,
+      "loss": 0.0362,
+      "step": 23820
+    },
+    {
+      "epoch": 34.24,
+      "learning_rate": 9.602623906705538e-05,
+      "loss": 0.0246,
+      "step": 23830
+    },
+    {
+      "epoch": 34.25,
+      "learning_rate": 9.593877551020407e-05,
+      "loss": 0.025,
+      "step": 23840
+    },
+    {
+      "epoch": 34.27,
+      "learning_rate": 9.585131195335277e-05,
+      "loss": 0.0399,
+      "step": 23850
+    },
+    {
+      "epoch": 34.28,
+      "learning_rate": 9.576384839650145e-05,
+      "loss": 0.0253,
+      "step": 23860
+    },
+    {
+      "epoch": 34.3,
+      "learning_rate": 9.567638483965013e-05,
+      "loss": 0.0316,
+      "step": 23870
+    },
+    {
+      "epoch": 34.31,
+      "learning_rate": 9.558892128279882e-05,
+      "loss": 0.0288,
+      "step": 23880
+    },
+    {
+      "epoch": 34.32,
+      "learning_rate": 9.550145772594751e-05,
+      "loss": 0.0204,
+      "step": 23890
+    },
+    {
+      "epoch": 34.34,
+      "learning_rate": 9.541399416909621e-05,
+      "loss": 0.026,
+      "step": 23900
+    },
+    {
+      "epoch": 34.34,
+      "eval_loss": 0.1346319168806076,
+      "eval_runtime": 40.2297,
+      "eval_samples_per_second": 11.36,
+      "eval_steps_per_second": 1.442,
+      "eval_wer": 0.1275410329769613,
+      "step": 23900
+    },
+    {
+      "epoch": 34.35,
+      "learning_rate": 9.532653061224489e-05,
+      "loss": 0.0241,
+      "step": 23910
+    },
+    {
+      "epoch": 34.37,
+      "learning_rate": 9.523906705539357e-05,
+      "loss": 0.0293,
+      "step": 23920
+    },
+    {
+      "epoch": 34.38,
+      "learning_rate": 9.515160349854226e-05,
+      "loss": 0.0265,
+      "step": 23930
+    },
+    {
+      "epoch": 34.4,
+      "learning_rate": 9.506413994169096e-05,
+      "loss": 0.0344,
+      "step": 23940
+    },
+    {
+      "epoch": 34.41,
+      "learning_rate": 9.497667638483964e-05,
+      "loss": 0.0308,
+      "step": 23950
+    },
+    {
+      "epoch": 34.43,
+      "learning_rate": 9.488921282798833e-05,
+      "loss": 0.0256,
+      "step": 23960
+    },
+    {
+      "epoch": 34.44,
+      "learning_rate": 9.480174927113701e-05,
+      "loss": 0.0286,
+      "step": 23970
+    },
+    {
+      "epoch": 34.45,
+      "learning_rate": 9.471428571428571e-05,
+      "loss": 0.0393,
+      "step": 23980
+    },
+    {
+      "epoch": 34.47,
+      "learning_rate": 9.46268221574344e-05,
+      "loss": 0.0203,
+      "step": 23990
+    },
+    {
+      "epoch": 34.48,
+      "learning_rate": 9.453935860058308e-05,
+      "loss": 0.0312,
+      "step": 24000
+    },
+    {
+      "epoch": 34.48,
+      "eval_loss": 0.1321752965450287,
+      "eval_runtime": 40.3973,
+      "eval_samples_per_second": 11.313,
+      "eval_steps_per_second": 1.436,
+      "eval_wer": 0.12919741002861015,
+      "step": 24000
+    },
+    {
+      "epoch": 34.5,
+      "learning_rate": 9.445189504373176e-05,
+      "loss": 0.0277,
+      "step": 24010
+    },
+    {
+      "epoch": 34.51,
+      "learning_rate": 9.436443148688047e-05,
+      "loss": 0.0386,
+      "step": 24020
+    },
+    {
+      "epoch": 34.53,
+      "learning_rate": 9.427696793002915e-05,
+      "loss": 0.0385,
+      "step": 24030
+    },
+    {
+      "epoch": 34.54,
+      "learning_rate": 9.418950437317783e-05,
+      "loss": 0.0285,
+      "step": 24040
+    },
+    {
+      "epoch": 34.55,
+      "learning_rate": 9.410204081632652e-05,
+      "loss": 0.0305,
+      "step": 24050
+    },
+    {
+      "epoch": 34.57,
+      "learning_rate": 9.40145772594752e-05,
+      "loss": 0.0206,
+      "step": 24060
+    },
+    {
+      "epoch": 34.58,
+      "learning_rate": 9.39271137026239e-05,
+      "loss": 0.0479,
+      "step": 24070
+    },
+    {
+      "epoch": 34.6,
+      "learning_rate": 9.38396501457726e-05,
+      "loss": 0.0314,
+      "step": 24080
+    },
+    {
+      "epoch": 34.61,
+      "learning_rate": 9.375218658892127e-05,
+      "loss": 0.0233,
+      "step": 24090
+    },
+    {
+      "epoch": 34.63,
+      "learning_rate": 9.366472303206996e-05,
+      "loss": 0.0285,
+      "step": 24100
+    },
+    {
+      "epoch": 34.63,
+      "eval_loss": 0.13412344455718994,
+      "eval_runtime": 40.2936,
+      "eval_samples_per_second": 11.342,
+      "eval_steps_per_second": 1.439,
+      "eval_wer": 0.13100436681222707,
+      "step": 24100
+    },
+    {
+      "epoch": 34.64,
+      "learning_rate": 9.357725947521866e-05,
+      "loss": 0.0229,
+      "step": 24110
+    },
+    {
+      "epoch": 34.66,
+      "learning_rate": 9.348979591836735e-05,
+      "loss": 0.0386,
+      "step": 24120
+    },
+    {
+      "epoch": 34.67,
+      "learning_rate": 9.340233236151602e-05,
+      "loss": 0.0386,
+      "step": 24130
+    },
+    {
+      "epoch": 34.68,
+      "learning_rate": 9.331486880466471e-05,
+      "loss": 0.0278,
+      "step": 24140
+    },
+    {
+      "epoch": 34.7,
+      "learning_rate": 9.322740524781341e-05,
+      "loss": 0.0295,
+      "step": 24150
+    },
+    {
+      "epoch": 34.71,
+      "learning_rate": 9.31399416909621e-05,
+      "loss": 0.019,
+      "step": 24160
+    },
+    {
+      "epoch": 34.73,
+      "learning_rate": 9.305247813411078e-05,
+      "loss": 0.031,
+      "step": 24170
+    },
+    {
+      "epoch": 34.74,
+      "learning_rate": 9.296501457725946e-05,
+      "loss": 0.0325,
+      "step": 24180
+    },
+    {
+      "epoch": 34.76,
+      "learning_rate": 9.287755102040815e-05,
+      "loss": 0.0343,
+      "step": 24190
+    },
+    {
+      "epoch": 34.77,
+      "learning_rate": 9.279008746355685e-05,
+      "loss": 0.0314,
+      "step": 24200
+    },
+    {
+      "epoch": 34.77,
+      "eval_loss": 0.1259945183992386,
+      "eval_runtime": 40.0768,
+      "eval_samples_per_second": 11.403,
+      "eval_steps_per_second": 1.447,
+      "eval_wer": 0.130853787080259,
+      "step": 24200
+    },
+    {
+      "epoch": 34.78,
+      "learning_rate": 9.270262390670554e-05,
+      "loss": 0.0263,
+      "step": 24210
+    },
+    {
+      "epoch": 34.8,
+      "learning_rate": 9.261516034985421e-05,
+      "loss": 0.026,
+      "step": 24220
+    },
+    {
+      "epoch": 34.81,
+      "learning_rate": 9.25276967930029e-05,
+      "loss": 0.0285,
+      "step": 24230
+    },
+    {
+      "epoch": 34.83,
+      "learning_rate": 9.24402332361516e-05,
+      "loss": 0.0262,
+      "step": 24240
+    },
+    {
+      "epoch": 34.84,
+      "learning_rate": 9.235276967930029e-05,
+      "loss": 0.0253,
+      "step": 24250
+    },
+    {
+      "epoch": 34.86,
+      "learning_rate": 9.226530612244898e-05,
+      "loss": 0.0243,
+      "step": 24260
+    },
+    {
+      "epoch": 34.87,
+      "learning_rate": 9.217784256559765e-05,
+      "loss": 0.0215,
+      "step": 24270
+    },
+    {
+      "epoch": 34.89,
+      "learning_rate": 9.209037900874635e-05,
+      "loss": 0.035,
+      "step": 24280
+    },
+    {
+      "epoch": 34.9,
+      "learning_rate": 9.200291545189504e-05,
+      "loss": 0.0308,
+      "step": 24290
+    },
+    {
+      "epoch": 34.91,
+      "learning_rate": 9.191545189504373e-05,
+      "loss": 0.0318,
+      "step": 24300
+    },
+    {
+      "epoch": 34.91,
+      "eval_loss": 0.13191795349121094,
+      "eval_runtime": 40.2331,
+      "eval_samples_per_second": 11.359,
+      "eval_steps_per_second": 1.442,
+      "eval_wer": 0.12995030868845053,
+      "step": 24300
+    },
+    {
+      "epoch": 34.93,
+      "learning_rate": 9.18279883381924e-05,
+      "loss": 0.0287,
+      "step": 24310
+    },
+    {
+      "epoch": 34.94,
+      "learning_rate": 9.174052478134109e-05,
+      "loss": 0.0317,
+      "step": 24320
+    },
+    {
+      "epoch": 34.96,
+      "learning_rate": 9.165306122448979e-05,
+      "loss": 0.0226,
+      "step": 24330
+    },
+    {
+      "epoch": 34.97,
+      "learning_rate": 9.156559766763848e-05,
+      "loss": 0.0232,
+      "step": 24340
+    },
+    {
+      "epoch": 34.99,
+      "learning_rate": 9.147813411078717e-05,
+      "loss": 0.0355,
+      "step": 24350
+    },
+    {
+      "epoch": 35.0,
+      "learning_rate": 9.139067055393584e-05,
+      "loss": 0.032,
+      "step": 24360
+    },
+    {
+      "epoch": 35.01,
+      "learning_rate": 9.130320699708454e-05,
+      "loss": 0.0381,
+      "step": 24370
+    },
+    {
+      "epoch": 35.03,
+      "learning_rate": 9.121574344023323e-05,
+      "loss": 0.0248,
+      "step": 24380
+    },
+    {
+      "epoch": 35.04,
+      "learning_rate": 9.112827988338192e-05,
+      "loss": 0.0202,
+      "step": 24390
+    },
+    {
+      "epoch": 35.06,
+      "learning_rate": 9.104081632653059e-05,
+      "loss": 0.0238,
+      "step": 24400
+    },
+    {
+      "epoch": 35.06,
+      "eval_loss": 0.13871027529239655,
+      "eval_runtime": 40.2335,
+      "eval_samples_per_second": 11.359,
+      "eval_steps_per_second": 1.442,
+      "eval_wer": 0.13010088842041861,
+      "step": 24400
+    },
+    {
+      "epoch": 35.07,
+      "learning_rate": 9.09533527696793e-05,
+      "loss": 0.0209,
+      "step": 24410
+    },
+    {
+      "epoch": 35.09,
+      "learning_rate": 9.086588921282798e-05,
+      "loss": 0.0266,
+      "step": 24420
+    },
+    {
+      "epoch": 35.1,
+      "learning_rate": 9.077842565597667e-05,
+      "loss": 0.0217,
+      "step": 24430
+    },
+    {
+      "epoch": 35.11,
+      "learning_rate": 9.069096209912536e-05,
+      "loss": 0.034,
+      "step": 24440
+    },
+    {
+      "epoch": 35.13,
+      "learning_rate": 9.060349854227405e-05,
+      "loss": 0.0335,
+      "step": 24450
+    },
+    {
+      "epoch": 35.14,
+      "learning_rate": 9.051603498542273e-05,
+      "loss": 0.0352,
+      "step": 24460
+    },
+    {
+      "epoch": 35.16,
+      "learning_rate": 9.042857142857142e-05,
+      "loss": 0.027,
+      "step": 24470
+    },
+    {
+      "epoch": 35.17,
+      "learning_rate": 9.034110787172011e-05,
+      "loss": 0.0254,
+      "step": 24480
+    },
+    {
+      "epoch": 35.19,
+      "learning_rate": 9.025364431486878e-05,
+      "loss": 0.0319,
+      "step": 24490
+    },
+    {
+      "epoch": 35.2,
+      "learning_rate": 9.016618075801749e-05,
+      "loss": 0.0213,
+      "step": 24500
+    },
+    {
+      "epoch": 35.2,
+      "eval_loss": 0.13293719291687012,
+      "eval_runtime": 40.6113,
+      "eval_samples_per_second": 11.253,
+      "eval_steps_per_second": 1.428,
+      "eval_wer": 0.1326607438638759,
+      "step": 24500
+    },
+    {
+      "epoch": 35.22,
+      "learning_rate": 9.007871720116617e-05,
+      "loss": 0.0234,
+      "step": 24510
+    },
+    {
+      "epoch": 35.23,
+      "learning_rate": 8.999125364431486e-05,
+      "loss": 0.029,
+      "step": 24520
+    },
+    {
+      "epoch": 35.24,
+      "learning_rate": 8.990379008746356e-05,
+      "loss": 0.0278,
+      "step": 24530
+    },
+    {
+      "epoch": 35.26,
+      "learning_rate": 8.981632653061224e-05,
+      "loss": 0.0288,
+      "step": 24540
+    },
+    {
+      "epoch": 35.27,
+      "learning_rate": 8.972886297376092e-05,
+      "loss": 0.0223,
+      "step": 24550
+    },
+    {
+      "epoch": 35.29,
+      "learning_rate": 8.964139941690961e-05,
+      "loss": 0.0208,
+      "step": 24560
+    },
+    {
+      "epoch": 35.3,
+      "learning_rate": 8.95539358600583e-05,
+      "loss": 0.0356,
+      "step": 24570
+    },
+    {
+      "epoch": 35.32,
+      "learning_rate": 8.946647230320699e-05,
+      "loss": 0.0262,
+      "step": 24580
+    },
+    {
+      "epoch": 35.33,
+      "learning_rate": 8.937900874635568e-05,
+      "loss": 0.0324,
+      "step": 24590
+    },
+    {
+      "epoch": 35.34,
+      "learning_rate": 8.929154518950436e-05,
+      "loss": 0.0261,
+      "step": 24600
+    },
+    {
+      "epoch": 35.34,
+      "eval_loss": 0.1314808577299118,
+      "eval_runtime": 40.3503,
+      "eval_samples_per_second": 11.326,
+      "eval_steps_per_second": 1.437,
+      "eval_wer": 0.12949856949254632,
+      "step": 24600
+    },
+    {
+      "epoch": 35.36,
+      "learning_rate": 8.920408163265305e-05,
+      "loss": 0.0216,
+      "step": 24610
+    },
+    {
+      "epoch": 35.37,
+      "learning_rate": 8.911661807580175e-05,
+      "loss": 0.0416,
+      "step": 24620
+    },
+    {
+      "epoch": 35.39,
+      "learning_rate": 8.902915451895043e-05,
+      "loss": 0.0351,
+      "step": 24630
+    },
+    {
+      "epoch": 35.4,
+      "learning_rate": 8.894169096209912e-05,
+      "loss": 0.029,
+      "step": 24640
+    },
+    {
+      "epoch": 35.42,
+      "learning_rate": 8.88542274052478e-05,
+      "loss": 0.0291,
+      "step": 24650
+    },
+    {
+      "epoch": 35.43,
+      "learning_rate": 8.87667638483965e-05,
+      "loss": 0.0255,
+      "step": 24660
+    },
+    {
+      "epoch": 35.45,
+      "learning_rate": 8.867930029154518e-05,
+      "loss": 0.0305,
+      "step": 24670
+    },
+    {
+      "epoch": 35.46,
+      "learning_rate": 8.859183673469387e-05,
+      "loss": 0.0222,
+      "step": 24680
+    },
+    {
+      "epoch": 35.47,
+      "learning_rate": 8.850437317784256e-05,
+      "loss": 0.0319,
+      "step": 24690
+    },
+    {
+      "epoch": 35.49,
+      "learning_rate": 8.841690962099124e-05,
+      "loss": 0.0289,
+      "step": 24700
+    },
+    {
+      "epoch": 35.49,
+      "eval_loss": 0.13069851696491241,
+      "eval_runtime": 40.6111,
+      "eval_samples_per_second": 11.253,
+      "eval_steps_per_second": 1.428,
+      "eval_wer": 0.1322090046679717,
+      "step": 24700
+    },
+    {
+      "epoch": 35.5,
+      "learning_rate": 8.832944606413994e-05,
+      "loss": 0.0261,
+      "step": 24710
+    },
+    {
+      "epoch": 35.52,
+      "learning_rate": 8.824198250728862e-05,
+      "loss": 0.0288,
+      "step": 24720
+    },
+    {
+      "epoch": 35.53,
+      "learning_rate": 8.81545189504373e-05,
+      "loss": 0.0295,
+      "step": 24730
+    },
+    {
+      "epoch": 35.55,
+      "learning_rate": 8.8067055393586e-05,
+      "loss": 0.0246,
+      "step": 24740
+    },
+    {
+      "epoch": 35.56,
+      "learning_rate": 8.79795918367347e-05,
+      "loss": 0.0296,
+      "step": 24750
+    },
+    {
+      "epoch": 35.57,
+      "learning_rate": 8.789212827988337e-05,
+      "loss": 0.0251,
+      "step": 24760
+    },
+    {
+      "epoch": 35.59,
+      "learning_rate": 8.780466472303206e-05,
+      "loss": 0.0357,
+      "step": 24770
+    },
+    {
+      "epoch": 35.6,
+      "learning_rate": 8.771720116618075e-05,
+      "loss": 0.0235,
+      "step": 24780
+    },
+    {
+      "epoch": 35.62,
+      "learning_rate": 8.762973760932945e-05,
+      "loss": 0.028,
+      "step": 24790
+    },
+    {
+      "epoch": 35.63,
+      "learning_rate": 8.754227405247814e-05,
+      "loss": 0.0204,
+      "step": 24800
+    },
+    {
+      "epoch": 35.63,
+      "eval_loss": 0.12542444467544556,
+      "eval_runtime": 40.1226,
+      "eval_samples_per_second": 11.39,
+      "eval_steps_per_second": 1.446,
+      "eval_wer": 0.13115494654419516,
+      "step": 24800
+    },
+    {
+      "epoch": 35.65,
+      "learning_rate": 8.745481049562681e-05,
+      "loss": 0.0211,
+      "step": 24810
+    },
+    {
+      "epoch": 35.66,
+      "learning_rate": 8.73673469387755e-05,
+      "loss": 0.0259,
+      "step": 24820
+    },
+    {
+      "epoch": 35.68,
+      "learning_rate": 8.72798833819242e-05,
+      "loss": 0.0275,
+      "step": 24830
+    },
+    {
+      "epoch": 35.69,
+      "learning_rate": 8.719241982507289e-05,
+      "loss": 0.024,
+      "step": 24840
+    },
+    {
+      "epoch": 35.7,
+      "learning_rate": 8.710495626822156e-05,
+      "loss": 0.0257,
+      "step": 24850
+    },
+    {
+      "epoch": 35.72,
+      "learning_rate": 8.701749271137025e-05,
+      "loss": 0.0225,
+      "step": 24860
+    },
+    {
+      "epoch": 35.73,
+      "learning_rate": 8.693002915451894e-05,
+      "loss": 0.0375,
+      "step": 24870
+    },
+    {
+      "epoch": 35.75,
+      "learning_rate": 8.684256559766764e-05,
+      "loss": 0.0252,
+      "step": 24880
+    },
+    {
+      "epoch": 35.76,
+      "learning_rate": 8.675510204081633e-05,
+      "loss": 0.0298,
+      "step": 24890
+    },
+    {
+      "epoch": 35.78,
+      "learning_rate": 8.6667638483965e-05,
+      "loss": 0.0266,
+      "step": 24900
+    },
+    {
+      "epoch": 35.78,
+      "eval_loss": 0.1274455338716507,
+      "eval_runtime": 40.3501,
+      "eval_samples_per_second": 11.326,
+      "eval_steps_per_second": 1.437,
+      "eval_wer": 0.128896250564674,
+      "step": 24900
+    },
+    {
+      "epoch": 35.79,
+      "learning_rate": 8.658017492711369e-05,
+      "loss": 0.0278,
+      "step": 24910
+    },
+    {
+      "epoch": 35.8,
+      "learning_rate": 8.649271137026239e-05,
+      "loss": 0.0288,
+      "step": 24920
+    },
+    {
+      "epoch": 35.82,
+      "learning_rate": 8.640524781341108e-05,
+      "loss": 0.0205,
+      "step": 24930
+    },
+    {
+      "epoch": 35.83,
+      "learning_rate": 8.631778425655975e-05,
+      "loss": 0.0247,
+      "step": 24940
+    },
+    {
+      "epoch": 35.85,
+      "learning_rate": 8.623032069970844e-05,
+      "loss": 0.0282,
+      "step": 24950
+    },
+    {
+      "epoch": 35.86,
+      "learning_rate": 8.614285714285714e-05,
+      "loss": 0.0243,
+      "step": 24960
+    },
+    {
+      "epoch": 35.88,
+      "learning_rate": 8.605539358600583e-05,
+      "loss": 0.0311,
+      "step": 24970
+    },
+    {
+      "epoch": 35.89,
+      "learning_rate": 8.596793002915452e-05,
+      "loss": 0.0327,
+      "step": 24980
+    },
+    {
+      "epoch": 35.91,
+      "learning_rate": 8.588046647230319e-05,
+      "loss": 0.0361,
+      "step": 24990
+    },
+    {
+      "epoch": 35.92,
+      "learning_rate": 8.579300291545188e-05,
+      "loss": 0.0287,
+      "step": 25000
+    },
+    {
+      "epoch": 35.92,
+      "eval_loss": 0.12421078979969025,
+      "eval_runtime": 40.3244,
+      "eval_samples_per_second": 11.333,
+      "eval_steps_per_second": 1.438,
+      "eval_wer": 0.12422827887366361,
+      "step": 25000
+    },
+    {
+      "epoch": 35.93,
+      "learning_rate": 8.570553935860058e-05,
+      "loss": 0.0287,
+      "step": 25010
+    },
+    {
+      "epoch": 35.95,
+      "learning_rate": 8.561807580174927e-05,
+      "loss": 0.031,
+      "step": 25020
+    },
+    {
+      "epoch": 35.96,
+      "learning_rate": 8.553061224489794e-05,
+      "loss": 0.0313,
+      "step": 25030
+    },
+    {
+      "epoch": 35.98,
+      "learning_rate": 8.544314868804663e-05,
+      "loss": 0.0255,
+      "step": 25040
+    },
+    {
+      "epoch": 35.99,
+      "learning_rate": 8.535568513119533e-05,
+      "loss": 0.0262,
+      "step": 25050
+    },
+    {
+      "epoch": 36.01,
+      "learning_rate": 8.526822157434402e-05,
+      "loss": 0.0238,
+      "step": 25060
+    },
+    {
+      "epoch": 36.02,
+      "learning_rate": 8.518075801749271e-05,
+      "loss": 0.0263,
+      "step": 25070
+    },
+    {
+      "epoch": 36.03,
+      "learning_rate": 8.509329446064138e-05,
+      "loss": 0.0324,
+      "step": 25080
+    },
+    {
+      "epoch": 36.05,
+      "learning_rate": 8.500583090379008e-05,
+      "loss": 0.0249,
+      "step": 25090
+    },
+    {
+      "epoch": 36.06,
+      "learning_rate": 8.491836734693877e-05,
+      "loss": 0.0208,
+      "step": 25100
+    },
+    {
+      "epoch": 36.06,
+      "eval_loss": 0.13018617033958435,
+      "eval_runtime": 40.7248,
+      "eval_samples_per_second": 11.222,
+      "eval_steps_per_second": 1.424,
+      "eval_wer": 0.12769161270892937,
+      "step": 25100
+    },
+    {
+      "epoch": 36.08,
+      "learning_rate": 8.483090379008746e-05,
+      "loss": 0.0227,
+      "step": 25110
+    },
+    {
+      "epoch": 36.09,
+      "learning_rate": 8.474344023323613e-05,
+      "loss": 0.0232,
+      "step": 25120
+    },
+    {
+      "epoch": 36.11,
+      "learning_rate": 8.465597667638482e-05,
+      "loss": 0.0199,
+      "step": 25130
+    },
+    {
+      "epoch": 36.12,
+      "learning_rate": 8.456851311953352e-05,
+      "loss": 0.0356,
+      "step": 25140
+    },
+    {
+      "epoch": 36.14,
+      "learning_rate": 8.448104956268221e-05,
+      "loss": 0.0257,
+      "step": 25150
+    },
+    {
+      "epoch": 36.15,
+      "learning_rate": 8.43935860058309e-05,
+      "loss": 0.0306,
+      "step": 25160
+    },
+    {
+      "epoch": 36.16,
+      "learning_rate": 8.430612244897957e-05,
+      "loss": 0.0214,
+      "step": 25170
+    },
+    {
+      "epoch": 36.18,
+      "learning_rate": 8.421865889212828e-05,
+      "loss": 0.0245,
+      "step": 25180
+    },
+    {
+      "epoch": 36.19,
+      "learning_rate": 8.413119533527696e-05,
+      "loss": 0.0287,
+      "step": 25190
+    },
+    {
+      "epoch": 36.21,
+      "learning_rate": 8.404373177842565e-05,
+      "loss": 0.0258,
+      "step": 25200
+    },
+    {
+      "epoch": 36.21,
+      "eval_loss": 0.13101942837238312,
+      "eval_runtime": 40.0217,
+      "eval_samples_per_second": 11.419,
+      "eval_steps_per_second": 1.449,
+      "eval_wer": 0.1307032073482909,
+      "step": 25200
+    },
+    {
+      "epoch": 36.22,
+      "learning_rate": 8.395626822157433e-05,
+      "loss": 0.0289,
+      "step": 25210
+    },
+    {
+      "epoch": 36.24,
+      "learning_rate": 8.386880466472303e-05,
+      "loss": 0.0227,
+      "step": 25220
+    },
+    {
+      "epoch": 36.25,
+      "learning_rate": 8.378134110787172e-05,
+      "loss": 0.0181,
+      "step": 25230
+    },
+    {
+      "epoch": 36.26,
+      "learning_rate": 8.36938775510204e-05,
+      "loss": 0.0276,
+      "step": 25240
+    },
+    {
+      "epoch": 36.28,
+      "learning_rate": 8.360641399416909e-05,
+      "loss": 0.0324,
+      "step": 25250
+    },
+    {
+      "epoch": 36.29,
+      "learning_rate": 8.351895043731778e-05,
+      "loss": 0.0306,
+      "step": 25260
+    },
+    {
+      "epoch": 36.31,
+      "learning_rate": 8.343148688046647e-05,
+      "loss": 0.0247,
+      "step": 25270
+    },
+    {
+      "epoch": 36.32,
+      "learning_rate": 8.334402332361515e-05,
+      "loss": 0.0293,
+      "step": 25280
+    },
+    {
+      "epoch": 36.34,
+      "learning_rate": 8.325655976676384e-05,
+      "loss": 0.0303,
+      "step": 25290
+    },
+    {
+      "epoch": 36.35,
+      "learning_rate": 8.316909620991252e-05,
+      "loss": 0.0222,
+      "step": 25300
+    },
+    {
+      "epoch": 36.35,
+      "eval_loss": 0.12964005768299103,
+      "eval_runtime": 40.3469,
+      "eval_samples_per_second": 11.327,
+      "eval_steps_per_second": 1.438,
+      "eval_wer": 0.1287456708327059,
+      "step": 25300
+    },
+    {
+      "epoch": 36.36,
+      "learning_rate": 8.308163265306122e-05,
+      "loss": 0.0316,
+      "step": 25310
+    },
+    {
+      "epoch": 36.38,
+      "learning_rate": 8.29941690962099e-05,
+      "loss": 0.032,
+      "step": 25320
+    },
+    {
+      "epoch": 36.39,
+      "learning_rate": 8.29067055393586e-05,
+      "loss": 0.0185,
+      "step": 25330
+    },
+    {
+      "epoch": 36.41,
+      "learning_rate": 8.28192419825073e-05,
+      "loss": 0.0351,
+      "step": 25340
+    },
+    {
+      "epoch": 36.42,
+      "learning_rate": 8.273177842565597e-05,
+      "loss": 0.0242,
+      "step": 25350
+    },
+    {
+      "epoch": 36.44,
+      "learning_rate": 8.264431486880466e-05,
+      "loss": 0.0205,
+      "step": 25360
+    },
+    {
+      "epoch": 36.45,
+      "learning_rate": 8.255685131195335e-05,
+      "loss": 0.0269,
+      "step": 25370
+    },
+    {
+      "epoch": 36.47,
+      "learning_rate": 8.246938775510203e-05,
+      "loss": 0.0199,
+      "step": 25380
+    },
+    {
+      "epoch": 36.48,
+      "learning_rate": 8.238192419825072e-05,
+      "loss": 0.0291,
+      "step": 25390
+    },
+    {
+      "epoch": 36.49,
+      "learning_rate": 8.229446064139941e-05,
+      "loss": 0.0219,
+      "step": 25400
+    },
+    {
+      "epoch": 36.49,
+      "eval_loss": 0.13100750744342804,
+      "eval_runtime": 40.0869,
+      "eval_samples_per_second": 11.4,
+      "eval_steps_per_second": 1.447,
+      "eval_wer": 0.12784219244089745,
+      "step": 25400
+    },
+    {
+      "epoch": 36.51,
+      "learning_rate": 8.22069970845481e-05,
+      "loss": 0.0358,
+      "step": 25410
+    },
+    {
+      "epoch": 36.52,
+      "learning_rate": 8.211953352769679e-05,
+      "loss": 0.0206,
+      "step": 25420
+    },
+    {
+      "epoch": 36.54,
+      "learning_rate": 8.203206997084549e-05,
+      "loss": 0.0204,
+      "step": 25430
+    },
+    {
+      "epoch": 36.55,
+      "learning_rate": 8.194460641399416e-05,
+      "loss": 0.0332,
+      "step": 25440
+    },
+    {
+      "epoch": 36.57,
+      "learning_rate": 8.185714285714285e-05,
+      "loss": 0.0256,
+      "step": 25450
+    },
+    {
+      "epoch": 36.58,
+      "learning_rate": 8.177842565597667e-05,
+      "loss": 0.025,
+      "step": 25460
+    },
+    {
+      "epoch": 36.59,
+      "learning_rate": 8.169096209912536e-05,
+      "loss": 0.0284,
+      "step": 25470
+    },
+    {
+      "epoch": 36.61,
+      "learning_rate": 8.160349854227404e-05,
+      "loss": 0.0213,
+      "step": 25480
+    },
+    {
+      "epoch": 36.62,
+      "learning_rate": 8.151603498542275e-05,
+      "loss": 0.0368,
+      "step": 25490
+    },
+    {
+      "epoch": 36.64,
+      "learning_rate": 8.142857142857142e-05,
+      "loss": 0.0221,
+      "step": 25500
+    },
+    {
+      "epoch": 36.64,
+      "eval_loss": 0.1341370940208435,
+      "eval_runtime": 40.1907,
+      "eval_samples_per_second": 11.371,
+      "eval_steps_per_second": 1.443,
+      "eval_wer": 0.12934798976057824,
+      "step": 25500
+    },
+    {
+      "epoch": 36.65,
+      "learning_rate": 8.134110787172011e-05,
+      "loss": 0.0307,
+      "step": 25510
+    },
+    {
+      "epoch": 36.67,
+      "learning_rate": 8.12536443148688e-05,
+      "loss": 0.0182,
+      "step": 25520
+    },
+    {
+      "epoch": 36.68,
+      "learning_rate": 8.11661807580175e-05,
+      "loss": 0.0293,
+      "step": 25530
+    },
+    {
+      "epoch": 36.7,
+      "learning_rate": 8.107871720116617e-05,
+      "loss": 0.026,
+      "step": 25540
+    },
+    {
+      "epoch": 36.71,
+      "learning_rate": 8.099125364431486e-05,
+      "loss": 0.0252,
+      "step": 25550
+    },
+    {
+      "epoch": 36.72,
+      "learning_rate": 8.090379008746355e-05,
+      "loss": 0.0259,
+      "step": 25560
+    },
+    {
+      "epoch": 36.74,
+      "learning_rate": 8.081632653061223e-05,
+      "loss": 0.0219,
+      "step": 25570
+    },
+    {
+      "epoch": 36.75,
+      "learning_rate": 8.072886297376094e-05,
+      "loss": 0.0239,
+      "step": 25580
+    },
+    {
+      "epoch": 36.77,
+      "learning_rate": 8.064139941690961e-05,
+      "loss": 0.0336,
+      "step": 25590
+    },
+    {
+      "epoch": 36.78,
+      "learning_rate": 8.05539358600583e-05,
+      "loss": 0.0182,
+      "step": 25600
+    },
+    {
+      "epoch": 36.78,
+      "eval_loss": 0.13404802978038788,
+      "eval_runtime": 40.1153,
+      "eval_samples_per_second": 11.392,
+      "eval_steps_per_second": 1.446,
+      "eval_wer": 0.126938714049089,
+      "step": 25600
+    },
+    {
+      "epoch": 36.8,
+      "learning_rate": 8.046647230320699e-05,
+      "loss": 0.0247,
+      "step": 25610
+    },
+    {
+      "epoch": 36.81,
+      "learning_rate": 8.037900874635569e-05,
+      "loss": 0.0214,
+      "step": 25620
+    },
+    {
+      "epoch": 36.82,
+      "learning_rate": 8.029154518950436e-05,
+      "loss": 0.0266,
+      "step": 25630
+    },
+    {
+      "epoch": 36.84,
+      "learning_rate": 8.020408163265305e-05,
+      "loss": 0.0301,
+      "step": 25640
+    },
+    {
+      "epoch": 36.85,
+      "learning_rate": 8.011661807580174e-05,
+      "loss": 0.0188,
+      "step": 25650
+    },
+    {
+      "epoch": 36.87,
+      "learning_rate": 8.002915451895044e-05,
+      "loss": 0.0229,
+      "step": 25660
+    },
+    {
+      "epoch": 36.88,
+      "learning_rate": 7.994169096209913e-05,
+      "loss": 0.0237,
+      "step": 25670
+    },
+    {
+      "epoch": 36.9,
+      "learning_rate": 7.98542274052478e-05,
+      "loss": 0.0203,
+      "step": 25680
+    },
+    {
+      "epoch": 36.91,
+      "learning_rate": 7.976676384839649e-05,
+      "loss": 0.0262,
+      "step": 25690
+    },
+    {
+      "epoch": 36.93,
+      "learning_rate": 7.967930029154518e-05,
+      "loss": 0.0176,
+      "step": 25700
+    },
+    {
+      "epoch": 36.93,
+      "eval_loss": 0.13346485793590546,
+      "eval_runtime": 40.5735,
+      "eval_samples_per_second": 11.264,
+      "eval_steps_per_second": 1.43,
+      "eval_wer": 0.12769161270892937,
+      "step": 25700
+    },
+    {
+      "epoch": 36.94,
+      "learning_rate": 7.959183673469388e-05,
+      "loss": 0.0291,
+      "step": 25710
+    },
+    {
+      "epoch": 36.95,
+      "learning_rate": 7.950437317784255e-05,
+      "loss": 0.0327,
+      "step": 25720
+    },
+    {
+      "epoch": 36.97,
+      "learning_rate": 7.941690962099124e-05,
+      "loss": 0.0228,
+      "step": 25730
+    },
+    {
+      "epoch": 36.98,
+      "learning_rate": 7.932944606413993e-05,
+      "loss": 0.0276,
+      "step": 25740
+    },
+    {
+      "epoch": 37.0,
+      "learning_rate": 7.924198250728863e-05,
+      "loss": 0.025,
+      "step": 25750
+    },
+    {
+      "epoch": 37.01,
+      "learning_rate": 7.915451895043732e-05,
+      "loss": 0.0251,
+      "step": 25760
+    },
+    {
+      "epoch": 37.03,
+      "learning_rate": 7.906705539358599e-05,
+      "loss": 0.0258,
+      "step": 25770
+    },
+    {
+      "epoch": 37.04,
+      "learning_rate": 7.897959183673468e-05,
+      "loss": 0.0209,
+      "step": 25780
+    },
+    {
+      "epoch": 37.05,
+      "learning_rate": 7.889212827988338e-05,
+      "loss": 0.0242,
+      "step": 25790
+    },
+    {
+      "epoch": 37.07,
+      "learning_rate": 7.880466472303207e-05,
+      "loss": 0.0301,
+      "step": 25800
+    },
+    {
+      "epoch": 37.07,
+      "eval_loss": 0.13566622138023376,
+      "eval_runtime": 40.4741,
+      "eval_samples_per_second": 11.291,
+      "eval_steps_per_second": 1.433,
+      "eval_wer": 0.12844451136876978,
+      "step": 25800
+    },
+    {
+      "epoch": 37.08,
+      "learning_rate": 7.871720116618074e-05,
+      "loss": 0.03,
+      "step": 25810
+    },
+    {
+      "epoch": 37.1,
+      "learning_rate": 7.862973760932943e-05,
+      "loss": 0.0265,
+      "step": 25820
+    },
+    {
+      "epoch": 37.11,
+      "learning_rate": 7.854227405247813e-05,
+      "loss": 0.0225,
+      "step": 25830
+    },
+    {
+      "epoch": 37.13,
+      "learning_rate": 7.845481049562682e-05,
+      "loss": 0.0234,
+      "step": 25840
+    },
+    {
+      "epoch": 37.14,
+      "learning_rate": 7.836734693877551e-05,
+      "loss": 0.028,
+      "step": 25850
+    },
+    {
+      "epoch": 37.16,
+      "learning_rate": 7.827988338192418e-05,
+      "loss": 0.0323,
+      "step": 25860
+    },
+    {
+      "epoch": 37.17,
+      "learning_rate": 7.819241982507287e-05,
+      "loss": 0.0207,
+      "step": 25870
+    },
+    {
+      "epoch": 37.18,
+      "learning_rate": 7.810495626822157e-05,
+      "loss": 0.0262,
+      "step": 25880
+    },
+    {
+      "epoch": 37.2,
+      "learning_rate": 7.801749271137026e-05,
+      "loss": 0.0247,
+      "step": 25890
+    },
+    {
+      "epoch": 37.21,
+      "learning_rate": 7.793002915451894e-05,
+      "loss": 0.0205,
+      "step": 25900
+    },
+    {
+      "epoch": 37.21,
+      "eval_loss": 0.1321299970149994,
+      "eval_runtime": 40.1895,
+      "eval_samples_per_second": 11.371,
+      "eval_steps_per_second": 1.443,
+      "eval_wer": 0.12919741002861015,
+      "step": 25900
+    },
+    {
+      "epoch": 37.23,
+      "learning_rate": 7.784256559766762e-05,
+      "loss": 0.0258,
+      "step": 25910
+    },
+    {
+      "epoch": 37.24,
+      "learning_rate": 7.775510204081632e-05,
+      "loss": 0.018,
+      "step": 25920
+    },
+    {
+      "epoch": 37.26,
+      "learning_rate": 7.766763848396501e-05,
+      "loss": 0.0195,
+      "step": 25930
+    },
+    {
+      "epoch": 37.27,
+      "learning_rate": 7.75801749271137e-05,
+      "loss": 0.0229,
+      "step": 25940
+    },
+    {
+      "epoch": 37.28,
+      "learning_rate": 7.749271137026237e-05,
+      "loss": 0.0225,
+      "step": 25950
+    },
+    {
+      "epoch": 37.3,
+      "learning_rate": 7.740524781341108e-05,
+      "loss": 0.0288,
+      "step": 25960
+    },
+    {
+      "epoch": 37.31,
+      "learning_rate": 7.731778425655976e-05,
+      "loss": 0.0283,
+      "step": 25970
+    },
+    {
+      "epoch": 37.33,
+      "learning_rate": 7.723032069970845e-05,
+      "loss": 0.0255,
+      "step": 25980
+    },
+    {
+      "epoch": 37.34,
+      "learning_rate": 7.714285714285713e-05,
+      "loss": 0.027,
+      "step": 25990
+    },
+    {
+      "epoch": 37.36,
+      "learning_rate": 7.705539358600581e-05,
+      "loss": 0.0193,
+      "step": 26000
+    },
+    {
+      "epoch": 37.36,
+      "eval_loss": 0.1375712752342224,
+      "eval_runtime": 40.3298,
+      "eval_samples_per_second": 11.332,
+      "eval_steps_per_second": 1.438,
+      "eval_wer": 0.12844451136876978,
+      "step": 26000
+    },
+    {
+      "epoch": 37.37,
+      "learning_rate": 7.696793002915452e-05,
+      "loss": 0.0275,
+      "step": 26010
+    },
+    {
+      "epoch": 37.39,
+      "learning_rate": 7.68804664723032e-05,
+      "loss": 0.0196,
+      "step": 26020
+    },
+    {
+      "epoch": 37.4,
+      "learning_rate": 7.679300291545189e-05,
+      "loss": 0.0192,
+      "step": 26030
+    },
+    {
+      "epoch": 37.41,
+      "learning_rate": 7.670553935860057e-05,
+      "loss": 0.0291,
+      "step": 26040
+    },
+    {
+      "epoch": 37.43,
+      "learning_rate": 7.661807580174927e-05,
+      "loss": 0.0179,
+      "step": 26050
+    },
+    {
+      "epoch": 37.44,
+      "learning_rate": 7.653061224489796e-05,
+      "loss": 0.0249,
+      "step": 26060
+    },
+    {
+      "epoch": 37.46,
+      "learning_rate": 7.644314868804664e-05,
+      "loss": 0.0202,
+      "step": 26070
+    },
+    {
+      "epoch": 37.47,
+      "learning_rate": 7.635568513119532e-05,
+      "loss": 0.0212,
+      "step": 26080
+    },
+    {
+      "epoch": 37.49,
+      "learning_rate": 7.626822157434402e-05,
+      "loss": 0.0258,
+      "step": 26090
+    },
+    {
+      "epoch": 37.5,
+      "learning_rate": 7.618075801749271e-05,
+      "loss": 0.019,
+      "step": 26100
+    },
+    {
+      "epoch": 37.5,
+      "eval_loss": 0.13397379219532013,
+      "eval_runtime": 40.1527,
+      "eval_samples_per_second": 11.382,
+      "eval_steps_per_second": 1.444,
+      "eval_wer": 0.13100436681222707,
+      "step": 26100
+    },
+    {
+      "epoch": 37.51,
+      "learning_rate": 7.60932944606414e-05,
+      "loss": 0.0264,
+      "step": 26110
+    },
+    {
+      "epoch": 37.53,
+      "learning_rate": 7.600583090379008e-05,
+      "loss": 0.0243,
+      "step": 26120
+    },
+    {
+      "epoch": 37.54,
+      "learning_rate": 7.591836734693876e-05,
+      "loss": 0.0252,
+      "step": 26130
+    },
+    {
+      "epoch": 37.56,
+      "learning_rate": 7.583090379008746e-05,
+      "loss": 0.0325,
+      "step": 26140
+    },
+    {
+      "epoch": 37.57,
+      "learning_rate": 7.574344023323615e-05,
+      "loss": 0.0228,
+      "step": 26150
+    },
+    {
+      "epoch": 37.59,
+      "learning_rate": 7.565597667638483e-05,
+      "loss": 0.028,
+      "step": 26160
+    },
+    {
+      "epoch": 37.6,
+      "learning_rate": 7.556851311953351e-05,
+      "loss": 0.0175,
+      "step": 26170
+    },
+    {
+      "epoch": 37.61,
+      "learning_rate": 7.548104956268221e-05,
+      "loss": 0.0315,
+      "step": 26180
+    },
+    {
+      "epoch": 37.63,
+      "learning_rate": 7.53935860058309e-05,
+      "loss": 0.0335,
+      "step": 26190
+    },
+    {
+      "epoch": 37.64,
+      "learning_rate": 7.530612244897959e-05,
+      "loss": 0.018,
+      "step": 26200
+    },
+    {
+      "epoch": 37.64,
+      "eval_loss": 0.13758954405784607,
+      "eval_runtime": 40.5804,
+      "eval_samples_per_second": 11.262,
+      "eval_steps_per_second": 1.429,
+      "eval_wer": 0.13326306279174824,
+      "step": 26200
+    },
+    {
+      "epoch": 37.66,
+      "learning_rate": 7.521865889212829e-05,
+      "loss": 0.0352,
+      "step": 26210
+    },
+    {
+      "epoch": 37.67,
+      "learning_rate": 7.513119533527696e-05,
+      "loss": 0.0237,
+      "step": 26220
+    },
+    {
+      "epoch": 37.69,
+      "learning_rate": 7.504373177842565e-05,
+      "loss": 0.0214,
+      "step": 26230
+    },
+    {
+      "epoch": 37.7,
+      "learning_rate": 7.495626822157434e-05,
+      "loss": 0.0237,
+      "step": 26240
+    },
+    {
+      "epoch": 37.72,
+      "learning_rate": 7.486880466472303e-05,
+      "loss": 0.0248,
+      "step": 26250
+    },
+    {
+      "epoch": 37.73,
+      "learning_rate": 7.478134110787171e-05,
+      "loss": 0.0333,
+      "step": 26260
+    },
+    {
+      "epoch": 37.74,
+      "learning_rate": 7.46938775510204e-05,
+      "loss": 0.0279,
+      "step": 26270
+    },
+    {
+      "epoch": 37.76,
+      "learning_rate": 7.460641399416909e-05,
+      "loss": 0.0245,
+      "step": 26280
+    },
+    {
+      "epoch": 37.77,
+      "learning_rate": 7.451895043731778e-05,
+      "loss": 0.0281,
+      "step": 26290
+    },
+    {
+      "epoch": 37.79,
+      "learning_rate": 7.443148688046646e-05,
+      "loss": 0.0254,
+      "step": 26300
+    },
+    {
+      "epoch": 37.79,
+      "eval_loss": 0.13217325508594513,
+      "eval_runtime": 40.6524,
+      "eval_samples_per_second": 11.242,
+      "eval_steps_per_second": 1.427,
+      "eval_wer": 0.13175726547206745,
+      "step": 26300
+    },
+    {
+      "epoch": 37.8,
+      "learning_rate": 7.434402332361515e-05,
+      "loss": 0.0281,
+      "step": 26310
+    },
+    {
+      "epoch": 37.82,
+      "learning_rate": 7.425655976676384e-05,
+      "loss": 0.0175,
+      "step": 26320
+    },
+    {
+      "epoch": 37.83,
+      "learning_rate": 7.416909620991253e-05,
+      "loss": 0.0375,
+      "step": 26330
+    },
+    {
+      "epoch": 37.84,
+      "learning_rate": 7.408163265306122e-05,
+      "loss": 0.0229,
+      "step": 26340
+    },
+    {
+      "epoch": 37.86,
+      "learning_rate": 7.39941690962099e-05,
+      "loss": 0.0193,
+      "step": 26350
+    },
+    {
+      "epoch": 37.87,
+      "learning_rate": 7.390670553935859e-05,
+      "loss": 0.0269,
+      "step": 26360
+    },
+    {
+      "epoch": 37.89,
+      "learning_rate": 7.381924198250728e-05,
+      "loss": 0.0233,
+      "step": 26370
+    },
+    {
+      "epoch": 37.9,
+      "learning_rate": 7.373177842565597e-05,
+      "loss": 0.0209,
+      "step": 26380
+    },
+    {
+      "epoch": 37.92,
+      "learning_rate": 7.364431486880466e-05,
+      "loss": 0.0232,
+      "step": 26390
+    },
+    {
+      "epoch": 37.93,
+      "learning_rate": 7.355685131195334e-05,
+      "loss": 0.0188,
+      "step": 26400
+    },
+    {
+      "epoch": 37.93,
+      "eval_loss": 0.1334916353225708,
+      "eval_runtime": 40.3434,
+      "eval_samples_per_second": 11.328,
+      "eval_steps_per_second": 1.438,
+      "eval_wer": 0.1263363951212167,
+      "step": 26400
+    },
+    {
+      "epoch": 37.95,
+      "learning_rate": 7.346938775510203e-05,
+      "loss": 0.0282,
+      "step": 26410
+    },
+    {
+      "epoch": 37.96,
+      "learning_rate": 7.338192419825072e-05,
+      "loss": 0.0246,
+      "step": 26420
+    },
+    {
+      "epoch": 37.97,
+      "learning_rate": 7.329446064139941e-05,
+      "loss": 0.0293,
+      "step": 26430
+    },
+    {
+      "epoch": 37.99,
+      "learning_rate": 7.32069970845481e-05,
+      "loss": 0.0206,
+      "step": 26440
+    },
+    {
+      "epoch": 38.0,
+      "learning_rate": 7.311953352769678e-05,
+      "loss": 0.0228,
+      "step": 26450
+    },
+    {
+      "epoch": 38.02,
+      "learning_rate": 7.303206997084547e-05,
+      "loss": 0.0207,
+      "step": 26460
+    },
+    {
+      "epoch": 38.03,
+      "learning_rate": 7.294460641399417e-05,
+      "loss": 0.0213,
+      "step": 26470
+    },
+    {
+      "epoch": 38.05,
+      "learning_rate": 7.285714285714285e-05,
+      "loss": 0.0211,
+      "step": 26480
+    },
+    {
+      "epoch": 38.06,
+      "learning_rate": 7.276967930029155e-05,
+      "loss": 0.029,
+      "step": 26490
+    },
+    {
+      "epoch": 38.07,
+      "learning_rate": 7.268221574344022e-05,
+      "loss": 0.023,
+      "step": 26500
+    },
+    {
+      "epoch": 38.07,
+      "eval_loss": 0.13524821400642395,
+      "eval_runtime": 40.7993,
+      "eval_samples_per_second": 11.201,
+      "eval_steps_per_second": 1.422,
+      "eval_wer": 0.1281433519048336,
+      "step": 26500
+    },
+    {
+      "epoch": 38.09,
+      "learning_rate": 7.259475218658891e-05,
+      "loss": 0.026,
+      "step": 26510
+    },
+    {
+      "epoch": 38.1,
+      "learning_rate": 7.25072886297376e-05,
+      "loss": 0.0165,
+      "step": 26520
+    },
+    {
+      "epoch": 38.12,
+      "learning_rate": 7.241982507288629e-05,
+      "loss": 0.03,
+      "step": 26530
+    },
+    {
+      "epoch": 38.13,
+      "learning_rate": 7.233236151603497e-05,
+      "loss": 0.0252,
+      "step": 26540
+    },
+    {
+      "epoch": 38.15,
+      "learning_rate": 7.224489795918366e-05,
+      "loss": 0.0309,
+      "step": 26550
+    },
+    {
+      "epoch": 38.16,
+      "learning_rate": 7.215743440233236e-05,
+      "loss": 0.0275,
+      "step": 26560
+    },
+    {
+      "epoch": 38.18,
+      "learning_rate": 7.206997084548104e-05,
+      "loss": 0.0196,
+      "step": 26570
+    },
+    {
+      "epoch": 38.19,
+      "learning_rate": 7.198250728862974e-05,
+      "loss": 0.0357,
+      "step": 26580
+    },
+    {
+      "epoch": 38.2,
+      "learning_rate": 7.189504373177841e-05,
+      "loss": 0.0248,
+      "step": 26590
+    },
+    {
+      "epoch": 38.22,
+      "learning_rate": 7.180758017492712e-05,
+      "loss": 0.0198,
+      "step": 26600
+    },
+    {
+      "epoch": 38.22,
+      "eval_loss": 0.12783843278884888,
+      "eval_runtime": 40.4736,
+      "eval_samples_per_second": 11.291,
+      "eval_steps_per_second": 1.433,
+      "eval_wer": 0.12513175726547207,
+      "step": 26600
+    },
+    {
+      "epoch": 38.23,
+      "learning_rate": 7.172011661807579e-05,
+      "loss": 0.0328,
+      "step": 26610
+    },
+    {
+      "epoch": 38.25,
+      "learning_rate": 7.163265306122449e-05,
+      "loss": 0.0212,
+      "step": 26620
+    },
+    {
+      "epoch": 38.26,
+      "learning_rate": 7.154518950437317e-05,
+      "loss": 0.0277,
+      "step": 26630
+    },
+    {
+      "epoch": 38.28,
+      "learning_rate": 7.145772594752185e-05,
+      "loss": 0.0255,
+      "step": 26640
+    },
+    {
+      "epoch": 38.29,
+      "learning_rate": 7.137026239067055e-05,
+      "loss": 0.0212,
+      "step": 26650
+    },
+    {
+      "epoch": 38.3,
+      "learning_rate": 7.128279883381923e-05,
+      "loss": 0.0313,
+      "step": 26660
+    },
+    {
+      "epoch": 38.32,
+      "learning_rate": 7.119533527696793e-05,
+      "loss": 0.0216,
+      "step": 26670
+    },
+    {
+      "epoch": 38.33,
+      "learning_rate": 7.11078717201166e-05,
+      "loss": 0.0226,
+      "step": 26680
+    },
+    {
+      "epoch": 38.35,
+      "learning_rate": 7.10204081632653e-05,
+      "loss": 0.0209,
+      "step": 26690
+    },
+    {
+      "epoch": 38.36,
+      "learning_rate": 7.093294460641398e-05,
+      "loss": 0.0182,
+      "step": 26700
+    },
+    {
+      "epoch": 38.36,
+      "eval_loss": 0.13292567431926727,
+      "eval_runtime": 40.4251,
+      "eval_samples_per_second": 11.305,
+      "eval_steps_per_second": 1.435,
+      "eval_wer": 0.12769161270892937,
+      "step": 26700
+    },
+    {
+      "epoch": 38.38,
+      "learning_rate": 7.084548104956268e-05,
+      "loss": 0.0236,
+      "step": 26710
+    },
+    {
+      "epoch": 38.39,
+      "learning_rate": 7.075801749271136e-05,
+      "loss": 0.0209,
+      "step": 26720
+    },
+    {
+      "epoch": 38.41,
+      "learning_rate": 7.067055393586006e-05,
+      "loss": 0.026,
+      "step": 26730
+    },
+    {
+      "epoch": 38.42,
+      "learning_rate": 7.058309037900875e-05,
+      "loss": 0.0235,
+      "step": 26740
+    },
+    {
+      "epoch": 38.43,
+      "learning_rate": 7.049562682215743e-05,
+      "loss": 0.0405,
+      "step": 26750
+    },
+    {
+      "epoch": 38.45,
+      "learning_rate": 7.040816326530612e-05,
+      "loss": 0.0295,
+      "step": 26760
+    },
+    {
+      "epoch": 38.46,
+      "learning_rate": 7.032069970845481e-05,
+      "loss": 0.0203,
+      "step": 26770
+    },
+    {
+      "epoch": 38.48,
+      "learning_rate": 7.02332361516035e-05,
+      "loss": 0.0207,
+      "step": 26780
+    },
+    {
+      "epoch": 38.49,
+      "learning_rate": 7.014577259475217e-05,
+      "loss": 0.0239,
+      "step": 26790
+    },
+    {
+      "epoch": 38.51,
+      "learning_rate": 7.005830903790087e-05,
+      "loss": 0.0257,
+      "step": 26800
+    },
+    {
+      "epoch": 38.51,
+      "eval_loss": 0.1308964639902115,
+      "eval_runtime": 40.3719,
+      "eval_samples_per_second": 11.32,
+      "eval_steps_per_second": 1.437,
+      "eval_wer": 0.1281433519048336,
+      "step": 26800
+    },
+    {
+      "epoch": 38.52,
+      "learning_rate": 6.997084548104955e-05,
+      "loss": 0.0285,
+      "step": 26810
+    },
+    {
+      "epoch": 38.53,
+      "learning_rate": 6.988338192419825e-05,
+      "loss": 0.0245,
+      "step": 26820
+    },
+    {
+      "epoch": 38.55,
+      "learning_rate": 6.979591836734692e-05,
+      "loss": 0.0231,
+      "step": 26830
+    },
+    {
+      "epoch": 38.56,
+      "learning_rate": 6.970845481049562e-05,
+      "loss": 0.032,
+      "step": 26840
+    },
+    {
+      "epoch": 38.58,
+      "learning_rate": 6.962099125364431e-05,
+      "loss": 0.0209,
+      "step": 26850
+    },
+    {
+      "epoch": 38.59,
+      "learning_rate": 6.9533527696793e-05,
+      "loss": 0.0254,
+      "step": 26860
+    },
+    {
+      "epoch": 38.61,
+      "learning_rate": 6.944606413994169e-05,
+      "loss": 0.022,
+      "step": 26870
+    },
+    {
+      "epoch": 38.62,
+      "learning_rate": 6.935860058309038e-05,
+      "loss": 0.022,
+      "step": 26880
+    },
+    {
+      "epoch": 38.64,
+      "learning_rate": 6.927113702623906e-05,
+      "loss": 0.0182,
+      "step": 26890
+    },
+    {
+      "epoch": 38.65,
+      "learning_rate": 6.918367346938775e-05,
+      "loss": 0.0259,
+      "step": 26900
+    },
+    {
+      "epoch": 38.65,
+      "eval_loss": 0.1267746239900589,
+      "eval_runtime": 40.23,
+      "eval_samples_per_second": 11.36,
+      "eval_steps_per_second": 1.442,
+      "eval_wer": 0.1281433519048336,
+      "step": 26900
+    },
+    {
+      "epoch": 38.66,
+      "learning_rate": 6.909620991253644e-05,
+      "loss": 0.034,
+      "step": 26910
+    },
+    {
+      "epoch": 38.68,
+      "learning_rate": 6.900874635568513e-05,
+      "loss": 0.0205,
+      "step": 26920
+    },
+    {
+      "epoch": 38.69,
+      "learning_rate": 6.892128279883382e-05,
+      "loss": 0.0194,
+      "step": 26930
+    },
+    {
+      "epoch": 38.71,
+      "learning_rate": 6.88338192419825e-05,
+      "loss": 0.0293,
+      "step": 26940
+    },
+    {
+      "epoch": 38.72,
+      "learning_rate": 6.874635568513119e-05,
+      "loss": 0.0194,
+      "step": 26950
+    },
+    {
+      "epoch": 38.74,
+      "learning_rate": 6.865889212827988e-05,
+      "loss": 0.026,
+      "step": 26960
+    },
+    {
+      "epoch": 38.75,
+      "learning_rate": 6.857142857142857e-05,
+      "loss": 0.0182,
+      "step": 26970
+    },
+    {
+      "epoch": 38.76,
+      "learning_rate": 6.848396501457726e-05,
+      "loss": 0.0279,
+      "step": 26980
+    },
+    {
+      "epoch": 38.78,
+      "learning_rate": 6.839650145772594e-05,
+      "loss": 0.0234,
+      "step": 26990
+    },
+    {
+      "epoch": 38.79,
+      "learning_rate": 6.830903790087463e-05,
+      "loss": 0.0306,
+      "step": 27000
+    },
+    {
+      "epoch": 38.79,
+      "eval_loss": 0.13185811042785645,
+      "eval_runtime": 40.2584,
+      "eval_samples_per_second": 11.352,
+      "eval_steps_per_second": 1.441,
+      "eval_wer": 0.12964914922451437,
+      "step": 27000
+    },
+    {
+      "epoch": 38.81,
+      "learning_rate": 6.822157434402332e-05,
+      "loss": 0.0247,
+      "step": 27010
+    },
+    {
+      "epoch": 38.82,
+      "learning_rate": 6.813411078717201e-05,
+      "loss": 0.018,
+      "step": 27020
+    },
+    {
+      "epoch": 38.84,
+      "learning_rate": 6.80466472303207e-05,
+      "loss": 0.022,
+      "step": 27030
+    },
+    {
+      "epoch": 38.85,
+      "learning_rate": 6.795918367346938e-05,
+      "loss": 0.0232,
+      "step": 27040
+    },
+    {
+      "epoch": 38.86,
+      "learning_rate": 6.787172011661807e-05,
+      "loss": 0.0234,
+      "step": 27050
+    },
+    {
+      "epoch": 38.88,
+      "learning_rate": 6.778425655976676e-05,
+      "loss": 0.0329,
+      "step": 27060
+    },
+    {
+      "epoch": 38.89,
+      "learning_rate": 6.769679300291545e-05,
+      "loss": 0.0216,
+      "step": 27070
+    },
+    {
+      "epoch": 38.91,
+      "learning_rate": 6.760932944606413e-05,
+      "loss": 0.0252,
+      "step": 27080
+    },
+    {
+      "epoch": 38.92,
+      "learning_rate": 6.752186588921282e-05,
+      "loss": 0.022,
+      "step": 27090
+    },
+    {
+      "epoch": 38.94,
+      "learning_rate": 6.743440233236151e-05,
+      "loss": 0.0212,
+      "step": 27100
+    },
+    {
+      "epoch": 38.94,
+      "eval_loss": 0.12713249027729034,
+      "eval_runtime": 40.5764,
+      "eval_samples_per_second": 11.263,
+      "eval_steps_per_second": 1.429,
+      "eval_wer": 0.1275410329769613,
+      "step": 27100
+    },
+    {
+      "epoch": 38.95,
+      "learning_rate": 6.73469387755102e-05,
+      "loss": 0.0213,
+      "step": 27110
+    },
+    {
+      "epoch": 38.97,
+      "learning_rate": 6.725947521865889e-05,
+      "loss": 0.0196,
+      "step": 27120
+    },
+    {
+      "epoch": 38.98,
+      "learning_rate": 6.717201166180757e-05,
+      "loss": 0.0269,
+      "step": 27130
+    },
+    {
+      "epoch": 38.99,
+      "learning_rate": 6.708454810495626e-05,
+      "loss": 0.0212,
+      "step": 27140
+    },
+    {
+      "epoch": 39.01,
+      "learning_rate": 6.699708454810495e-05,
+      "loss": 0.0208,
+      "step": 27150
+    },
+    {
+      "epoch": 39.02,
+      "learning_rate": 6.690962099125364e-05,
+      "loss": 0.0191,
+      "step": 27160
+    },
+    {
+      "epoch": 39.04,
+      "learning_rate": 6.682215743440233e-05,
+      "loss": 0.02,
+      "step": 27170
+    },
+    {
+      "epoch": 39.05,
+      "learning_rate": 6.673469387755101e-05,
+      "loss": 0.0265,
+      "step": 27180
+    },
+    {
+      "epoch": 39.07,
+      "learning_rate": 6.66472303206997e-05,
+      "loss": 0.0179,
+      "step": 27190
+    },
+    {
+      "epoch": 39.08,
+      "learning_rate": 6.655976676384839e-05,
+      "loss": 0.0214,
+      "step": 27200
+    },
+    {
+      "epoch": 39.08,
+      "eval_loss": 0.13211554288864136,
+      "eval_runtime": 40.5187,
+      "eval_samples_per_second": 11.279,
+      "eval_steps_per_second": 1.431,
+      "eval_wer": 0.12739045324499323,
+      "step": 27200
+    },
+    {
+      "epoch": 39.09,
+      "learning_rate": 6.647230320699708e-05,
+      "loss": 0.0229,
+      "step": 27210
+    },
+    {
+      "epoch": 39.11,
+      "learning_rate": 6.638483965014576e-05,
+      "loss": 0.0187,
+      "step": 27220
+    },
+    {
+      "epoch": 39.12,
+      "learning_rate": 6.629737609329445e-05,
+      "loss": 0.0347,
+      "step": 27230
+    },
+    {
+      "epoch": 39.14,
+      "learning_rate": 6.620991253644314e-05,
+      "loss": 0.0202,
+      "step": 27240
+    },
+    {
+      "epoch": 39.15,
+      "learning_rate": 6.612244897959183e-05,
+      "loss": 0.0274,
+      "step": 27250
+    },
+    {
+      "epoch": 39.17,
+      "learning_rate": 6.603498542274052e-05,
+      "loss": 0.0294,
+      "step": 27260
+    },
+    {
+      "epoch": 39.18,
+      "learning_rate": 6.59475218658892e-05,
+      "loss": 0.0238,
+      "step": 27270
+    },
+    {
+      "epoch": 39.2,
+      "learning_rate": 6.586005830903789e-05,
+      "loss": 0.0228,
+      "step": 27280
+    },
+    {
+      "epoch": 39.21,
+      "learning_rate": 6.577259475218658e-05,
+      "loss": 0.0224,
+      "step": 27290
+    },
+    {
+      "epoch": 39.22,
+      "learning_rate": 6.568513119533528e-05,
+      "loss": 0.0385,
+      "step": 27300
+    },
+    {
+      "epoch": 39.22,
+      "eval_loss": 0.13131336867809296,
+      "eval_runtime": 40.7082,
+      "eval_samples_per_second": 11.226,
+      "eval_steps_per_second": 1.425,
+      "eval_wer": 0.128896250564674,
+      "step": 27300
+    },
+    {
+      "epoch": 39.24,
+      "learning_rate": 6.559766763848396e-05,
+      "loss": 0.0241,
+      "step": 27310
+    },
+    {
+      "epoch": 39.25,
+      "learning_rate": 6.551020408163264e-05,
+      "loss": 0.0323,
+      "step": 27320
+    },
+    {
+      "epoch": 39.27,
+      "learning_rate": 6.542274052478133e-05,
+      "loss": 0.0283,
+      "step": 27330
+    },
+    {
+      "epoch": 39.28,
+      "learning_rate": 6.533527696793002e-05,
+      "loss": 0.0223,
+      "step": 27340
+    },
+    {
+      "epoch": 39.3,
+      "learning_rate": 6.524781341107871e-05,
+      "loss": 0.023,
+      "step": 27350
+    },
+    {
+      "epoch": 39.31,
+      "learning_rate": 6.51603498542274e-05,
+      "loss": 0.0232,
+      "step": 27360
+    },
+    {
+      "epoch": 39.32,
+      "learning_rate": 6.507288629737608e-05,
+      "loss": 0.0214,
+      "step": 27370
+    },
+    {
+      "epoch": 39.34,
+      "learning_rate": 6.498542274052477e-05,
+      "loss": 0.0239,
+      "step": 27380
+    },
+    {
+      "epoch": 39.35,
+      "learning_rate": 6.489795918367347e-05,
+      "loss": 0.0254,
+      "step": 27390
+    },
+    {
+      "epoch": 39.37,
+      "learning_rate": 6.481049562682215e-05,
+      "loss": 0.0315,
+      "step": 27400
+    },
+    {
+      "epoch": 39.37,
+      "eval_loss": 0.1282867193222046,
+      "eval_runtime": 40.3945,
+      "eval_samples_per_second": 11.313,
+      "eval_steps_per_second": 1.436,
+      "eval_wer": 0.12904683029664207,
+      "step": 27400
+    },
+    {
+      "epoch": 39.38,
+      "learning_rate": 6.472303206997085e-05,
+      "loss": 0.0206,
+      "step": 27410
+    },
+    {
+      "epoch": 39.4,
+      "learning_rate": 6.463556851311952e-05,
+      "loss": 0.0206,
+      "step": 27420
+    },
+    {
+      "epoch": 39.41,
+      "learning_rate": 6.454810495626822e-05,
+      "loss": 0.0248,
+      "step": 27430
+    },
+    {
+      "epoch": 39.43,
+      "learning_rate": 6.44606413994169e-05,
+      "loss": 0.0141,
+      "step": 27440
+    },
+    {
+      "epoch": 39.44,
+      "learning_rate": 6.437317784256559e-05,
+      "loss": 0.0348,
+      "step": 27450
+    },
+    {
+      "epoch": 39.45,
+      "learning_rate": 6.428571428571427e-05,
+      "loss": 0.0207,
+      "step": 27460
+    },
+    {
+      "epoch": 39.47,
+      "learning_rate": 6.419825072886296e-05,
+      "loss": 0.0248,
+      "step": 27470
+    },
+    {
+      "epoch": 39.48,
+      "learning_rate": 6.411078717201166e-05,
+      "loss": 0.0209,
+      "step": 27480
+    },
+    {
+      "epoch": 39.5,
+      "learning_rate": 6.402332361516034e-05,
+      "loss": 0.0215,
+      "step": 27490
+    },
+    {
+      "epoch": 39.51,
+      "learning_rate": 6.393586005830904e-05,
+      "loss": 0.0208,
+      "step": 27500
+    },
+    {
+      "epoch": 39.51,
+      "eval_loss": 0.1325148493051529,
+      "eval_runtime": 40.3921,
+      "eval_samples_per_second": 11.314,
+      "eval_steps_per_second": 1.436,
+      "eval_wer": 0.1263363951212167,
+      "step": 27500
+    },
+    {
+      "epoch": 39.53,
+      "learning_rate": 6.384839650145771e-05,
+      "loss": 0.0152,
+      "step": 27510
+    },
+    {
+      "epoch": 39.54,
+      "learning_rate": 6.376093294460642e-05,
+      "loss": 0.0182,
+      "step": 27520
+    },
+    {
+      "epoch": 39.55,
+      "learning_rate": 6.367346938775509e-05,
+      "loss": 0.0226,
+      "step": 27530
+    },
+    {
+      "epoch": 39.57,
+      "learning_rate": 6.358600583090379e-05,
+      "loss": 0.0204,
+      "step": 27540
+    },
+    {
+      "epoch": 39.58,
+      "learning_rate": 6.349854227405247e-05,
+      "loss": 0.024,
+      "step": 27550
+    },
+    {
+      "epoch": 39.6,
+      "learning_rate": 6.341107871720117e-05,
+      "loss": 0.0212,
+      "step": 27560
+    },
+    {
+      "epoch": 39.61,
+      "learning_rate": 6.332361516034985e-05,
+      "loss": 0.0247,
+      "step": 27570
+    },
+    {
+      "epoch": 39.63,
+      "learning_rate": 6.323615160349854e-05,
+      "loss": 0.0294,
+      "step": 27580
+    },
+    {
+      "epoch": 39.64,
+      "learning_rate": 6.314868804664723e-05,
+      "loss": 0.0198,
+      "step": 27590
+    },
+    {
+      "epoch": 39.66,
+      "learning_rate": 6.30612244897959e-05,
+      "loss": 0.029,
+      "step": 27600
+    },
+    {
+      "epoch": 39.66,
+      "eval_loss": 0.12957048416137695,
+      "eval_runtime": 40.5377,
+      "eval_samples_per_second": 11.273,
+      "eval_steps_per_second": 1.431,
+      "eval_wer": 0.12919741002861015,
+      "step": 27600
+    },
+    {
+      "epoch": 39.67,
+      "learning_rate": 6.29737609329446e-05,
+      "loss": 0.0228,
+      "step": 27610
+    },
+    {
+      "epoch": 39.68,
+      "learning_rate": 6.288629737609328e-05,
+      "loss": 0.0301,
+      "step": 27620
+    },
+    {
+      "epoch": 39.7,
+      "learning_rate": 6.279883381924198e-05,
+      "loss": 0.0275,
+      "step": 27630
+    },
+    {
+      "epoch": 39.71,
+      "learning_rate": 6.271137026239066e-05,
+      "loss": 0.0215,
+      "step": 27640
+    },
+    {
+      "epoch": 39.73,
+      "learning_rate": 6.262390670553936e-05,
+      "loss": 0.0207,
+      "step": 27650
+    },
+    {
+      "epoch": 39.74,
+      "learning_rate": 6.253644314868805e-05,
+      "loss": 0.0204,
+      "step": 27660
+    },
+    {
+      "epoch": 39.76,
+      "learning_rate": 6.244897959183673e-05,
+      "loss": 0.0222,
+      "step": 27670
+    },
+    {
+      "epoch": 39.77,
+      "learning_rate": 6.236151603498542e-05,
+      "loss": 0.0264,
+      "step": 27680
+    },
+    {
+      "epoch": 39.78,
+      "learning_rate": 6.227405247813411e-05,
+      "loss": 0.0174,
+      "step": 27690
+    },
+    {
+      "epoch": 39.8,
+      "learning_rate": 6.21865889212828e-05,
+      "loss": 0.0263,
+      "step": 27700
+    },
+    {
+      "epoch": 39.8,
+      "eval_loss": 0.12812817096710205,
+      "eval_runtime": 41.192,
+      "eval_samples_per_second": 11.094,
+      "eval_steps_per_second": 1.408,
+      "eval_wer": 0.12723987351302515,
+      "step": 27700
+    },
+    {
+      "epoch": 39.81,
+      "learning_rate": 6.209912536443149e-05,
+      "loss": 0.0186,
+      "step": 27710
+    },
+    {
+      "epoch": 39.83,
+      "learning_rate": 6.201166180758017e-05,
+      "loss": 0.0252,
+      "step": 27720
+    },
+    {
+      "epoch": 39.84,
+      "learning_rate": 6.192419825072885e-05,
+      "loss": 0.0235,
+      "step": 27730
+    },
+    {
+      "epoch": 39.86,
+      "learning_rate": 6.183673469387755e-05,
+      "loss": 0.0255,
+      "step": 27740
+    },
+    {
+      "epoch": 39.87,
+      "learning_rate": 6.174927113702624e-05,
+      "loss": 0.0327,
+      "step": 27750
+    },
+    {
+      "epoch": 39.89,
+      "learning_rate": 6.166180758017492e-05,
+      "loss": 0.0238,
+      "step": 27760
+    },
+    {
+      "epoch": 39.9,
+      "learning_rate": 6.157434402332361e-05,
+      "loss": 0.0144,
+      "step": 27770
+    },
+    {
+      "epoch": 39.91,
+      "learning_rate": 6.14868804664723e-05,
+      "loss": 0.0244,
+      "step": 27780
+    },
+    {
+      "epoch": 39.93,
+      "learning_rate": 6.139941690962099e-05,
+      "loss": 0.0205,
+      "step": 27790
+    },
+    {
+      "epoch": 39.94,
+      "learning_rate": 6.131195335276968e-05,
+      "loss": 0.0234,
+      "step": 27800
+    },
+    {
+      "epoch": 39.94,
+      "eval_loss": 0.12909665703773499,
+      "eval_runtime": 40.2598,
+      "eval_samples_per_second": 11.351,
+      "eval_steps_per_second": 1.441,
+      "eval_wer": 0.12498117753350399,
+      "step": 27800
+    },
+    {
+      "epoch": 39.96,
+      "learning_rate": 6.122448979591836e-05,
+      "loss": 0.0253,
+      "step": 27810
+    },
+    {
+      "epoch": 39.97,
+      "learning_rate": 6.113702623906705e-05,
+      "loss": 0.0199,
+      "step": 27820
+    },
+    {
+      "epoch": 39.99,
+      "learning_rate": 6.104956268221574e-05,
+      "loss": 0.0212,
+      "step": 27830
+    },
+    {
+      "epoch": 40.0,
+      "learning_rate": 6.096209912536443e-05,
+      "loss": 0.0206,
+      "step": 27840
+    },
+    {
+      "epoch": 40.01,
+      "learning_rate": 6.0874635568513116e-05,
+      "loss": 0.0249,
+      "step": 27850
+    },
+    {
+      "epoch": 40.03,
+      "learning_rate": 6.0787172011661804e-05,
+      "loss": 0.021,
+      "step": 27860
+    },
+    {
+      "epoch": 40.04,
+      "learning_rate": 6.069970845481049e-05,
+      "loss": 0.027,
+      "step": 27870
+    },
+    {
+      "epoch": 40.06,
+      "learning_rate": 6.061224489795918e-05,
+      "loss": 0.0231,
+      "step": 27880
+    },
+    {
+      "epoch": 40.07,
+      "learning_rate": 6.052478134110787e-05,
+      "loss": 0.0224,
+      "step": 27890
+    },
+    {
+      "epoch": 40.09,
+      "learning_rate": 6.0437317784256555e-05,
+      "loss": 0.0229,
+      "step": 27900
+    },
+    {
+      "epoch": 40.09,
+      "eval_loss": 0.13076798617839813,
+      "eval_runtime": 40.4671,
+      "eval_samples_per_second": 11.293,
+      "eval_steps_per_second": 1.433,
+      "eval_wer": 0.1287456708327059,
+      "step": 27900
+    },
+    {
+      "epoch": 40.1,
+      "learning_rate": 6.0349854227405236e-05,
+      "loss": 0.0167,
+      "step": 27910
+    },
+    {
+      "epoch": 40.11,
+      "learning_rate": 6.026239067055393e-05,
+      "loss": 0.0221,
+      "step": 27920
+    },
+    {
+      "epoch": 40.13,
+      "learning_rate": 6.017492711370261e-05,
+      "loss": 0.0211,
+      "step": 27930
+    },
+    {
+      "epoch": 40.14,
+      "learning_rate": 6.008746355685131e-05,
+      "loss": 0.0211,
+      "step": 27940
+    },
+    {
+      "epoch": 40.16,
+      "learning_rate": 5.9999999999999995e-05,
+      "loss": 0.0239,
+      "step": 27950
+    },
+    {
+      "epoch": 40.17,
+      "learning_rate": 5.991253644314868e-05,
+      "loss": 0.0241,
+      "step": 27960
+    },
+    {
+      "epoch": 40.19,
+      "learning_rate": 5.982507288629737e-05,
+      "loss": 0.029,
+      "step": 27970
+    },
+    {
+      "epoch": 40.2,
+      "learning_rate": 5.973760932944606e-05,
+      "loss": 0.0173,
+      "step": 27980
+    },
+    {
+      "epoch": 40.22,
+      "learning_rate": 5.9650145772594746e-05,
+      "loss": 0.019,
+      "step": 27990
+    },
+    {
+      "epoch": 40.23,
+      "learning_rate": 5.9562682215743434e-05,
+      "loss": 0.0254,
+      "step": 28000
+    },
+    {
+      "epoch": 40.23,
+      "eval_loss": 0.13149315118789673,
+      "eval_runtime": 40.5502,
+      "eval_samples_per_second": 11.27,
+      "eval_steps_per_second": 1.43,
+      "eval_wer": 0.12573407619334437,
+      "step": 28000
+    },
+    {
+      "epoch": 40.24,
+      "learning_rate": 5.947521865889212e-05,
+      "loss": 0.0205,
+      "step": 28010
+    },
+    {
+      "epoch": 40.26,
+      "learning_rate": 5.938775510204081e-05,
+      "loss": 0.0242,
+      "step": 28020
+    },
+    {
+      "epoch": 40.27,
+      "learning_rate": 5.93002915451895e-05,
+      "loss": 0.0186,
+      "step": 28030
+    },
+    {
+      "epoch": 40.29,
+      "learning_rate": 5.921282798833819e-05,
+      "loss": 0.0153,
+      "step": 28040
+    },
+    {
+      "epoch": 40.3,
+      "learning_rate": 5.9125364431486874e-05,
+      "loss": 0.0259,
+      "step": 28050
+    },
+    {
+      "epoch": 40.32,
+      "learning_rate": 5.903790087463557e-05,
+      "loss": 0.0264,
+      "step": 28060
+    },
+    {
+      "epoch": 40.33,
+      "learning_rate": 5.895043731778425e-05,
+      "loss": 0.0212,
+      "step": 28070
+    },
+    {
+      "epoch": 40.34,
+      "learning_rate": 5.8862973760932944e-05,
+      "loss": 0.0212,
+      "step": 28080
+    },
+    {
+      "epoch": 40.36,
+      "learning_rate": 5.8775510204081625e-05,
+      "loss": 0.0189,
+      "step": 28090
+    },
+    {
+      "epoch": 40.37,
+      "learning_rate": 5.868804664723031e-05,
+      "loss": 0.0266,
+      "step": 28100
+    },
+    {
+      "epoch": 40.37,
+      "eval_loss": 0.132347971200943,
+      "eval_runtime": 40.6163,
+      "eval_samples_per_second": 11.252,
+      "eval_steps_per_second": 1.428,
+      "eval_wer": 0.12498117753350399,
+      "step": 28100
+    },
+    {
+      "epoch": 40.39,
+      "learning_rate": 5.8600583090379e-05,
+      "loss": 0.0224,
+      "step": 28110
+    },
+    {
+      "epoch": 40.4,
+      "learning_rate": 5.851311953352769e-05,
+      "loss": 0.0255,
+      "step": 28120
+    },
+    {
+      "epoch": 40.42,
+      "learning_rate": 5.8425655976676384e-05,
+      "loss": 0.027,
+      "step": 28130
+    },
+    {
+      "epoch": 40.43,
+      "learning_rate": 5.8338192419825065e-05,
+      "loss": 0.022,
+      "step": 28140
+    },
+    {
+      "epoch": 40.45,
+      "learning_rate": 5.825072886297376e-05,
+      "loss": 0.0227,
+      "step": 28150
+    },
+    {
+      "epoch": 40.46,
+      "learning_rate": 5.816326530612244e-05,
+      "loss": 0.0202,
+      "step": 28160
+    },
+    {
+      "epoch": 40.47,
+      "learning_rate": 5.8075801749271135e-05,
+      "loss": 0.0226,
+      "step": 28170
+    },
+    {
+      "epoch": 40.49,
+      "learning_rate": 5.7988338192419816e-05,
+      "loss": 0.0195,
+      "step": 28180
+    },
+    {
+      "epoch": 40.5,
+      "learning_rate": 5.790087463556851e-05,
+      "loss": 0.0167,
+      "step": 28190
+    },
+    {
+      "epoch": 40.52,
+      "learning_rate": 5.781341107871719e-05,
+      "loss": 0.0275,
+      "step": 28200
+    },
+    {
+      "epoch": 40.52,
+      "eval_loss": 0.12815876305103302,
+      "eval_runtime": 40.653,
+      "eval_samples_per_second": 11.241,
+      "eval_steps_per_second": 1.427,
+      "eval_wer": 0.12422827887366361,
+      "step": 28200
+    },
+    {
+      "epoch": 40.53,
+      "learning_rate": 5.772594752186589e-05,
+      "loss": 0.0188,
+      "step": 28210
+    },
+    {
+      "epoch": 40.55,
+      "learning_rate": 5.7638483965014575e-05,
+      "loss": 0.0204,
+      "step": 28220
+    },
+    {
+      "epoch": 40.56,
+      "learning_rate": 5.7551020408163256e-05,
+      "loss": 0.0216,
+      "step": 28230
+    },
+    {
+      "epoch": 40.57,
+      "learning_rate": 5.746355685131195e-05,
+      "loss": 0.0156,
+      "step": 28240
+    },
+    {
+      "epoch": 40.59,
+      "learning_rate": 5.737609329446063e-05,
+      "loss": 0.0211,
+      "step": 28250
+    },
+    {
+      "epoch": 40.6,
+      "learning_rate": 5.7288629737609326e-05,
+      "loss": 0.0167,
+      "step": 28260
+    },
+    {
+      "epoch": 40.62,
+      "learning_rate": 5.720116618075801e-05,
+      "loss": 0.0221,
+      "step": 28270
+    },
+    {
+      "epoch": 40.63,
+      "learning_rate": 5.71137026239067e-05,
+      "loss": 0.0177,
+      "step": 28280
+    },
+    {
+      "epoch": 40.65,
+      "learning_rate": 5.702623906705538e-05,
+      "loss": 0.015,
+      "step": 28290
+    },
+    {
+      "epoch": 40.66,
+      "learning_rate": 5.693877551020408e-05,
+      "loss": 0.0266,
+      "step": 28300
+    },
+    {
+      "epoch": 40.66,
+      "eval_loss": 0.1275702863931656,
+      "eval_runtime": 40.6042,
+      "eval_samples_per_second": 11.255,
+      "eval_steps_per_second": 1.428,
+      "eval_wer": 0.12543291672940823,
+      "step": 28300
+    },
+    {
+      "epoch": 40.68,
+      "learning_rate": 5.6851311953352766e-05,
+      "loss": 0.0201,
+      "step": 28310
+    },
+    {
+      "epoch": 40.69,
+      "learning_rate": 5.6763848396501454e-05,
+      "loss": 0.0284,
+      "step": 28320
+    },
+    {
+      "epoch": 40.7,
+      "learning_rate": 5.667638483965014e-05,
+      "loss": 0.0228,
+      "step": 28330
+    },
+    {
+      "epoch": 40.72,
+      "learning_rate": 5.658892128279883e-05,
+      "loss": 0.0192,
+      "step": 28340
+    },
+    {
+      "epoch": 40.73,
+      "learning_rate": 5.650145772594752e-05,
+      "loss": 0.0214,
+      "step": 28350
+    },
+    {
+      "epoch": 40.75,
+      "learning_rate": 5.6413994169096205e-05,
+      "loss": 0.0156,
+      "step": 28360
+    },
+    {
+      "epoch": 40.76,
+      "learning_rate": 5.632653061224489e-05,
+      "loss": 0.0229,
+      "step": 28370
+    },
+    {
+      "epoch": 40.78,
+      "learning_rate": 5.6239067055393574e-05,
+      "loss": 0.0287,
+      "step": 28380
+    },
+    {
+      "epoch": 40.79,
+      "learning_rate": 5.615160349854227e-05,
+      "loss": 0.0176,
+      "step": 28390
+    },
+    {
+      "epoch": 40.8,
+      "learning_rate": 5.6064139941690964e-05,
+      "loss": 0.0237,
+      "step": 28400
+    },
+    {
+      "epoch": 40.8,
+      "eval_loss": 0.13049447536468506,
+      "eval_runtime": 40.609,
+      "eval_samples_per_second": 11.254,
+      "eval_steps_per_second": 1.428,
+      "eval_wer": 0.12543291672940823,
+      "step": 28400
+    },
+    {
+      "epoch": 40.82,
+      "learning_rate": 5.5976676384839645e-05,
+      "loss": 0.022,
+      "step": 28410
+    },
+    {
+      "epoch": 40.83,
+      "learning_rate": 5.588921282798834e-05,
+      "loss": 0.03,
+      "step": 28420
+    },
+    {
+      "epoch": 40.85,
+      "learning_rate": 5.580174927113702e-05,
+      "loss": 0.0224,
+      "step": 28430
+    },
+    {
+      "epoch": 40.86,
+      "learning_rate": 5.571428571428571e-05,
+      "loss": 0.0192,
+      "step": 28440
+    },
+    {
+      "epoch": 40.88,
+      "learning_rate": 5.5626822157434396e-05,
+      "loss": 0.0267,
+      "step": 28450
+    },
+    {
+      "epoch": 40.89,
+      "learning_rate": 5.5539358600583084e-05,
+      "loss": 0.0171,
+      "step": 28460
+    },
+    {
+      "epoch": 40.91,
+      "learning_rate": 5.545189504373177e-05,
+      "loss": 0.019,
+      "step": 28470
+    },
+    {
+      "epoch": 40.92,
+      "learning_rate": 5.536443148688046e-05,
+      "loss": 0.0261,
+      "step": 28480
+    },
+    {
+      "epoch": 40.93,
+      "learning_rate": 5.5276967930029155e-05,
+      "loss": 0.0168,
+      "step": 28490
+    },
+    {
+      "epoch": 40.95,
+      "learning_rate": 5.5189504373177836e-05,
+      "loss": 0.0229,
+      "step": 28500
+    },
+    {
+      "epoch": 40.95,
+      "eval_loss": 0.1289079487323761,
+      "eval_runtime": 40.595,
+      "eval_samples_per_second": 11.258,
+      "eval_steps_per_second": 1.429,
+      "eval_wer": 0.12588465592531245,
+      "step": 28500
+    },
+    {
+      "epoch": 40.96,
+      "learning_rate": 5.510204081632653e-05,
+      "loss": 0.0198,
+      "step": 28510
+    },
+    {
+      "epoch": 40.98,
+      "learning_rate": 5.501457725947521e-05,
+      "loss": 0.0157,
+      "step": 28520
+    },
+    {
+      "epoch": 40.99,
+      "learning_rate": 5.4927113702623906e-05,
+      "loss": 0.022,
+      "step": 28530
+    },
+    {
+      "epoch": 41.01,
+      "learning_rate": 5.483965014577259e-05,
+      "loss": 0.0195,
+      "step": 28540
+    },
+    {
+      "epoch": 41.02,
+      "learning_rate": 5.475218658892128e-05,
+      "loss": 0.0206,
+      "step": 28550
+    },
+    {
+      "epoch": 41.03,
+      "learning_rate": 5.466472303206996e-05,
+      "loss": 0.0188,
+      "step": 28560
+    },
+    {
+      "epoch": 41.05,
+      "learning_rate": 5.457725947521865e-05,
+      "loss": 0.0226,
+      "step": 28570
+    },
+    {
+      "epoch": 41.06,
+      "learning_rate": 5.4489795918367346e-05,
+      "loss": 0.0208,
+      "step": 28580
+    },
+    {
+      "epoch": 41.08,
+      "learning_rate": 5.440233236151603e-05,
+      "loss": 0.0192,
+      "step": 28590
+    },
+    {
+      "epoch": 41.09,
+      "learning_rate": 5.431486880466472e-05,
+      "loss": 0.0194,
+      "step": 28600
+    },
+    {
+      "epoch": 41.09,
+      "eval_loss": 0.12941156327724457,
+      "eval_runtime": 40.6472,
+      "eval_samples_per_second": 11.243,
+      "eval_steps_per_second": 1.427,
+      "eval_wer": 0.12377653967775937,
+      "step": 28600
+    },
+    {
+      "epoch": 41.11,
+      "learning_rate": 5.42274052478134e-05,
+      "loss": 0.0246,
+      "step": 28610
+    },
+    {
+      "epoch": 41.12,
+      "learning_rate": 5.41399416909621e-05,
+      "loss": 0.0232,
+      "step": 28620
+    },
+    {
+      "epoch": 41.14,
+      "learning_rate": 5.405247813411078e-05,
+      "loss": 0.0203,
+      "step": 28630
+    },
+    {
+      "epoch": 41.15,
+      "learning_rate": 5.396501457725947e-05,
+      "loss": 0.0203,
+      "step": 28640
+    },
+    {
+      "epoch": 41.16,
+      "learning_rate": 5.3877551020408154e-05,
+      "loss": 0.0254,
+      "step": 28650
+    },
+    {
+      "epoch": 41.18,
+      "learning_rate": 5.379008746355685e-05,
+      "loss": 0.0217,
+      "step": 28660
+    },
+    {
+      "epoch": 41.19,
+      "learning_rate": 5.370262390670554e-05,
+      "loss": 0.0227,
+      "step": 28670
+    },
+    {
+      "epoch": 41.21,
+      "learning_rate": 5.3615160349854225e-05,
+      "loss": 0.0204,
+      "step": 28680
+    },
+    {
+      "epoch": 41.22,
+      "learning_rate": 5.352769679300291e-05,
+      "loss": 0.0185,
+      "step": 28690
+    },
+    {
+      "epoch": 41.24,
+      "learning_rate": 5.34402332361516e-05,
+      "loss": 0.0222,
+      "step": 28700
+    },
+    {
+      "epoch": 41.24,
+      "eval_loss": 0.12966987490653992,
+      "eval_runtime": 41.1373,
+      "eval_samples_per_second": 11.109,
+      "eval_steps_per_second": 1.41,
+      "eval_wer": 0.12046378557446168,
+      "step": 28700
+    },
+    {
+      "epoch": 41.25,
+      "learning_rate": 5.335276967930029e-05,
+      "loss": 0.0133,
+      "step": 28710
+    },
+    {
+      "epoch": 41.26,
+      "learning_rate": 5.326530612244897e-05,
+      "loss": 0.0226,
+      "step": 28720
+    },
+    {
+      "epoch": 41.28,
+      "learning_rate": 5.3177842565597664e-05,
+      "loss": 0.0225,
+      "step": 28730
+    },
+    {
+      "epoch": 41.29,
+      "learning_rate": 5.3090379008746345e-05,
+      "loss": 0.0152,
+      "step": 28740
+    },
+    {
+      "epoch": 41.31,
+      "learning_rate": 5.300291545189504e-05,
+      "loss": 0.0199,
+      "step": 28750
+    },
+    {
+      "epoch": 41.32,
+      "learning_rate": 5.291545189504373e-05,
+      "loss": 0.0216,
+      "step": 28760
+    },
+    {
+      "epoch": 41.34,
+      "learning_rate": 5.2827988338192416e-05,
+      "loss": 0.0309,
+      "step": 28770
+    },
+    {
+      "epoch": 41.35,
+      "learning_rate": 5.2740524781341104e-05,
+      "loss": 0.0205,
+      "step": 28780
+    },
+    {
+      "epoch": 41.36,
+      "learning_rate": 5.265306122448979e-05,
+      "loss": 0.0252,
+      "step": 28790
+    },
+    {
+      "epoch": 41.38,
+      "learning_rate": 5.256559766763848e-05,
+      "loss": 0.024,
+      "step": 28800
+    },
+    {
+      "epoch": 41.38,
+      "eval_loss": 0.12944242358207703,
+      "eval_runtime": 40.5436,
+      "eval_samples_per_second": 11.272,
+      "eval_steps_per_second": 1.431,
+      "eval_wer": 0.11910856798674899,
+      "step": 28800
+    },
+    {
+      "epoch": 41.39,
+      "learning_rate": 5.247813411078717e-05,
+      "loss": 0.022,
+      "step": 28810
+    },
+    {
+      "epoch": 41.41,
+      "learning_rate": 5.2390670553935855e-05,
+      "loss": 0.0296,
+      "step": 28820
+    },
+    {
+      "epoch": 41.42,
+      "learning_rate": 5.230320699708454e-05,
+      "loss": 0.015,
+      "step": 28830
+    },
+    {
+      "epoch": 41.44,
+      "learning_rate": 5.221574344023323e-05,
+      "loss": 0.0226,
+      "step": 28840
+    },
+    {
+      "epoch": 41.45,
+      "learning_rate": 5.2128279883381926e-05,
+      "loss": 0.0188,
+      "step": 28850
+    },
+    {
+      "epoch": 41.47,
+      "learning_rate": 5.204081632653061e-05,
+      "loss": 0.0173,
+      "step": 28860
+    },
+    {
+      "epoch": 41.48,
+      "learning_rate": 5.19533527696793e-05,
+      "loss": 0.0203,
+      "step": 28870
+    },
+    {
+      "epoch": 41.49,
+      "learning_rate": 5.186588921282798e-05,
+      "loss": 0.0209,
+      "step": 28880
+    },
+    {
+      "epoch": 41.51,
+      "learning_rate": 5.177842565597668e-05,
+      "loss": 0.0245,
+      "step": 28890
+    },
+    {
+      "epoch": 41.52,
+      "learning_rate": 5.169096209912536e-05,
+      "loss": 0.0185,
+      "step": 28900
+    },
+    {
+      "epoch": 41.52,
+      "eval_loss": 0.13065434992313385,
+      "eval_runtime": 40.6007,
+      "eval_samples_per_second": 11.256,
+      "eval_steps_per_second": 1.429,
+      "eval_wer": 0.12121668423430207,
+      "step": 28900
+    },
+    {
+      "epoch": 41.54,
+      "learning_rate": 5.1603498542274046e-05,
+      "loss": 0.0172,
+      "step": 28910
+    },
+    {
+      "epoch": 41.55,
+      "learning_rate": 5.1516034985422734e-05,
+      "loss": 0.0213,
+      "step": 28920
+    },
+    {
+      "epoch": 41.57,
+      "learning_rate": 5.142857142857142e-05,
+      "loss": 0.0175,
+      "step": 28930
+    },
+    {
+      "epoch": 41.58,
+      "learning_rate": 5.134110787172011e-05,
+      "loss": 0.0166,
+      "step": 28940
+    },
+    {
+      "epoch": 41.59,
+      "learning_rate": 5.12536443148688e-05,
+      "loss": 0.0285,
+      "step": 28950
+    },
+    {
+      "epoch": 41.61,
+      "learning_rate": 5.117492711370262e-05,
+      "loss": 0.0188,
+      "step": 28960
+    },
+    {
+      "epoch": 41.62,
+      "learning_rate": 5.1087463556851305e-05,
+      "loss": 0.0234,
+      "step": 28970
+    },
+    {
+      "epoch": 41.64,
+      "learning_rate": 5.1e-05,
+      "loss": 0.0206,
+      "step": 28980
+    },
+    {
+      "epoch": 41.65,
+      "learning_rate": 5.091253644314868e-05,
+      "loss": 0.0241,
+      "step": 28990
+    },
+    {
+      "epoch": 41.67,
+      "learning_rate": 5.0825072886297375e-05,
+      "loss": 0.0199,
+      "step": 29000
+    },
+    {
+      "epoch": 41.67,
+      "eval_loss": 0.1251639872789383,
+      "eval_runtime": 40.5816,
+      "eval_samples_per_second": 11.261,
+      "eval_steps_per_second": 1.429,
+      "eval_wer": 0.12181900316217437,
+      "step": 29000
+    },
+    {
+      "epoch": 41.68,
+      "learning_rate": 5.0737609329446057e-05,
+      "loss": 0.0231,
+      "step": 29010
+    },
+    {
+      "epoch": 41.7,
+      "learning_rate": 5.065014577259475e-05,
+      "loss": 0.0183,
+      "step": 29020
+    },
+    {
+      "epoch": 41.71,
+      "learning_rate": 5.056268221574343e-05,
+      "loss": 0.0179,
+      "step": 29030
+    },
+    {
+      "epoch": 41.72,
+      "learning_rate": 5.047521865889213e-05,
+      "loss": 0.0214,
+      "step": 29040
+    },
+    {
+      "epoch": 41.74,
+      "learning_rate": 5.038775510204081e-05,
+      "loss": 0.0223,
+      "step": 29050
+    },
+    {
+      "epoch": 41.75,
+      "learning_rate": 5.03002915451895e-05,
+      "loss": 0.0151,
+      "step": 29060
+    },
+    {
+      "epoch": 41.77,
+      "learning_rate": 5.021282798833819e-05,
+      "loss": 0.0207,
+      "step": 29070
+    },
+    {
+      "epoch": 41.78,
+      "learning_rate": 5.012536443148688e-05,
+      "loss": 0.0217,
+      "step": 29080
+    },
+    {
+      "epoch": 41.8,
+      "learning_rate": 5.0037900874635566e-05,
+      "loss": 0.0185,
+      "step": 29090
+    },
+    {
+      "epoch": 41.81,
+      "learning_rate": 4.995043731778425e-05,
+      "loss": 0.0311,
+      "step": 29100
+    },
+    {
+      "epoch": 41.81,
+      "eval_loss": 0.12811152637004852,
+      "eval_runtime": 40.4122,
+      "eval_samples_per_second": 11.308,
+      "eval_steps_per_second": 1.435,
+      "eval_wer": 0.1236259599457913,
+      "step": 29100
+    },
+    {
+      "epoch": 41.82,
+      "learning_rate": 4.986297376093294e-05,
+      "loss": 0.0205,
+      "step": 29110
+    },
+    {
+      "epoch": 41.84,
+      "learning_rate": 4.977551020408162e-05,
+      "loss": 0.0294,
+      "step": 29120
+    },
+    {
+      "epoch": 41.85,
+      "learning_rate": 4.968804664723032e-05,
+      "loss": 0.0138,
+      "step": 29130
+    },
+    {
+      "epoch": 41.87,
+      "learning_rate": 4.9600583090379e-05,
+      "loss": 0.0263,
+      "step": 29140
+    },
+    {
+      "epoch": 41.88,
+      "learning_rate": 4.9513119533527694e-05,
+      "loss": 0.0214,
+      "step": 29150
+    },
+    {
+      "epoch": 41.9,
+      "learning_rate": 4.942565597667638e-05,
+      "loss": 0.0162,
+      "step": 29160
+    },
+    {
+      "epoch": 41.91,
+      "learning_rate": 4.933819241982507e-05,
+      "loss": 0.0251,
+      "step": 29170
+    },
+    {
+      "epoch": 41.93,
+      "learning_rate": 4.925072886297376e-05,
+      "loss": 0.0186,
+      "step": 29180
+    },
+    {
+      "epoch": 41.94,
+      "learning_rate": 4.9163265306122445e-05,
+      "loss": 0.0242,
+      "step": 29190
+    },
+    {
+      "epoch": 41.95,
+      "learning_rate": 4.907580174927113e-05,
+      "loss": 0.0235,
+      "step": 29200
+    },
+    {
+      "epoch": 41.95,
+      "eval_loss": 0.12631256878376007,
+      "eval_runtime": 41.101,
+      "eval_samples_per_second": 11.119,
+      "eval_steps_per_second": 1.411,
+      "eval_wer": 0.1267881343171209,
+      "step": 29200
+    },
+    {
+      "epoch": 41.97,
+      "learning_rate": 4.898833819241982e-05,
+      "loss": 0.0178,
+      "step": 29210
+    },
+    {
+      "epoch": 41.98,
+      "learning_rate": 4.890087463556851e-05,
+      "loss": 0.0259,
+      "step": 29220
+    },
+    {
+      "epoch": 42.0,
+      "learning_rate": 4.881341107871719e-05,
+      "loss": 0.019,
+      "step": 29230
+    },
+    {
+      "epoch": 42.01,
+      "learning_rate": 4.8725947521865885e-05,
+      "loss": 0.0196,
+      "step": 29240
+    },
+    {
+      "epoch": 42.03,
+      "learning_rate": 4.863848396501458e-05,
+      "loss": 0.0212,
+      "step": 29250
+    },
+    {
+      "epoch": 42.04,
+      "learning_rate": 4.855102040816326e-05,
+      "loss": 0.0214,
+      "step": 29260
+    },
+    {
+      "epoch": 42.05,
+      "learning_rate": 4.8463556851311955e-05,
+      "loss": 0.0219,
+      "step": 29270
+    },
+    {
+      "epoch": 42.07,
+      "learning_rate": 4.8376093294460636e-05,
+      "loss": 0.027,
+      "step": 29280
+    },
+    {
+      "epoch": 42.08,
+      "learning_rate": 4.8288629737609324e-05,
+      "loss": 0.0186,
+      "step": 29290
+    },
+    {
+      "epoch": 42.1,
+      "learning_rate": 4.820116618075801e-05,
+      "loss": 0.0175,
+      "step": 29300
+    },
+    {
+      "epoch": 42.1,
+      "eval_loss": 0.12914027273654938,
+      "eval_runtime": 40.8425,
+      "eval_samples_per_second": 11.189,
+      "eval_steps_per_second": 1.42,
+      "eval_wer": 0.1255834964613763,
+      "step": 29300
+    },
+    {
+      "epoch": 42.11,
+      "learning_rate": 4.81137026239067e-05,
+      "loss": 0.0227,
+      "step": 29310
+    },
+    {
+      "epoch": 42.13,
+      "learning_rate": 4.802623906705539e-05,
+      "loss": 0.0246,
+      "step": 29320
+    },
+    {
+      "epoch": 42.14,
+      "learning_rate": 4.7938775510204076e-05,
+      "loss": 0.0188,
+      "step": 29330
+    },
+    {
+      "epoch": 42.16,
+      "learning_rate": 4.785131195335277e-05,
+      "loss": 0.0217,
+      "step": 29340
+    },
+    {
+      "epoch": 42.17,
+      "learning_rate": 4.776384839650145e-05,
+      "loss": 0.0219,
+      "step": 29350
+    },
+    {
+      "epoch": 42.18,
+      "learning_rate": 4.7676384839650146e-05,
+      "loss": 0.019,
+      "step": 29360
+    },
+    {
+      "epoch": 42.2,
+      "learning_rate": 4.758892128279883e-05,
+      "loss": 0.0241,
+      "step": 29370
+    },
+    {
+      "epoch": 42.21,
+      "learning_rate": 4.750145772594752e-05,
+      "loss": 0.016,
+      "step": 29380
+    },
+    {
+      "epoch": 42.23,
+      "learning_rate": 4.74139941690962e-05,
+      "loss": 0.0277,
+      "step": 29390
+    },
+    {
+      "epoch": 42.24,
+      "learning_rate": 4.73265306122449e-05,
+      "loss": 0.0188,
+      "step": 29400
+    },
+    {
+      "epoch": 42.24,
+      "eval_loss": 0.12900404632091522,
+      "eval_runtime": 40.6753,
+      "eval_samples_per_second": 11.235,
+      "eval_steps_per_second": 1.426,
+      "eval_wer": 0.12573407619334437,
+      "step": 29400
+    },
+    {
+      "epoch": 42.26,
+      "learning_rate": 4.723906705539358e-05,
+      "loss": 0.0176,
+      "step": 29410
+    },
+    {
+      "epoch": 42.27,
+      "learning_rate": 4.715160349854227e-05,
+      "loss": 0.021,
+      "step": 29420
+    },
+    {
+      "epoch": 42.28,
+      "learning_rate": 4.706413994169096e-05,
+      "loss": 0.0168,
+      "step": 29430
+    },
+    {
+      "epoch": 42.3,
+      "learning_rate": 4.697667638483964e-05,
+      "loss": 0.0258,
+      "step": 29440
+    },
+    {
+      "epoch": 42.31,
+      "learning_rate": 4.688921282798834e-05,
+      "loss": 0.0234,
+      "step": 29450
+    },
+    {
+      "epoch": 42.33,
+      "learning_rate": 4.680174927113702e-05,
+      "loss": 0.0248,
+      "step": 29460
+    },
+    {
+      "epoch": 42.34,
+      "learning_rate": 4.671428571428571e-05,
+      "loss": 0.0213,
+      "step": 29470
+    },
+    {
+      "epoch": 42.36,
+      "learning_rate": 4.6626822157434394e-05,
+      "loss": 0.0246,
+      "step": 29480
+    },
+    {
+      "epoch": 42.37,
+      "learning_rate": 4.653935860058309e-05,
+      "loss": 0.0244,
+      "step": 29490
+    },
+    {
+      "epoch": 42.39,
+      "learning_rate": 4.645189504373177e-05,
+      "loss": 0.0288,
+      "step": 29500
+    },
+    {
+      "epoch": 42.39,
+      "eval_loss": 0.12424959987401962,
+      "eval_runtime": 40.6911,
+      "eval_samples_per_second": 11.231,
+      "eval_steps_per_second": 1.425,
+      "eval_wer": 0.1261858153892486,
+      "step": 29500
+    },
+    {
+      "epoch": 42.4,
+      "learning_rate": 4.6364431486880465e-05,
+      "loss": 0.0208,
+      "step": 29510
+    },
+    {
+      "epoch": 42.41,
+      "learning_rate": 4.627696793002915e-05,
+      "loss": 0.0203,
+      "step": 29520
+    },
+    {
+      "epoch": 42.43,
+      "learning_rate": 4.618950437317784e-05,
+      "loss": 0.0182,
+      "step": 29530
+    },
+    {
+      "epoch": 42.44,
+      "learning_rate": 4.610204081632653e-05,
+      "loss": 0.0262,
+      "step": 29540
+    },
+    {
+      "epoch": 42.46,
+      "learning_rate": 4.6014577259475216e-05,
+      "loss": 0.0173,
+      "step": 29550
+    },
+    {
+      "epoch": 42.47,
+      "learning_rate": 4.5927113702623904e-05,
+      "loss": 0.0202,
+      "step": 29560
+    },
+    {
+      "epoch": 42.49,
+      "learning_rate": 4.5839650145772585e-05,
+      "loss": 0.0233,
+      "step": 29570
+    },
+    {
+      "epoch": 42.5,
+      "learning_rate": 4.575218658892128e-05,
+      "loss": 0.0164,
+      "step": 29580
+    },
+    {
+      "epoch": 42.51,
+      "learning_rate": 4.566472303206996e-05,
+      "loss": 0.0245,
+      "step": 29590
+    },
+    {
+      "epoch": 42.53,
+      "learning_rate": 4.5577259475218656e-05,
+      "loss": 0.0223,
+      "step": 29600
+    },
+    {
+      "epoch": 42.53,
+      "eval_loss": 0.12688687443733215,
+      "eval_runtime": 40.9929,
+      "eval_samples_per_second": 11.148,
+      "eval_steps_per_second": 1.415,
+      "eval_wer": 0.12663755458515283,
+      "step": 29600
+    },
+    {
+      "epoch": 42.54,
+      "learning_rate": 4.548979591836734e-05,
+      "loss": 0.0254,
+      "step": 29610
+    },
+    {
+      "epoch": 42.56,
+      "learning_rate": 4.540233236151603e-05,
+      "loss": 0.0263,
+      "step": 29620
+    },
+    {
+      "epoch": 42.57,
+      "learning_rate": 4.531486880466472e-05,
+      "loss": 0.024,
+      "step": 29630
+    },
+    {
+      "epoch": 42.59,
+      "learning_rate": 4.522740524781341e-05,
+      "loss": 0.0236,
+      "step": 29640
+    },
+    {
+      "epoch": 42.6,
+      "learning_rate": 4.5139941690962095e-05,
+      "loss": 0.0223,
+      "step": 29650
+    },
+    {
+      "epoch": 42.61,
+      "learning_rate": 4.505247813411078e-05,
+      "loss": 0.0233,
+      "step": 29660
+    },
+    {
+      "epoch": 42.63,
+      "learning_rate": 4.496501457725947e-05,
+      "loss": 0.027,
+      "step": 29670
+    },
+    {
+      "epoch": 42.64,
+      "learning_rate": 4.487755102040816e-05,
+      "loss": 0.0146,
+      "step": 29680
+    },
+    {
+      "epoch": 42.66,
+      "learning_rate": 4.479008746355685e-05,
+      "loss": 0.0316,
+      "step": 29690
+    },
+    {
+      "epoch": 42.67,
+      "learning_rate": 4.470262390670553e-05,
+      "loss": 0.0125,
+      "step": 29700
+    },
+    {
+      "epoch": 42.67,
+      "eval_loss": 0.13046014308929443,
+      "eval_runtime": 40.7681,
+      "eval_samples_per_second": 11.21,
+      "eval_steps_per_second": 1.423,
+      "eval_wer": 0.12302364101791899,
+      "step": 29700
+    },
+    {
+      "epoch": 42.69,
+      "learning_rate": 4.461516034985422e-05,
+      "loss": 0.0232,
+      "step": 29710
+    },
+    {
+      "epoch": 42.7,
+      "learning_rate": 4.452769679300292e-05,
+      "loss": 0.0207,
+      "step": 29720
+    },
+    {
+      "epoch": 42.72,
+      "learning_rate": 4.44402332361516e-05,
+      "loss": 0.0174,
+      "step": 29730
+    },
+    {
+      "epoch": 42.73,
+      "learning_rate": 4.435276967930029e-05,
+      "loss": 0.0201,
+      "step": 29740
+    },
+    {
+      "epoch": 42.74,
+      "learning_rate": 4.4265306122448974e-05,
+      "loss": 0.022,
+      "step": 29750
+    },
+    {
+      "epoch": 42.76,
+      "learning_rate": 4.417784256559766e-05,
+      "loss": 0.0154,
+      "step": 29760
+    },
+    {
+      "epoch": 42.77,
+      "learning_rate": 4.409037900874635e-05,
+      "loss": 0.0194,
+      "step": 29770
+    },
+    {
+      "epoch": 42.79,
+      "learning_rate": 4.400291545189504e-05,
+      "loss": 0.0198,
+      "step": 29780
+    },
+    {
+      "epoch": 42.8,
+      "learning_rate": 4.3915451895043726e-05,
+      "loss": 0.0199,
+      "step": 29790
+    },
+    {
+      "epoch": 42.82,
+      "learning_rate": 4.3827988338192414e-05,
+      "loss": 0.0182,
+      "step": 29800
+    },
+    {
+      "epoch": 42.82,
+      "eval_loss": 0.131119966506958,
+      "eval_runtime": 41.0012,
+      "eval_samples_per_second": 11.146,
+      "eval_steps_per_second": 1.415,
+      "eval_wer": 0.1222707423580786,
+      "step": 29800
+    },
+    {
+      "epoch": 42.83,
+      "learning_rate": 4.374052478134111e-05,
+      "loss": 0.0175,
+      "step": 29810
+    },
+    {
+      "epoch": 42.84,
+      "learning_rate": 4.365306122448979e-05,
+      "loss": 0.0252,
+      "step": 29820
+    },
+    {
+      "epoch": 42.86,
+      "learning_rate": 4.3565597667638484e-05,
+      "loss": 0.0176,
+      "step": 29830
+    },
+    {
+      "epoch": 42.87,
+      "learning_rate": 4.3478134110787165e-05,
+      "loss": 0.0397,
+      "step": 29840
+    },
+    {
+      "epoch": 42.89,
+      "learning_rate": 4.339067055393586e-05,
+      "loss": 0.0147,
+      "step": 29850
+    },
+    {
+      "epoch": 42.9,
+      "learning_rate": 4.330320699708454e-05,
+      "loss": 0.0154,
+      "step": 29860
+    },
+    {
+      "epoch": 42.92,
+      "learning_rate": 4.3215743440233236e-05,
+      "loss": 0.0181,
+      "step": 29870
+    },
+    {
+      "epoch": 42.93,
+      "learning_rate": 4.312827988338192e-05,
+      "loss": 0.0121,
+      "step": 29880
+    },
+    {
+      "epoch": 42.95,
+      "learning_rate": 4.3040816326530605e-05,
+      "loss": 0.0298,
+      "step": 29890
+    },
+    {
+      "epoch": 42.96,
+      "learning_rate": 4.29533527696793e-05,
+      "loss": 0.0194,
+      "step": 29900
+    },
+    {
+      "epoch": 42.96,
+      "eval_loss": 0.131606325507164,
+      "eval_runtime": 40.5304,
+      "eval_samples_per_second": 11.275,
+      "eval_steps_per_second": 1.431,
+      "eval_wer": 0.12708929378105707,
+      "step": 29900
+    },
+    {
+      "epoch": 42.97,
+      "learning_rate": 4.286588921282798e-05,
+      "loss": 0.019,
+      "step": 29910
+    },
+    {
+      "epoch": 42.99,
+      "learning_rate": 4.2778425655976675e-05,
+      "loss": 0.0251,
+      "step": 29920
+    },
+    {
+      "epoch": 43.0,
+      "learning_rate": 4.2690962099125357e-05,
+      "loss": 0.0166,
+      "step": 29930
+    },
+    {
+      "epoch": 43.02,
+      "learning_rate": 4.260349854227405e-05,
+      "loss": 0.0201,
+      "step": 29940
+    },
+    {
+      "epoch": 43.03,
+      "learning_rate": 4.251603498542273e-05,
+      "loss": 0.0187,
+      "step": 29950
+    },
+    {
+      "epoch": 43.05,
+      "learning_rate": 4.242857142857143e-05,
+      "loss": 0.0256,
+      "step": 29960
+    },
+    {
+      "epoch": 43.06,
+      "learning_rate": 4.234110787172011e-05,
+      "loss": 0.0249,
+      "step": 29970
+    },
+    {
+      "epoch": 43.07,
+      "learning_rate": 4.22536443148688e-05,
+      "loss": 0.0209,
+      "step": 29980
+    },
+    {
+      "epoch": 43.09,
+      "learning_rate": 4.216618075801749e-05,
+      "loss": 0.0222,
+      "step": 29990
+    },
+    {
+      "epoch": 43.1,
+      "learning_rate": 4.208746355685131e-05,
+      "loss": 0.0148,
+      "step": 30000
+    },
+    {
+      "epoch": 43.1,
+      "eval_loss": 0.1303335726261139,
+      "eval_runtime": 40.5862,
+      "eval_samples_per_second": 11.26,
+      "eval_steps_per_second": 1.429,
+      "eval_wer": 0.12708929378105707,
+      "step": 30000
+    },
+    {
+      "epoch": 43.12,
+      "learning_rate": 4.2e-05,
+      "loss": 0.0199,
+      "step": 30010
+    },
+    {
+      "epoch": 43.13,
+      "learning_rate": 4.1912536443148686e-05,
+      "loss": 0.0153,
+      "step": 30020
+    },
+    {
+      "epoch": 43.15,
+      "learning_rate": 4.1825072886297373e-05,
+      "loss": 0.0187,
+      "step": 30030
+    },
+    {
+      "epoch": 43.16,
+      "learning_rate": 4.173760932944606e-05,
+      "loss": 0.0203,
+      "step": 30040
+    },
+    {
+      "epoch": 43.18,
+      "learning_rate": 4.165014577259475e-05,
+      "loss": 0.0207,
+      "step": 30050
+    },
+    {
+      "epoch": 43.19,
+      "learning_rate": 4.156268221574344e-05,
+      "loss": 0.0232,
+      "step": 30060
+    },
+    {
+      "epoch": 43.2,
+      "learning_rate": 4.1475218658892125e-05,
+      "loss": 0.0217,
+      "step": 30070
+    },
+    {
+      "epoch": 43.22,
+      "learning_rate": 4.1387755102040806e-05,
+      "loss": 0.0182,
+      "step": 30080
+    },
+    {
+      "epoch": 43.23,
+      "learning_rate": 4.13002915451895e-05,
+      "loss": 0.0257,
+      "step": 30090
+    },
+    {
+      "epoch": 43.25,
+      "learning_rate": 4.1212827988338196e-05,
+      "loss": 0.0125,
+      "step": 30100
+    },
+    {
+      "epoch": 43.25,
+      "eval_loss": 0.12872837483882904,
+      "eval_runtime": 40.7552,
+      "eval_samples_per_second": 11.213,
+      "eval_steps_per_second": 1.423,
+      "eval_wer": 0.12468001806956784,
+      "step": 30100
+    },
+    {
+      "epoch": 43.26,
+      "learning_rate": 4.112536443148688e-05,
+      "loss": 0.0231,
+      "step": 30110
+    },
+    {
+      "epoch": 43.28,
+      "learning_rate": 4.103790087463557e-05,
+      "loss": 0.0132,
+      "step": 30120
+    },
+    {
+      "epoch": 43.29,
+      "learning_rate": 4.095043731778425e-05,
+      "loss": 0.0208,
+      "step": 30130
+    },
+    {
+      "epoch": 43.3,
+      "learning_rate": 4.086297376093294e-05,
+      "loss": 0.0193,
+      "step": 30140
+    },
+    {
+      "epoch": 43.32,
+      "learning_rate": 4.077551020408163e-05,
+      "loss": 0.0142,
+      "step": 30150
+    },
+    {
+      "epoch": 43.33,
+      "learning_rate": 4.0688046647230316e-05,
+      "loss": 0.0343,
+      "step": 30160
+    },
+    {
+      "epoch": 43.35,
+      "learning_rate": 4.0600583090379004e-05,
+      "loss": 0.0209,
+      "step": 30170
+    },
+    {
+      "epoch": 43.36,
+      "learning_rate": 4.051311953352769e-05,
+      "loss": 0.0164,
+      "step": 30180
+    },
+    {
+      "epoch": 43.38,
+      "learning_rate": 4.0425655976676387e-05,
+      "loss": 0.026,
+      "step": 30190
+    },
+    {
+      "epoch": 43.39,
+      "learning_rate": 4.033819241982507e-05,
+      "loss": 0.0197,
+      "step": 30200
+    },
+    {
+      "epoch": 43.39,
+      "eval_loss": 0.12666457891464233,
+      "eval_runtime": 40.7476,
+      "eval_samples_per_second": 11.215,
+      "eval_steps_per_second": 1.423,
+      "eval_wer": 0.12287306128595091,
+      "step": 30200
+    },
+    {
+      "epoch": 43.41,
+      "learning_rate": 4.025072886297376e-05,
+      "loss": 0.0222,
+      "step": 30210
+    },
+    {
+      "epoch": 43.42,
+      "learning_rate": 4.0163265306122443e-05,
+      "loss": 0.0146,
+      "step": 30220
+    },
+    {
+      "epoch": 43.43,
+      "learning_rate": 4.007580174927114e-05,
+      "loss": 0.0144,
+      "step": 30230
+    },
+    {
+      "epoch": 43.45,
+      "learning_rate": 3.998833819241982e-05,
+      "loss": 0.0253,
+      "step": 30240
+    },
+    {
+      "epoch": 43.46,
+      "learning_rate": 3.9900874635568514e-05,
+      "loss": 0.0207,
+      "step": 30250
+    },
+    {
+      "epoch": 43.48,
+      "learning_rate": 3.9813411078717195e-05,
+      "loss": 0.0235,
+      "step": 30260
+    },
+    {
+      "epoch": 43.49,
+      "learning_rate": 3.972594752186588e-05,
+      "loss": 0.0147,
+      "step": 30270
+    },
+    {
+      "epoch": 43.51,
+      "learning_rate": 3.963848396501457e-05,
+      "loss": 0.0172,
+      "step": 30280
+    },
+    {
+      "epoch": 43.52,
+      "learning_rate": 3.955102040816326e-05,
+      "loss": 0.0248,
+      "step": 30290
+    },
+    {
+      "epoch": 43.53,
+      "learning_rate": 3.9463556851311953e-05,
+      "loss": 0.0149,
+      "step": 30300
+    },
+    {
+      "epoch": 43.53,
+      "eval_loss": 0.12862493097782135,
+      "eval_runtime": 40.7412,
+      "eval_samples_per_second": 11.217,
+      "eval_steps_per_second": 1.424,
+      "eval_wer": 0.12212016262611053,
+      "step": 30300
+    },
+    {
+      "epoch": 43.55,
+      "learning_rate": 3.9376093294460635e-05,
+      "loss": 0.0221,
+      "step": 30310
+    },
+    {
+      "epoch": 43.56,
+      "learning_rate": 3.928862973760933e-05,
+      "loss": 0.0152,
+      "step": 30320
+    },
+    {
+      "epoch": 43.58,
+      "learning_rate": 3.920116618075801e-05,
+      "loss": 0.0186,
+      "step": 30330
+    },
+    {
+      "epoch": 43.59,
+      "learning_rate": 3.9113702623906705e-05,
+      "loss": 0.0161,
+      "step": 30340
+    },
+    {
+      "epoch": 43.61,
+      "learning_rate": 3.9026239067055386e-05,
+      "loss": 0.0157,
+      "step": 30350
+    },
+    {
+      "epoch": 43.62,
+      "learning_rate": 3.893877551020408e-05,
+      "loss": 0.0231,
+      "step": 30360
+    },
+    {
+      "epoch": 43.64,
+      "learning_rate": 3.885131195335276e-05,
+      "loss": 0.0174,
+      "step": 30370
+    },
+    {
+      "epoch": 43.65,
+      "learning_rate": 3.876384839650146e-05,
+      "loss": 0.0181,
+      "step": 30380
+    },
+    {
+      "epoch": 43.66,
+      "learning_rate": 3.8676384839650145e-05,
+      "loss": 0.021,
+      "step": 30390
+    },
+    {
+      "epoch": 43.68,
+      "learning_rate": 3.858892128279883e-05,
+      "loss": 0.0268,
+      "step": 30400
+    },
+    {
+      "epoch": 43.68,
+      "eval_loss": 0.1292026787996292,
+      "eval_runtime": 40.6933,
+      "eval_samples_per_second": 11.23,
+      "eval_steps_per_second": 1.425,
+      "eval_wer": 0.12272248155398284,
+      "step": 30400
+    },
+    {
+      "epoch": 43.69,
+      "learning_rate": 3.850145772594752e-05,
+      "loss": 0.0213,
+      "step": 30410
+    },
+    {
+      "epoch": 43.71,
+      "learning_rate": 3.84139941690962e-05,
+      "loss": 0.0202,
+      "step": 30420
+    },
+    {
+      "epoch": 43.72,
+      "learning_rate": 3.8326530612244896e-05,
+      "loss": 0.0204,
+      "step": 30430
+    },
+    {
+      "epoch": 43.74,
+      "learning_rate": 3.823906705539358e-05,
+      "loss": 0.0215,
+      "step": 30440
+    },
+    {
+      "epoch": 43.75,
+      "learning_rate": 3.815160349854227e-05,
+      "loss": 0.0113,
+      "step": 30450
+    },
+    {
+      "epoch": 43.76,
+      "learning_rate": 3.806413994169095e-05,
+      "loss": 0.0184,
+      "step": 30460
+    },
+    {
+      "epoch": 43.78,
+      "learning_rate": 3.797667638483965e-05,
+      "loss": 0.0226,
+      "step": 30470
+    },
+    {
+      "epoch": 43.79,
+      "learning_rate": 3.7889212827988336e-05,
+      "loss": 0.0262,
+      "step": 30480
+    },
+    {
+      "epoch": 43.81,
+      "learning_rate": 3.7801749271137023e-05,
+      "loss": 0.018,
+      "step": 30490
+    },
+    {
+      "epoch": 43.82,
+      "learning_rate": 3.771428571428571e-05,
+      "loss": 0.0173,
+      "step": 30500
+    },
+    {
+      "epoch": 43.82,
+      "eval_loss": 0.1283373236656189,
+      "eval_runtime": 40.7197,
+      "eval_samples_per_second": 11.223,
+      "eval_steps_per_second": 1.424,
+      "eval_wer": 0.1216684234302063,
+      "step": 30500
+    },
+    {
+      "epoch": 43.84,
+      "learning_rate": 3.76268221574344e-05,
+      "loss": 0.0201,
+      "step": 30510
+    },
+    {
+      "epoch": 43.85,
+      "learning_rate": 3.753935860058309e-05,
+      "loss": 0.0161,
+      "step": 30520
+    },
+    {
+      "epoch": 43.86,
+      "learning_rate": 3.7451895043731775e-05,
+      "loss": 0.0172,
+      "step": 30530
+    },
+    {
+      "epoch": 43.88,
+      "learning_rate": 3.736443148688046e-05,
+      "loss": 0.0206,
+      "step": 30540
+    },
+    {
+      "epoch": 43.89,
+      "learning_rate": 3.727696793002915e-05,
+      "loss": 0.0183,
+      "step": 30550
+    },
+    {
+      "epoch": 43.91,
+      "learning_rate": 3.718950437317784e-05,
+      "loss": 0.0198,
+      "step": 30560
+    },
+    {
+      "epoch": 43.92,
+      "learning_rate": 3.710204081632653e-05,
+      "loss": 0.0176,
+      "step": 30570
+    },
+    {
+      "epoch": 43.94,
+      "learning_rate": 3.7014577259475215e-05,
+      "loss": 0.0174,
+      "step": 30580
+    },
+    {
+      "epoch": 43.95,
+      "learning_rate": 3.69271137026239e-05,
+      "loss": 0.0219,
+      "step": 30590
+    },
+    {
+      "epoch": 43.97,
+      "learning_rate": 3.683965014577259e-05,
+      "loss": 0.0171,
+      "step": 30600
+    },
+    {
+      "epoch": 43.97,
+      "eval_loss": 0.12676407396793365,
+      "eval_runtime": 40.7159,
+      "eval_samples_per_second": 11.224,
+      "eval_steps_per_second": 1.425,
+      "eval_wer": 0.11986146664658937,
+      "step": 30600
+    },
+    {
+      "epoch": 43.98,
+      "learning_rate": 3.675218658892128e-05,
+      "loss": 0.0201,
+      "step": 30610
+    },
+    {
+      "epoch": 43.99,
+      "learning_rate": 3.6664723032069966e-05,
+      "loss": 0.0183,
+      "step": 30620
+    },
+    {
+      "epoch": 44.01,
+      "learning_rate": 3.6577259475218654e-05,
+      "loss": 0.0218,
+      "step": 30630
+    },
+    {
+      "epoch": 44.02,
+      "learning_rate": 3.648979591836734e-05,
+      "loss": 0.0207,
+      "step": 30640
+    },
+    {
+      "epoch": 44.04,
+      "learning_rate": 3.640233236151603e-05,
+      "loss": 0.0164,
+      "step": 30650
+    },
+    {
+      "epoch": 44.05,
+      "learning_rate": 3.631486880466472e-05,
+      "loss": 0.0188,
+      "step": 30660
+    },
+    {
+      "epoch": 44.07,
+      "learning_rate": 3.6227405247813406e-05,
+      "loss": 0.022,
+      "step": 30670
+    },
+    {
+      "epoch": 44.08,
+      "learning_rate": 3.6139941690962094e-05,
+      "loss": 0.0192,
+      "step": 30680
+    },
+    {
+      "epoch": 44.09,
+      "learning_rate": 3.605247813411078e-05,
+      "loss": 0.0197,
+      "step": 30690
+    },
+    {
+      "epoch": 44.11,
+      "learning_rate": 3.5965014577259476e-05,
+      "loss": 0.0184,
+      "step": 30700
+    },
+    {
+      "epoch": 44.11,
+      "eval_loss": 0.12772968411445618,
+      "eval_runtime": 40.7769,
+      "eval_samples_per_second": 11.207,
+      "eval_steps_per_second": 1.422,
+      "eval_wer": 0.12121668423430207,
+      "step": 30700
+    },
+    {
+      "epoch": 44.12,
+      "learning_rate": 3.5877551020408164e-05,
+      "loss": 0.0306,
+      "step": 30710
+    },
+    {
+      "epoch": 44.14,
+      "learning_rate": 3.579008746355685e-05,
+      "loss": 0.0197,
+      "step": 30720
+    },
+    {
+      "epoch": 44.15,
+      "learning_rate": 3.570262390670554e-05,
+      "loss": 0.0297,
+      "step": 30730
+    },
+    {
+      "epoch": 44.17,
+      "learning_rate": 3.561516034985423e-05,
+      "loss": 0.0166,
+      "step": 30740
+    },
+    {
+      "epoch": 44.18,
+      "learning_rate": 3.552769679300291e-05,
+      "loss": 0.0183,
+      "step": 30750
+    },
+    {
+      "epoch": 44.2,
+      "learning_rate": 3.54402332361516e-05,
+      "loss": 0.0208,
+      "step": 30760
+    },
+    {
+      "epoch": 44.21,
+      "learning_rate": 3.5352769679300285e-05,
+      "loss": 0.0184,
+      "step": 30770
+    },
+    {
+      "epoch": 44.22,
+      "learning_rate": 3.526530612244897e-05,
+      "loss": 0.0181,
+      "step": 30780
+    },
+    {
+      "epoch": 44.24,
+      "learning_rate": 3.517784256559767e-05,
+      "loss": 0.017,
+      "step": 30790
+    },
+    {
+      "epoch": 44.25,
+      "learning_rate": 3.5090379008746355e-05,
+      "loss": 0.0176,
+      "step": 30800
+    },
+    {
+      "epoch": 44.25,
+      "eval_loss": 0.1296500861644745,
+      "eval_runtime": 40.688,
+      "eval_samples_per_second": 11.232,
+      "eval_steps_per_second": 1.425,
+      "eval_wer": 0.1203132058424936,
+      "step": 30800
+    },
+    {
+      "epoch": 44.27,
+      "learning_rate": 3.500291545189504e-05,
+      "loss": 0.0158,
+      "step": 30810
+    },
+    {
+      "epoch": 44.28,
+      "learning_rate": 3.491545189504373e-05,
+      "loss": 0.0138,
+      "step": 30820
+    },
+    {
+      "epoch": 44.3,
+      "learning_rate": 3.482798833819242e-05,
+      "loss": 0.0197,
+      "step": 30830
+    },
+    {
+      "epoch": 44.31,
+      "learning_rate": 3.474052478134111e-05,
+      "loss": 0.0173,
+      "step": 30840
+    },
+    {
+      "epoch": 44.32,
+      "learning_rate": 3.4653061224489795e-05,
+      "loss": 0.0138,
+      "step": 30850
+    },
+    {
+      "epoch": 44.34,
+      "learning_rate": 3.456559766763848e-05,
+      "loss": 0.0177,
+      "step": 30860
+    },
+    {
+      "epoch": 44.35,
+      "learning_rate": 3.447813411078717e-05,
+      "loss": 0.0153,
+      "step": 30870
+    },
+    {
+      "epoch": 44.37,
+      "learning_rate": 3.439067055393586e-05,
+      "loss": 0.0169,
+      "step": 30880
+    },
+    {
+      "epoch": 44.38,
+      "learning_rate": 3.4303206997084546e-05,
+      "loss": 0.0168,
+      "step": 30890
+    },
+    {
+      "epoch": 44.4,
+      "learning_rate": 3.4215743440233234e-05,
+      "loss": 0.0172,
+      "step": 30900
+    },
+    {
+      "epoch": 44.4,
+      "eval_loss": 0.12854306399822235,
+      "eval_runtime": 41.8837,
+      "eval_samples_per_second": 10.911,
+      "eval_steps_per_second": 1.385,
+      "eval_wer": 0.12106610450233399,
+      "step": 30900
+    },
+    {
+      "epoch": 44.41,
+      "learning_rate": 3.412827988338192e-05,
+      "loss": 0.0215,
+      "step": 30910
+    },
+    {
+      "epoch": 44.43,
+      "learning_rate": 3.404081632653061e-05,
+      "loss": 0.0137,
+      "step": 30920
+    },
+    {
+      "epoch": 44.44,
+      "learning_rate": 3.39533527696793e-05,
+      "loss": 0.0201,
+      "step": 30930
+    },
+    {
+      "epoch": 44.45,
+      "learning_rate": 3.3865889212827986e-05,
+      "loss": 0.0171,
+      "step": 30940
+    },
+    {
+      "epoch": 44.47,
+      "learning_rate": 3.3778425655976673e-05,
+      "loss": 0.0164,
+      "step": 30950
+    },
+    {
+      "epoch": 44.48,
+      "learning_rate": 3.369096209912536e-05,
+      "loss": 0.0221,
+      "step": 30960
+    },
+    {
+      "epoch": 44.5,
+      "learning_rate": 3.360349854227405e-05,
+      "loss": 0.013,
+      "step": 30970
+    },
+    {
+      "epoch": 44.51,
+      "learning_rate": 3.351603498542274e-05,
+      "loss": 0.0203,
+      "step": 30980
+    },
+    {
+      "epoch": 44.53,
+      "learning_rate": 3.3428571428571425e-05,
+      "loss": 0.0196,
+      "step": 30990
+    },
+    {
+      "epoch": 44.54,
+      "learning_rate": 3.334110787172011e-05,
+      "loss": 0.0155,
+      "step": 31000
+    },
+    {
+      "epoch": 44.54,
+      "eval_loss": 0.12705247104167938,
+      "eval_runtime": 40.7901,
+      "eval_samples_per_second": 11.204,
+      "eval_steps_per_second": 1.422,
+      "eval_wer": 0.12212016262611053,
+      "step": 31000
+    },
+    {
+      "epoch": 44.55,
+      "learning_rate": 3.32536443148688e-05,
+      "loss": 0.0255,
+      "step": 31010
+    },
+    {
+      "epoch": 44.57,
+      "learning_rate": 3.316618075801749e-05,
+      "loss": 0.0139,
+      "step": 31020
+    },
+    {
+      "epoch": 44.58,
+      "learning_rate": 3.307871720116618e-05,
+      "loss": 0.0225,
+      "step": 31030
+    },
+    {
+      "epoch": 44.6,
+      "learning_rate": 3.2991253644314865e-05,
+      "loss": 0.0192,
+      "step": 31040
+    },
+    {
+      "epoch": 44.61,
+      "learning_rate": 3.290379008746355e-05,
+      "loss": 0.0167,
+      "step": 31050
+    },
+    {
+      "epoch": 44.63,
+      "learning_rate": 3.281632653061224e-05,
+      "loss": 0.0159,
+      "step": 31060
+    },
+    {
+      "epoch": 44.64,
+      "learning_rate": 3.2728862973760935e-05,
+      "loss": 0.017,
+      "step": 31070
+    },
+    {
+      "epoch": 44.66,
+      "learning_rate": 3.2641399416909616e-05,
+      "loss": 0.0287,
+      "step": 31080
+    },
+    {
+      "epoch": 44.67,
+      "learning_rate": 3.2553935860058304e-05,
+      "loss": 0.0176,
+      "step": 31090
+    },
+    {
+      "epoch": 44.68,
+      "learning_rate": 3.246647230320699e-05,
+      "loss": 0.0195,
+      "step": 31100
+    },
+    {
+      "epoch": 44.68,
+      "eval_loss": 0.12698258459568024,
+      "eval_runtime": 40.4576,
+      "eval_samples_per_second": 11.296,
+      "eval_steps_per_second": 1.434,
+      "eval_wer": 0.12076494503839783,
+      "step": 31100
+    },
+    {
+      "epoch": 44.7,
+      "learning_rate": 3.237900874635568e-05,
+      "loss": 0.0197,
+      "step": 31110
+    },
+    {
+      "epoch": 44.71,
+      "learning_rate": 3.229154518950437e-05,
+      "loss": 0.0102,
+      "step": 31120
+    },
+    {
+      "epoch": 44.73,
+      "learning_rate": 3.2204081632653056e-05,
+      "loss": 0.0215,
+      "step": 31130
+    },
+    {
+      "epoch": 44.74,
+      "learning_rate": 3.2116618075801744e-05,
+      "loss": 0.0217,
+      "step": 31140
+    },
+    {
+      "epoch": 44.76,
+      "learning_rate": 3.202915451895043e-05,
+      "loss": 0.0167,
+      "step": 31150
+    },
+    {
+      "epoch": 44.77,
+      "learning_rate": 3.1941690962099126e-05,
+      "loss": 0.0266,
+      "step": 31160
+    },
+    {
+      "epoch": 44.78,
+      "learning_rate": 3.1854227405247814e-05,
+      "loss": 0.0148,
+      "step": 31170
+    },
+    {
+      "epoch": 44.8,
+      "learning_rate": 3.17667638483965e-05,
+      "loss": 0.0191,
+      "step": 31180
+    },
+    {
+      "epoch": 44.81,
+      "learning_rate": 3.167930029154519e-05,
+      "loss": 0.019,
+      "step": 31190
+    },
+    {
+      "epoch": 44.83,
+      "learning_rate": 3.159183673469388e-05,
+      "loss": 0.0179,
+      "step": 31200
+    },
+    {
+      "epoch": 44.83,
+      "eval_loss": 0.12674599885940552,
+      "eval_runtime": 40.9642,
+      "eval_samples_per_second": 11.156,
+      "eval_steps_per_second": 1.416,
+      "eval_wer": 0.11925914771871705,
+      "step": 31200
+    },
+    {
+      "epoch": 44.84,
+      "learning_rate": 3.1504373177842566e-05,
+      "loss": 0.0213,
+      "step": 31210
+    },
+    {
+      "epoch": 44.86,
+      "learning_rate": 3.141690962099125e-05,
+      "loss": 0.0187,
+      "step": 31220
+    },
+    {
+      "epoch": 44.87,
+      "learning_rate": 3.1329446064139935e-05,
+      "loss": 0.0268,
+      "step": 31230
+    },
+    {
+      "epoch": 44.89,
+      "learning_rate": 3.124198250728862e-05,
+      "loss": 0.0225,
+      "step": 31240
+    },
+    {
+      "epoch": 44.9,
+      "learning_rate": 3.115451895043732e-05,
+      "loss": 0.0183,
+      "step": 31250
+    },
+    {
+      "epoch": 44.91,
+      "learning_rate": 3.1067055393586005e-05,
+      "loss": 0.0191,
+      "step": 31260
+    },
+    {
+      "epoch": 44.93,
+      "learning_rate": 3.097959183673469e-05,
+      "loss": 0.0206,
+      "step": 31270
+    },
+    {
+      "epoch": 44.94,
+      "learning_rate": 3.089212827988338e-05,
+      "loss": 0.0163,
+      "step": 31280
+    },
+    {
+      "epoch": 44.96,
+      "learning_rate": 3.080466472303207e-05,
+      "loss": 0.0181,
+      "step": 31290
+    },
+    {
+      "epoch": 44.97,
+      "learning_rate": 3.071720116618076e-05,
+      "loss": 0.0171,
+      "step": 31300
+    },
+    {
+      "epoch": 44.97,
+      "eval_loss": 0.1255967915058136,
+      "eval_runtime": 40.8446,
+      "eval_samples_per_second": 11.189,
+      "eval_steps_per_second": 1.42,
+      "eval_wer": 0.11850624905887668,
+      "step": 31300
+    },
+    {
+      "epoch": 44.99,
+      "learning_rate": 3.0629737609329445e-05,
+      "loss": 0.0236,
+      "step": 31310
+    },
+    {
+      "epoch": 45.0,
+      "learning_rate": 3.054227405247813e-05,
+      "loss": 0.0213,
+      "step": 31320
+    },
+    {
+      "epoch": 45.01,
+      "learning_rate": 3.0454810495626817e-05,
+      "loss": 0.0203,
+      "step": 31330
+    },
+    {
+      "epoch": 45.03,
+      "learning_rate": 3.0367346938775508e-05,
+      "loss": 0.0158,
+      "step": 31340
+    },
+    {
+      "epoch": 45.04,
+      "learning_rate": 3.0279883381924196e-05,
+      "loss": 0.028,
+      "step": 31350
+    },
+    {
+      "epoch": 45.06,
+      "learning_rate": 3.0192419825072884e-05,
+      "loss": 0.0196,
+      "step": 31360
+    },
+    {
+      "epoch": 45.07,
+      "learning_rate": 3.0104956268221572e-05,
+      "loss": 0.0144,
+      "step": 31370
+    },
+    {
+      "epoch": 45.09,
+      "learning_rate": 3.001749271137026e-05,
+      "loss": 0.0283,
+      "step": 31380
+    },
+    {
+      "epoch": 45.1,
+      "learning_rate": 2.9930029154518948e-05,
+      "loss": 0.0167,
+      "step": 31390
+    },
+    {
+      "epoch": 45.11,
+      "learning_rate": 2.9842565597667636e-05,
+      "loss": 0.0221,
+      "step": 31400
+    },
+    {
+      "epoch": 45.11,
+      "eval_loss": 0.1247408539056778,
+      "eval_runtime": 40.8758,
+      "eval_samples_per_second": 11.18,
+      "eval_steps_per_second": 1.419,
+      "eval_wer": 0.1183556693269086,
+      "step": 31400
+    },
+    {
+      "epoch": 45.13,
+      "learning_rate": 2.9755102040816323e-05,
+      "loss": 0.0251,
+      "step": 31410
+    },
+    {
+      "epoch": 45.14,
+      "learning_rate": 2.966763848396501e-05,
+      "loss": 0.0201,
+      "step": 31420
+    },
+    {
+      "epoch": 45.16,
+      "learning_rate": 2.9580174927113703e-05,
+      "loss": 0.0254,
+      "step": 31430
+    },
+    {
+      "epoch": 45.17,
+      "learning_rate": 2.949271137026239e-05,
+      "loss": 0.0194,
+      "step": 31440
+    },
+    {
+      "epoch": 45.19,
+      "learning_rate": 2.940524781341108e-05,
+      "loss": 0.0157,
+      "step": 31450
+    },
+    {
+      "epoch": 45.2,
+      "learning_rate": 2.9317784256559766e-05,
+      "loss": 0.0162,
+      "step": 31460
+    },
+    {
+      "epoch": 45.22,
+      "learning_rate": 2.923032069970845e-05,
+      "loss": 0.0142,
+      "step": 31470
+    },
+    {
+      "epoch": 45.23,
+      "learning_rate": 2.914285714285714e-05,
+      "loss": 0.0222,
+      "step": 31480
+    },
+    {
+      "epoch": 45.24,
+      "learning_rate": 2.9055393586005827e-05,
+      "loss": 0.0183,
+      "step": 31490
+    },
+    {
+      "epoch": 45.26,
+      "learning_rate": 2.8967930029154515e-05,
+      "loss": 0.0177,
+      "step": 31500
+    },
+    {
+      "epoch": 45.26,
+      "eval_loss": 0.1244613528251648,
+      "eval_runtime": 41.3124,
+      "eval_samples_per_second": 11.062,
+      "eval_steps_per_second": 1.404,
+      "eval_wer": 0.11925914771871705,
+      "step": 31500
+    },
+    {
+      "epoch": 45.27,
+      "learning_rate": 2.8880466472303202e-05,
+      "loss": 0.0196,
+      "step": 31510
+    },
+    {
+      "epoch": 45.29,
+      "learning_rate": 2.8793002915451894e-05,
+      "loss": 0.0131,
+      "step": 31520
+    },
+    {
+      "epoch": 45.3,
+      "learning_rate": 2.870553935860058e-05,
+      "loss": 0.0242,
+      "step": 31530
+    },
+    {
+      "epoch": 45.32,
+      "learning_rate": 2.861807580174927e-05,
+      "loss": 0.0172,
+      "step": 31540
+    },
+    {
+      "epoch": 45.33,
+      "learning_rate": 2.8530612244897957e-05,
+      "loss": 0.0177,
+      "step": 31550
+    },
+    {
+      "epoch": 45.34,
+      "learning_rate": 2.8443148688046645e-05,
+      "loss": 0.023,
+      "step": 31560
+    },
+    {
+      "epoch": 45.36,
+      "learning_rate": 2.8355685131195333e-05,
+      "loss": 0.0157,
+      "step": 31570
+    },
+    {
+      "epoch": 45.37,
+      "learning_rate": 2.826822157434402e-05,
+      "loss": 0.0215,
+      "step": 31580
+    },
+    {
+      "epoch": 45.39,
+      "learning_rate": 2.818075801749271e-05,
+      "loss": 0.017,
+      "step": 31590
+    },
+    {
+      "epoch": 45.4,
+      "learning_rate": 2.8093294460641397e-05,
+      "loss": 0.0174,
+      "step": 31600
+    },
+    {
+      "epoch": 45.4,
+      "eval_loss": 0.12396515160799026,
+      "eval_runtime": 40.8394,
+      "eval_samples_per_second": 11.19,
+      "eval_steps_per_second": 1.42,
+      "eval_wer": 0.11850624905887668,
+      "step": 31600
+    },
+    {
+      "epoch": 45.42,
+      "learning_rate": 2.800583090379008e-05,
+      "loss": 0.0211,
+      "step": 31610
+    },
+    {
+      "epoch": 45.43,
+      "learning_rate": 2.7918367346938776e-05,
+      "loss": 0.0171,
+      "step": 31620
+    },
+    {
+      "epoch": 45.45,
+      "learning_rate": 2.7830903790087464e-05,
+      "loss": 0.0171,
+      "step": 31630
+    },
+    {
+      "epoch": 45.46,
+      "learning_rate": 2.774344023323615e-05,
+      "loss": 0.0188,
+      "step": 31640
+    },
+    {
+      "epoch": 45.47,
+      "learning_rate": 2.7655976676384836e-05,
+      "loss": 0.023,
+      "step": 31650
+    },
+    {
+      "epoch": 45.49,
+      "learning_rate": 2.7568513119533524e-05,
+      "loss": 0.0168,
+      "step": 31660
+    },
+    {
+      "epoch": 45.5,
+      "learning_rate": 2.7481049562682212e-05,
+      "loss": 0.0179,
+      "step": 31670
+    },
+    {
+      "epoch": 45.52,
+      "learning_rate": 2.73935860058309e-05,
+      "loss": 0.0195,
+      "step": 31680
+    },
+    {
+      "epoch": 45.53,
+      "learning_rate": 2.7306122448979588e-05,
+      "loss": 0.017,
+      "step": 31690
+    },
+    {
+      "epoch": 45.55,
+      "learning_rate": 2.7218658892128276e-05,
+      "loss": 0.0149,
+      "step": 31700
+    },
+    {
+      "epoch": 45.55,
+      "eval_loss": 0.12391093373298645,
+      "eval_runtime": 40.7869,
+      "eval_samples_per_second": 11.205,
+      "eval_steps_per_second": 1.422,
+      "eval_wer": 0.1189579882547809,
+      "step": 31700
+    },
+    {
+      "epoch": 45.56,
+      "learning_rate": 2.7131195335276967e-05,
+      "loss": 0.0149,
+      "step": 31710
+    },
+    {
+      "epoch": 45.57,
+      "learning_rate": 2.7043731778425655e-05,
+      "loss": 0.0146,
+      "step": 31720
+    },
+    {
+      "epoch": 45.59,
+      "learning_rate": 2.6956268221574343e-05,
+      "loss": 0.0257,
+      "step": 31730
+    },
+    {
+      "epoch": 45.6,
+      "learning_rate": 2.686880466472303e-05,
+      "loss": 0.0181,
+      "step": 31740
+    },
+    {
+      "epoch": 45.62,
+      "learning_rate": 2.678134110787172e-05,
+      "loss": 0.0266,
+      "step": 31750
+    },
+    {
+      "epoch": 45.63,
+      "learning_rate": 2.6693877551020407e-05,
+      "loss": 0.0197,
+      "step": 31760
+    },
+    {
+      "epoch": 45.65,
+      "learning_rate": 2.6606413994169095e-05,
+      "loss": 0.0158,
+      "step": 31770
+    },
+    {
+      "epoch": 45.66,
+      "learning_rate": 2.651895043731778e-05,
+      "loss": 0.0191,
+      "step": 31780
+    },
+    {
+      "epoch": 45.68,
+      "learning_rate": 2.6431486880466467e-05,
+      "loss": 0.0206,
+      "step": 31790
+    },
+    {
+      "epoch": 45.69,
+      "learning_rate": 2.6344023323615158e-05,
+      "loss": 0.0156,
+      "step": 31800
+    },
+    {
+      "epoch": 45.69,
+      "eval_loss": 0.12179351598024368,
+      "eval_runtime": 40.676,
+      "eval_samples_per_second": 11.235,
+      "eval_steps_per_second": 1.426,
+      "eval_wer": 0.11910856798674899,
+      "step": 31800
+    },
+    {
+      "epoch": 45.7,
+      "learning_rate": 2.6256559766763846e-05,
+      "loss": 0.0167,
+      "step": 31810
+    },
+    {
+      "epoch": 45.72,
+      "learning_rate": 2.6169096209912534e-05,
+      "loss": 0.0165,
+      "step": 31820
+    },
+    {
+      "epoch": 45.73,
+      "learning_rate": 2.6081632653061222e-05,
+      "loss": 0.0179,
+      "step": 31830
+    },
+    {
+      "epoch": 45.75,
+      "learning_rate": 2.599416909620991e-05,
+      "loss": 0.0183,
+      "step": 31840
+    },
+    {
+      "epoch": 45.76,
+      "learning_rate": 2.5906705539358598e-05,
+      "loss": 0.0241,
+      "step": 31850
+    },
+    {
+      "epoch": 45.78,
+      "learning_rate": 2.5819241982507286e-05,
+      "loss": 0.0191,
+      "step": 31860
+    },
+    {
+      "epoch": 45.79,
+      "learning_rate": 2.5731778425655973e-05,
+      "loss": 0.0159,
+      "step": 31870
+    },
+    {
+      "epoch": 45.8,
+      "learning_rate": 2.564431486880466e-05,
+      "loss": 0.021,
+      "step": 31880
+    },
+    {
+      "epoch": 45.82,
+      "learning_rate": 2.5556851311953353e-05,
+      "loss": 0.0194,
+      "step": 31890
+    },
+    {
+      "epoch": 45.83,
+      "learning_rate": 2.546938775510204e-05,
+      "loss": 0.0276,
+      "step": 31900
+    },
+    {
+      "epoch": 45.83,
+      "eval_loss": 0.12291625887155533,
+      "eval_runtime": 40.8657,
+      "eval_samples_per_second": 11.183,
+      "eval_steps_per_second": 1.419,
+      "eval_wer": 0.11986146664658937,
+      "step": 31900
+    },
+    {
+      "epoch": 45.85,
+      "learning_rate": 2.538192419825073e-05,
+      "loss": 0.0214,
+      "step": 31910
+    },
+    {
+      "epoch": 45.86,
+      "learning_rate": 2.5294460641399416e-05,
+      "loss": 0.0105,
+      "step": 31920
+    },
+    {
+      "epoch": 45.88,
+      "learning_rate": 2.5206997084548104e-05,
+      "loss": 0.0221,
+      "step": 31930
+    },
+    {
+      "epoch": 45.89,
+      "learning_rate": 2.511953352769679e-05,
+      "loss": 0.0202,
+      "step": 31940
+    },
+    {
+      "epoch": 45.91,
+      "learning_rate": 2.5032069970845477e-05,
+      "loss": 0.0197,
+      "step": 31950
+    },
+    {
+      "epoch": 45.92,
+      "learning_rate": 2.4944606413994165e-05,
+      "loss": 0.0204,
+      "step": 31960
+    },
+    {
+      "epoch": 45.93,
+      "learning_rate": 2.4857142857142852e-05,
+      "loss": 0.0206,
+      "step": 31970
+    },
+    {
+      "epoch": 45.95,
+      "learning_rate": 2.4769679300291544e-05,
+      "loss": 0.0172,
+      "step": 31980
+    },
+    {
+      "epoch": 45.96,
+      "learning_rate": 2.468221574344023e-05,
+      "loss": 0.0149,
+      "step": 31990
+    },
+    {
+      "epoch": 45.98,
+      "learning_rate": 2.459475218658892e-05,
+      "loss": 0.0178,
+      "step": 32000
+    },
+    {
+      "epoch": 45.98,
+      "eval_loss": 0.12346994876861572,
+      "eval_runtime": 42.8222,
+      "eval_samples_per_second": 10.672,
+      "eval_steps_per_second": 1.354,
+      "eval_wer": 0.1189579882547809,
+      "step": 32000
+    },
+    {
+      "epoch": 45.99,
+      "learning_rate": 2.4507288629737607e-05,
+      "loss": 0.0166,
+      "step": 32010
+    },
+    {
+      "epoch": 46.01,
+      "learning_rate": 2.4419825072886295e-05,
+      "loss": 0.0176,
+      "step": 32020
+    },
+    {
+      "epoch": 46.02,
+      "learning_rate": 2.4332361516034983e-05,
+      "loss": 0.0188,
+      "step": 32030
+    },
+    {
+      "epoch": 46.03,
+      "learning_rate": 2.424489795918367e-05,
+      "loss": 0.0142,
+      "step": 32040
+    },
+    {
+      "epoch": 46.05,
+      "learning_rate": 2.415743440233236e-05,
+      "loss": 0.0315,
+      "step": 32050
+    },
+    {
+      "epoch": 46.06,
+      "learning_rate": 2.4069970845481047e-05,
+      "loss": 0.0154,
+      "step": 32060
+    },
+    {
+      "epoch": 46.08,
+      "learning_rate": 2.3982507288629738e-05,
+      "loss": 0.0166,
+      "step": 32070
+    },
+    {
+      "epoch": 46.09,
+      "learning_rate": 2.3895043731778426e-05,
+      "loss": 0.0164,
+      "step": 32080
+    },
+    {
+      "epoch": 46.11,
+      "learning_rate": 2.3807580174927114e-05,
+      "loss": 0.0166,
+      "step": 32090
+    },
+    {
+      "epoch": 46.12,
+      "learning_rate": 2.3720116618075802e-05,
+      "loss": 0.0171,
+      "step": 32100
+    },
+    {
+      "epoch": 46.12,
+      "eval_loss": 0.12300096452236176,
+      "eval_runtime": 40.7996,
+      "eval_samples_per_second": 11.201,
+      "eval_steps_per_second": 1.422,
+      "eval_wer": 0.11956030718265322,
+      "step": 32100
+    },
+    {
+      "epoch": 46.14,
+      "learning_rate": 2.3632653061224486e-05,
+      "loss": 0.0176,
+      "step": 32110
+    },
+    {
+      "epoch": 46.15,
+      "learning_rate": 2.3545189504373174e-05,
+      "loss": 0.0163,
+      "step": 32120
+    },
+    {
+      "epoch": 46.16,
+      "learning_rate": 2.3457725947521862e-05,
+      "loss": 0.0188,
+      "step": 32130
+    },
+    {
+      "epoch": 46.18,
+      "learning_rate": 2.337026239067055e-05,
+      "loss": 0.0153,
+      "step": 32140
+    },
+    {
+      "epoch": 46.19,
+      "learning_rate": 2.3282798833819238e-05,
+      "loss": 0.0167,
+      "step": 32150
+    },
+    {
+      "epoch": 46.21,
+      "learning_rate": 2.3195335276967926e-05,
+      "loss": 0.0179,
+      "step": 32160
+    },
+    {
+      "epoch": 46.22,
+      "learning_rate": 2.3107871720116617e-05,
+      "loss": 0.0158,
+      "step": 32170
+    },
+    {
+      "epoch": 46.24,
+      "learning_rate": 2.3020408163265305e-05,
+      "loss": 0.0192,
+      "step": 32180
+    },
+    {
+      "epoch": 46.25,
+      "learning_rate": 2.2932944606413993e-05,
+      "loss": 0.0139,
+      "step": 32190
+    },
+    {
+      "epoch": 46.26,
+      "learning_rate": 2.284548104956268e-05,
+      "loss": 0.0226,
+      "step": 32200
+    },
+    {
+      "epoch": 46.26,
+      "eval_loss": 0.1228545680642128,
+      "eval_runtime": 41.5098,
+      "eval_samples_per_second": 11.009,
+      "eval_steps_per_second": 1.397,
+      "eval_wer": 0.11925914771871705,
+      "step": 32200
+    },
+    {
+      "epoch": 46.28,
+      "learning_rate": 2.275801749271137e-05,
+      "loss": 0.0159,
+      "step": 32210
+    },
+    {
+      "epoch": 46.29,
+      "learning_rate": 2.2670553935860057e-05,
+      "loss": 0.0174,
+      "step": 32220
+    },
+    {
+      "epoch": 46.31,
+      "learning_rate": 2.2583090379008745e-05,
+      "loss": 0.0213,
+      "step": 32230
+    },
+    {
+      "epoch": 46.32,
+      "learning_rate": 2.2495626822157432e-05,
+      "loss": 0.0167,
+      "step": 32240
+    },
+    {
+      "epoch": 46.34,
+      "learning_rate": 2.2408163265306117e-05,
+      "loss": 0.0219,
+      "step": 32250
+    },
+    {
+      "epoch": 46.35,
+      "learning_rate": 2.232069970845481e-05,
+      "loss": 0.0137,
+      "step": 32260
+    },
+    {
+      "epoch": 46.36,
+      "learning_rate": 2.22332361516035e-05,
+      "loss": 0.0177,
+      "step": 32270
+    },
+    {
+      "epoch": 46.38,
+      "learning_rate": 2.2145772594752184e-05,
+      "loss": 0.0201,
+      "step": 32280
+    },
+    {
+      "epoch": 46.39,
+      "learning_rate": 2.2058309037900872e-05,
+      "loss": 0.0156,
+      "step": 32290
+    },
+    {
+      "epoch": 46.41,
+      "learning_rate": 2.197084548104956e-05,
+      "loss": 0.0222,
+      "step": 32300
+    },
+    {
+      "epoch": 46.41,
+      "eval_loss": 0.12293291836977005,
+      "eval_runtime": 40.8016,
+      "eval_samples_per_second": 11.201,
+      "eval_steps_per_second": 1.422,
+      "eval_wer": 0.1189579882547809,
+      "step": 32300
+    },
+    {
+      "epoch": 46.42,
+      "learning_rate": 2.1883381924198248e-05,
+      "loss": 0.0194,
+      "step": 32310
+    },
+    {
+      "epoch": 46.44,
+      "learning_rate": 2.1795918367346936e-05,
+      "loss": 0.0132,
+      "step": 32320
+    },
+    {
+      "epoch": 46.45,
+      "learning_rate": 2.1708454810495623e-05,
+      "loss": 0.0185,
+      "step": 32330
+    },
+    {
+      "epoch": 46.47,
+      "learning_rate": 2.162099125364431e-05,
+      "loss": 0.0228,
+      "step": 32340
+    },
+    {
+      "epoch": 46.48,
+      "learning_rate": 2.1533527696793003e-05,
+      "loss": 0.0249,
+      "step": 32350
+    },
+    {
+      "epoch": 46.49,
+      "learning_rate": 2.144606413994169e-05,
+      "loss": 0.0166,
+      "step": 32360
+    },
+    {
+      "epoch": 46.51,
+      "learning_rate": 2.135860058309038e-05,
+      "loss": 0.0187,
+      "step": 32370
+    },
+    {
+      "epoch": 46.52,
+      "learning_rate": 2.1271137026239066e-05,
+      "loss": 0.0215,
+      "step": 32380
+    },
+    {
+      "epoch": 46.54,
+      "learning_rate": 2.1183673469387754e-05,
+      "loss": 0.0113,
+      "step": 32390
+    },
+    {
+      "epoch": 46.55,
+      "learning_rate": 2.1096209912536442e-05,
+      "loss": 0.0208,
+      "step": 32400
+    },
+    {
+      "epoch": 46.55,
+      "eval_loss": 0.12255549430847168,
+      "eval_runtime": 41.0134,
+      "eval_samples_per_second": 11.143,
+      "eval_steps_per_second": 1.414,
+      "eval_wer": 0.11820508959494053,
+      "step": 32400
+    },
+    {
+      "epoch": 46.57,
+      "learning_rate": 2.100874635568513e-05,
+      "loss": 0.0179,
+      "step": 32410
+    },
+    {
+      "epoch": 46.58,
+      "learning_rate": 2.0921282798833815e-05,
+      "loss": 0.0192,
+      "step": 32420
+    },
+    {
+      "epoch": 46.59,
+      "learning_rate": 2.0833819241982502e-05,
+      "loss": 0.0183,
+      "step": 32430
+    },
+    {
+      "epoch": 46.61,
+      "learning_rate": 2.0746355685131197e-05,
+      "loss": 0.0142,
+      "step": 32440
+    },
+    {
+      "epoch": 46.62,
+      "learning_rate": 2.065889212827988e-05,
+      "loss": 0.021,
+      "step": 32450
+    },
+    {
+      "epoch": 46.64,
+      "learning_rate": 2.057142857142857e-05,
+      "loss": 0.0161,
+      "step": 32460
+    },
+    {
+      "epoch": 46.65,
+      "learning_rate": 2.0483965014577257e-05,
+      "loss": 0.0138,
+      "step": 32470
+    },
+    {
+      "epoch": 46.67,
+      "learning_rate": 2.0396501457725945e-05,
+      "loss": 0.0183,
+      "step": 32480
+    },
+    {
+      "epoch": 46.68,
+      "learning_rate": 2.0309037900874633e-05,
+      "loss": 0.0184,
+      "step": 32490
+    },
+    {
+      "epoch": 46.7,
+      "learning_rate": 2.022157434402332e-05,
+      "loss": 0.0176,
+      "step": 32500
+    },
+    {
+      "epoch": 46.7,
+      "eval_loss": 0.12446217983961105,
+      "eval_runtime": 40.8413,
+      "eval_samples_per_second": 11.19,
+      "eval_steps_per_second": 1.42,
+      "eval_wer": 0.11850624905887668,
+      "step": 32500
+    },
+    {
+      "epoch": 46.71,
+      "learning_rate": 2.013411078717201e-05,
+      "loss": 0.019,
+      "step": 32510
+    },
+    {
+      "epoch": 46.72,
+      "learning_rate": 2.0046647230320697e-05,
+      "loss": 0.0198,
+      "step": 32520
+    },
+    {
+      "epoch": 46.74,
+      "learning_rate": 1.9959183673469388e-05,
+      "loss": 0.0139,
+      "step": 32530
+    },
+    {
+      "epoch": 46.75,
+      "learning_rate": 1.9871720116618076e-05,
+      "loss": 0.0155,
+      "step": 32540
+    },
+    {
+      "epoch": 46.77,
+      "learning_rate": 1.9784256559766764e-05,
+      "loss": 0.0259,
+      "step": 32550
+    },
+    {
+      "epoch": 46.78,
+      "learning_rate": 1.9696793002915452e-05,
+      "loss": 0.017,
+      "step": 32560
+    },
+    {
+      "epoch": 46.8,
+      "learning_rate": 1.960932944606414e-05,
+      "loss": 0.0182,
+      "step": 32570
+    },
+    {
+      "epoch": 46.81,
+      "learning_rate": 1.9521865889212828e-05,
+      "loss": 0.0195,
+      "step": 32580
+    },
+    {
+      "epoch": 46.82,
+      "learning_rate": 1.9434402332361512e-05,
+      "loss": 0.0172,
+      "step": 32590
+    },
+    {
+      "epoch": 46.84,
+      "learning_rate": 1.93469387755102e-05,
+      "loss": 0.0202,
+      "step": 32600
+    },
+    {
+      "epoch": 46.84,
+      "eval_loss": 0.12425705790519714,
+      "eval_runtime": 42.4199,
+      "eval_samples_per_second": 10.773,
+      "eval_steps_per_second": 1.367,
+      "eval_wer": 0.11956030718265322,
+      "step": 32600
+    },
+    {
+      "epoch": 46.85,
+      "learning_rate": 1.9259475218658888e-05,
+      "loss": 0.0155,
+      "step": 32610
+    },
+    {
+      "epoch": 46.87,
+      "learning_rate": 1.9172011661807576e-05,
+      "loss": 0.0186,
+      "step": 32620
+    },
+    {
+      "epoch": 46.88,
+      "learning_rate": 1.9084548104956267e-05,
+      "loss": 0.0262,
+      "step": 32630
+    },
+    {
+      "epoch": 46.9,
+      "learning_rate": 1.8997084548104955e-05,
+      "loss": 0.0176,
+      "step": 32640
+    },
+    {
+      "epoch": 46.91,
+      "learning_rate": 1.8909620991253643e-05,
+      "loss": 0.0215,
+      "step": 32650
+    },
+    {
+      "epoch": 46.93,
+      "learning_rate": 1.882215743440233e-05,
+      "loss": 0.0141,
+      "step": 32660
+    },
+    {
+      "epoch": 46.94,
+      "learning_rate": 1.873469387755102e-05,
+      "loss": 0.0224,
+      "step": 32670
+    },
+    {
+      "epoch": 46.95,
+      "learning_rate": 1.8647230320699707e-05,
+      "loss": 0.0199,
+      "step": 32680
+    },
+    {
+      "epoch": 46.97,
+      "learning_rate": 1.8559766763848395e-05,
+      "loss": 0.0129,
+      "step": 32690
+    },
+    {
+      "epoch": 46.98,
+      "learning_rate": 1.8472303206997082e-05,
+      "loss": 0.0176,
+      "step": 32700
+    },
+    {
+      "epoch": 46.98,
+      "eval_loss": 0.12350737303495407,
+      "eval_runtime": 40.9221,
+      "eval_samples_per_second": 11.168,
+      "eval_steps_per_second": 1.417,
+      "eval_wer": 0.11956030718265322,
+      "step": 32700
+    },
+    {
+      "epoch": 47.0,
+      "learning_rate": 1.838483965014577e-05,
+      "loss": 0.017,
+      "step": 32710
+    },
+    {
+      "epoch": 47.01,
+      "learning_rate": 1.8297376093294458e-05,
+      "loss": 0.0208,
+      "step": 32720
+    },
+    {
+      "epoch": 47.03,
+      "learning_rate": 1.8209912536443146e-05,
+      "loss": 0.0144,
+      "step": 32730
+    },
+    {
+      "epoch": 47.04,
+      "learning_rate": 1.8122448979591837e-05,
+      "loss": 0.0164,
+      "step": 32740
+    },
+    {
+      "epoch": 47.05,
+      "learning_rate": 1.8034985422740522e-05,
+      "loss": 0.0174,
+      "step": 32750
+    },
+    {
+      "epoch": 47.07,
+      "learning_rate": 1.794752186588921e-05,
+      "loss": 0.0194,
+      "step": 32760
+    },
+    {
+      "epoch": 47.08,
+      "learning_rate": 1.7860058309037898e-05,
+      "loss": 0.0219,
+      "step": 32770
+    },
+    {
+      "epoch": 47.1,
+      "learning_rate": 1.7772594752186586e-05,
+      "loss": 0.0143,
+      "step": 32780
+    },
+    {
+      "epoch": 47.11,
+      "learning_rate": 1.7685131195335277e-05,
+      "loss": 0.0125,
+      "step": 32790
+    },
+    {
+      "epoch": 47.13,
+      "learning_rate": 1.7597667638483965e-05,
+      "loss": 0.0226,
+      "step": 32800
+    },
+    {
+      "epoch": 47.13,
+      "eval_loss": 0.12387290596961975,
+      "eval_runtime": 40.9699,
+      "eval_samples_per_second": 11.155,
+      "eval_steps_per_second": 1.416,
+      "eval_wer": 0.12001204637855745,
+      "step": 32800
+    },
+    {
+      "epoch": 47.14,
+      "learning_rate": 1.7510204081632653e-05,
+      "loss": 0.0169,
+      "step": 32810
+    },
+    {
+      "epoch": 47.16,
+      "learning_rate": 1.7422740524781337e-05,
+      "loss": 0.0186,
+      "step": 32820
+    },
+    {
+      "epoch": 47.17,
+      "learning_rate": 1.733527696793003e-05,
+      "loss": 0.0147,
+      "step": 32830
+    },
+    {
+      "epoch": 47.18,
+      "learning_rate": 1.7247813411078716e-05,
+      "loss": 0.0132,
+      "step": 32840
+    },
+    {
+      "epoch": 47.2,
+      "learning_rate": 1.7160349854227404e-05,
+      "loss": 0.0159,
+      "step": 32850
+    },
+    {
+      "epoch": 47.21,
+      "learning_rate": 1.7072886297376092e-05,
+      "loss": 0.0161,
+      "step": 32860
+    },
+    {
+      "epoch": 47.23,
+      "learning_rate": 1.698542274052478e-05,
+      "loss": 0.0151,
+      "step": 32870
+    },
+    {
+      "epoch": 47.24,
+      "learning_rate": 1.6897959183673468e-05,
+      "loss": 0.013,
+      "step": 32880
+    },
+    {
+      "epoch": 47.26,
+      "learning_rate": 1.6810495626822156e-05,
+      "loss": 0.0146,
+      "step": 32890
+    },
+    {
+      "epoch": 47.27,
+      "learning_rate": 1.6723032069970844e-05,
+      "loss": 0.0122,
+      "step": 32900
+    },
+    {
+      "epoch": 47.27,
+      "eval_loss": 0.12316489219665527,
+      "eval_runtime": 42.0627,
+      "eval_samples_per_second": 10.865,
+      "eval_steps_per_second": 1.379,
+      "eval_wer": 0.11940972745068514,
+      "step": 32900
+    },
+    {
+      "epoch": 47.28,
+      "learning_rate": 1.663556851311953e-05,
+      "loss": 0.0185,
+      "step": 32910
+    },
+    {
+      "epoch": 47.3,
+      "learning_rate": 1.654810495626822e-05,
+      "loss": 0.0215,
+      "step": 32920
+    },
+    {
+      "epoch": 47.31,
+      "learning_rate": 1.6460641399416907e-05,
+      "loss": 0.0183,
+      "step": 32930
+    },
+    {
+      "epoch": 47.33,
+      "learning_rate": 1.6373177842565595e-05,
+      "loss": 0.0147,
+      "step": 32940
+    },
+    {
+      "epoch": 47.34,
+      "learning_rate": 1.6285714285714283e-05,
+      "loss": 0.0171,
+      "step": 32950
+    },
+    {
+      "epoch": 47.36,
+      "learning_rate": 1.619825072886297e-05,
+      "loss": 0.0137,
+      "step": 32960
+    },
+    {
+      "epoch": 47.37,
+      "learning_rate": 1.6110787172011662e-05,
+      "loss": 0.0158,
+      "step": 32970
+    },
+    {
+      "epoch": 47.39,
+      "learning_rate": 1.602332361516035e-05,
+      "loss": 0.0102,
+      "step": 32980
+    },
+    {
+      "epoch": 47.4,
+      "learning_rate": 1.5935860058309035e-05,
+      "loss": 0.0119,
+      "step": 32990
+    },
+    {
+      "epoch": 47.41,
+      "learning_rate": 1.5848396501457723e-05,
+      "loss": 0.0185,
+      "step": 33000
+    },
+    {
+      "epoch": 47.41,
+      "eval_loss": 0.12409212440252304,
+      "eval_runtime": 41.2942,
+      "eval_samples_per_second": 11.067,
+      "eval_steps_per_second": 1.405,
+      "eval_wer": 0.11880740852281282,
+      "step": 33000
+    },
+    {
+      "epoch": 47.43,
+      "learning_rate": 1.5760932944606414e-05,
+      "loss": 0.0144,
+      "step": 33010
+    },
+    {
+      "epoch": 47.44,
+      "learning_rate": 1.5673469387755102e-05,
+      "loss": 0.0204,
+      "step": 33020
+    },
+    {
+      "epoch": 47.46,
+      "learning_rate": 1.558600583090379e-05,
+      "loss": 0.0172,
+      "step": 33030
+    },
+    {
+      "epoch": 47.47,
+      "learning_rate": 1.5498542274052478e-05,
+      "loss": 0.0179,
+      "step": 33040
+    },
+    {
+      "epoch": 47.49,
+      "learning_rate": 1.5411078717201166e-05,
+      "loss": 0.0153,
+      "step": 33050
+    },
+    {
+      "epoch": 47.5,
+      "learning_rate": 1.5323615160349853e-05,
+      "loss": 0.0206,
+      "step": 33060
+    },
+    {
+      "epoch": 47.51,
+      "learning_rate": 1.5236151603498541e-05,
+      "loss": 0.0158,
+      "step": 33070
+    },
+    {
+      "epoch": 47.53,
+      "learning_rate": 1.514868804664723e-05,
+      "loss": 0.0187,
+      "step": 33080
+    },
+    {
+      "epoch": 47.54,
+      "learning_rate": 1.5061224489795917e-05,
+      "loss": 0.0162,
+      "step": 33090
+    },
+    {
+      "epoch": 47.56,
+      "learning_rate": 1.4973760932944603e-05,
+      "loss": 0.0198,
+      "step": 33100
+    },
+    {
+      "epoch": 47.56,
+      "eval_loss": 0.12408657371997833,
+      "eval_runtime": 41.0394,
+      "eval_samples_per_second": 11.136,
+      "eval_steps_per_second": 1.413,
+      "eval_wer": 0.11925914771871705,
+      "step": 33100
+    },
+    {
+      "epoch": 47.57,
+      "learning_rate": 1.4886297376093295e-05,
+      "loss": 0.016,
+      "step": 33110
+    },
+    {
+      "epoch": 47.59,
+      "learning_rate": 1.4798833819241981e-05,
+      "loss": 0.0162,
+      "step": 33120
+    },
+    {
+      "epoch": 47.6,
+      "learning_rate": 1.4711370262390669e-05,
+      "loss": 0.0177,
+      "step": 33130
+    },
+    {
+      "epoch": 47.61,
+      "learning_rate": 1.4623906705539357e-05,
+      "loss": 0.0133,
+      "step": 33140
+    },
+    {
+      "epoch": 47.63,
+      "learning_rate": 1.4536443148688046e-05,
+      "loss": 0.0181,
+      "step": 33150
+    },
+    {
+      "epoch": 47.64,
+      "learning_rate": 1.4448979591836734e-05,
+      "loss": 0.0191,
+      "step": 33160
+    },
+    {
+      "epoch": 47.66,
+      "learning_rate": 1.4361516034985422e-05,
+      "loss": 0.0241,
+      "step": 33170
+    },
+    {
+      "epoch": 47.67,
+      "learning_rate": 1.427405247813411e-05,
+      "loss": 0.0197,
+      "step": 33180
+    },
+    {
+      "epoch": 47.69,
+      "learning_rate": 1.4186588921282796e-05,
+      "loss": 0.0198,
+      "step": 33190
+    },
+    {
+      "epoch": 47.7,
+      "learning_rate": 1.4099125364431486e-05,
+      "loss": 0.0184,
+      "step": 33200
+    },
+    {
+      "epoch": 47.7,
+      "eval_loss": 0.12426743656396866,
+      "eval_runtime": 41.2473,
+      "eval_samples_per_second": 11.08,
+      "eval_steps_per_second": 1.406,
+      "eval_wer": 0.11820508959494053,
+      "step": 33200
+    },
+    {
+      "epoch": 47.72,
+      "learning_rate": 1.4011661807580174e-05,
+      "loss": 0.0178,
+      "step": 33210
+    },
+    {
+      "epoch": 47.73,
+      "learning_rate": 1.3924198250728862e-05,
+      "loss": 0.0225,
+      "step": 33220
+    },
+    {
+      "epoch": 47.74,
+      "learning_rate": 1.383673469387755e-05,
+      "loss": 0.013,
+      "step": 33230
+    },
+    {
+      "epoch": 47.76,
+      "learning_rate": 1.3749271137026239e-05,
+      "loss": 0.0106,
+      "step": 33240
+    },
+    {
+      "epoch": 47.77,
+      "learning_rate": 1.3661807580174927e-05,
+      "loss": 0.0202,
+      "step": 33250
+    },
+    {
+      "epoch": 47.79,
+      "learning_rate": 1.3574344023323615e-05,
+      "loss": 0.015,
+      "step": 33260
+    },
+    {
+      "epoch": 47.8,
+      "learning_rate": 1.3486880466472301e-05,
+      "loss": 0.0233,
+      "step": 33270
+    },
+    {
+      "epoch": 47.82,
+      "learning_rate": 1.3399416909620989e-05,
+      "loss": 0.0155,
+      "step": 33280
+    },
+    {
+      "epoch": 47.83,
+      "learning_rate": 1.3311953352769678e-05,
+      "loss": 0.0182,
+      "step": 33290
+    },
+    {
+      "epoch": 47.84,
+      "learning_rate": 1.3224489795918366e-05,
+      "loss": 0.0233,
+      "step": 33300
+    },
+    {
+      "epoch": 47.84,
+      "eval_loss": 0.12506160140037537,
+      "eval_runtime": 41.1898,
+      "eval_samples_per_second": 11.095,
+      "eval_steps_per_second": 1.408,
+      "eval_wer": 0.11745219093510013,
+      "step": 33300
+    },
+    {
+      "epoch": 47.86,
+      "learning_rate": 1.3137026239067054e-05,
+      "loss": 0.0202,
+      "step": 33310
+    },
+    {
+      "epoch": 47.87,
+      "learning_rate": 1.3049562682215742e-05,
+      "loss": 0.0223,
+      "step": 33320
+    },
+    {
+      "epoch": 47.89,
+      "learning_rate": 1.296209912536443e-05,
+      "loss": 0.0122,
+      "step": 33330
+    },
+    {
+      "epoch": 47.9,
+      "learning_rate": 1.287463556851312e-05,
+      "loss": 0.0144,
+      "step": 33340
+    },
+    {
+      "epoch": 47.92,
+      "learning_rate": 1.2787172011661806e-05,
+      "loss": 0.0169,
+      "step": 33350
+    },
+    {
+      "epoch": 47.93,
+      "learning_rate": 1.2699708454810494e-05,
+      "loss": 0.0131,
+      "step": 33360
+    },
+    {
+      "epoch": 47.95,
+      "learning_rate": 1.2612244897959182e-05,
+      "loss": 0.021,
+      "step": 33370
+    },
+    {
+      "epoch": 47.96,
+      "learning_rate": 1.2524781341107871e-05,
+      "loss": 0.0194,
+      "step": 33380
+    },
+    {
+      "epoch": 47.97,
+      "learning_rate": 1.2437317784256559e-05,
+      "loss": 0.0174,
+      "step": 33390
+    },
+    {
+      "epoch": 47.99,
+      "learning_rate": 1.2349854227405247e-05,
+      "loss": 0.0137,
+      "step": 33400
+    },
+    {
+      "epoch": 47.99,
+      "eval_loss": 0.12359973043203354,
+      "eval_runtime": 40.9137,
+      "eval_samples_per_second": 11.17,
+      "eval_steps_per_second": 1.418,
+      "eval_wer": 0.11669929227525976,
+      "step": 33400
+    },
+    {
+      "epoch": 48.0,
+      "learning_rate": 1.2262390670553935e-05,
+      "loss": 0.0138,
+      "step": 33410
+    },
+    {
+      "epoch": 48.02,
+      "learning_rate": 1.2174927113702621e-05,
+      "loss": 0.0119,
+      "step": 33420
+    },
+    {
+      "epoch": 48.03,
+      "learning_rate": 1.2087463556851312e-05,
+      "loss": 0.014,
+      "step": 33430
+    },
+    {
+      "epoch": 48.05,
+      "learning_rate": 1.1999999999999999e-05,
+      "loss": 0.0235,
+      "step": 33440
+    },
+    {
+      "epoch": 48.06,
+      "learning_rate": 1.1912536443148687e-05,
+      "loss": 0.0219,
+      "step": 33450
+    },
+    {
+      "epoch": 48.07,
+      "learning_rate": 1.1825072886297374e-05,
+      "loss": 0.014,
+      "step": 33460
+    },
+    {
+      "epoch": 48.09,
+      "learning_rate": 1.1737609329446064e-05,
+      "loss": 0.0245,
+      "step": 33470
+    },
+    {
+      "epoch": 48.1,
+      "learning_rate": 1.1650145772594752e-05,
+      "loss": 0.0179,
+      "step": 33480
+    },
+    {
+      "epoch": 48.12,
+      "learning_rate": 1.156268221574344e-05,
+      "loss": 0.0327,
+      "step": 33490
+    },
+    {
+      "epoch": 48.13,
+      "learning_rate": 1.1475218658892128e-05,
+      "loss": 0.0149,
+      "step": 33500
+    },
+    {
+      "epoch": 48.13,
+      "eval_loss": 0.12391505390405655,
+      "eval_runtime": 40.4562,
+      "eval_samples_per_second": 11.296,
+      "eval_steps_per_second": 1.434,
+      "eval_wer": 0.11654871254329167,
+      "step": 33500
+    },
+    {
+      "epoch": 48.15,
+      "learning_rate": 1.1387755102040814e-05,
+      "loss": 0.0124,
+      "step": 33510
+    },
+    {
+      "epoch": 48.16,
+      "learning_rate": 1.1300291545189503e-05,
+      "loss": 0.0215,
+      "step": 33520
+    },
+    {
+      "epoch": 48.18,
+      "learning_rate": 1.1212827988338191e-05,
+      "loss": 0.0111,
+      "step": 33530
+    },
+    {
+      "epoch": 48.19,
+      "learning_rate": 1.112536443148688e-05,
+      "loss": 0.0195,
+      "step": 33540
+    },
+    {
+      "epoch": 48.2,
+      "learning_rate": 1.1037900874635567e-05,
+      "loss": 0.0186,
+      "step": 33550
+    },
+    {
+      "epoch": 48.22,
+      "learning_rate": 1.0950437317784255e-05,
+      "loss": 0.0154,
+      "step": 33560
+    },
+    {
+      "epoch": 48.23,
+      "learning_rate": 1.0862973760932945e-05,
+      "loss": 0.0162,
+      "step": 33570
+    },
+    {
+      "epoch": 48.25,
+      "learning_rate": 1.0775510204081633e-05,
+      "loss": 0.0134,
+      "step": 33580
+    },
+    {
+      "epoch": 48.26,
+      "learning_rate": 1.0688046647230319e-05,
+      "loss": 0.022,
+      "step": 33590
+    },
+    {
+      "epoch": 48.28,
+      "learning_rate": 1.0600583090379007e-05,
+      "loss": 0.0129,
+      "step": 33600
+    },
+    {
+      "epoch": 48.28,
+      "eval_loss": 0.12475401163101196,
+      "eval_runtime": 40.7809,
+      "eval_samples_per_second": 11.206,
+      "eval_steps_per_second": 1.422,
+      "eval_wer": 0.11624755307935553,
+      "step": 33600
+    },
+    {
+      "epoch": 48.29,
+      "learning_rate": 1.0513119533527696e-05,
+      "loss": 0.0161,
+      "step": 33610
+    },
+    {
+      "epoch": 48.3,
+      "learning_rate": 1.0425655976676384e-05,
+      "loss": 0.0174,
+      "step": 33620
+    },
+    {
+      "epoch": 48.32,
+      "learning_rate": 1.0338192419825072e-05,
+      "loss": 0.0114,
+      "step": 33630
+    },
+    {
+      "epoch": 48.33,
+      "learning_rate": 1.025072886297376e-05,
+      "loss": 0.0237,
+      "step": 33640
+    },
+    {
+      "epoch": 48.35,
+      "learning_rate": 1.0163265306122448e-05,
+      "loss": 0.0142,
+      "step": 33650
+    },
+    {
+      "epoch": 48.36,
+      "learning_rate": 1.0075801749271137e-05,
+      "loss": 0.0162,
+      "step": 33660
+    },
+    {
+      "epoch": 48.38,
+      "learning_rate": 9.988338192419824e-06,
+      "loss": 0.0225,
+      "step": 33670
+    },
+    {
+      "epoch": 48.39,
+      "learning_rate": 9.900874635568512e-06,
+      "loss": 0.0172,
+      "step": 33680
+    },
+    {
+      "epoch": 48.41,
+      "learning_rate": 9.8134110787172e-06,
+      "loss": 0.0178,
+      "step": 33690
+    },
+    {
+      "epoch": 48.42,
+      "learning_rate": 9.725947521865889e-06,
+      "loss": 0.0162,
+      "step": 33700
+    },
+    {
+      "epoch": 48.42,
+      "eval_loss": 0.12392963469028473,
+      "eval_runtime": 40.7699,
+      "eval_samples_per_second": 11.209,
+      "eval_steps_per_second": 1.423,
+      "eval_wer": 0.11624755307935553,
+      "step": 33700
+    },
+    {
+      "epoch": 48.43,
+      "learning_rate": 9.638483965014577e-06,
+      "loss": 0.0156,
+      "step": 33710
+    },
+    {
+      "epoch": 48.45,
+      "learning_rate": 9.551020408163265e-06,
+      "loss": 0.0183,
+      "step": 33720
+    },
+    {
+      "epoch": 48.46,
+      "learning_rate": 9.463556851311953e-06,
+      "loss": 0.013,
+      "step": 33730
+    },
+    {
+      "epoch": 48.48,
+      "learning_rate": 9.376093294460639e-06,
+      "loss": 0.0163,
+      "step": 33740
+    },
+    {
+      "epoch": 48.49,
+      "learning_rate": 9.288629737609328e-06,
+      "loss": 0.017,
+      "step": 33750
+    },
+    {
+      "epoch": 48.51,
+      "learning_rate": 9.201166180758016e-06,
+      "loss": 0.0142,
+      "step": 33760
+    },
+    {
+      "epoch": 48.52,
+      "learning_rate": 9.113702623906704e-06,
+      "loss": 0.0212,
+      "step": 33770
+    },
+    {
+      "epoch": 48.53,
+      "learning_rate": 9.026239067055394e-06,
+      "loss": 0.0181,
+      "step": 33780
+    },
+    {
+      "epoch": 48.55,
+      "learning_rate": 8.93877551020408e-06,
+      "loss": 0.0258,
+      "step": 33790
+    },
+    {
+      "epoch": 48.56,
+      "learning_rate": 8.851311953352768e-06,
+      "loss": 0.0173,
+      "step": 33800
+    },
+    {
+      "epoch": 48.56,
+      "eval_loss": 0.12351784110069275,
+      "eval_runtime": 40.8564,
+      "eval_samples_per_second": 11.186,
+      "eval_steps_per_second": 1.42,
+      "eval_wer": 0.11669929227525976,
+      "step": 33800
+    },
+    {
+      "epoch": 48.58,
+      "learning_rate": 8.763848396501458e-06,
+      "loss": 0.0156,
+      "step": 33810
+    },
+    {
+      "epoch": 48.59,
+      "learning_rate": 8.676384839650145e-06,
+      "loss": 0.0223,
+      "step": 33820
+    },
+    {
+      "epoch": 48.61,
+      "learning_rate": 8.588921282798833e-06,
+      "loss": 0.0157,
+      "step": 33830
+    },
+    {
+      "epoch": 48.62,
+      "learning_rate": 8.501457725947521e-06,
+      "loss": 0.0158,
+      "step": 33840
+    },
+    {
+      "epoch": 48.64,
+      "learning_rate": 8.413994169096209e-06,
+      "loss": 0.0166,
+      "step": 33850
+    },
+    {
+      "epoch": 48.65,
+      "learning_rate": 8.326530612244897e-06,
+      "loss": 0.0195,
+      "step": 33860
+    },
+    {
+      "epoch": 48.66,
+      "learning_rate": 8.239067055393587e-06,
+      "loss": 0.0182,
+      "step": 33870
+    },
+    {
+      "epoch": 48.68,
+      "learning_rate": 8.151603498542273e-06,
+      "loss": 0.0131,
+      "step": 33880
+    },
+    {
+      "epoch": 48.69,
+      "learning_rate": 8.06413994169096e-06,
+      "loss": 0.0186,
+      "step": 33890
+    },
+    {
+      "epoch": 48.71,
+      "learning_rate": 7.97667638483965e-06,
+      "loss": 0.0144,
+      "step": 33900
+    },
+    {
+      "epoch": 48.71,
+      "eval_loss": 0.12300607562065125,
+      "eval_runtime": 40.8193,
+      "eval_samples_per_second": 11.196,
+      "eval_steps_per_second": 1.421,
+      "eval_wer": 0.11624755307935553,
+      "step": 33900
+    },
+    {
+      "epoch": 48.72,
+      "learning_rate": 7.889212827988337e-06,
+      "loss": 0.0113,
+      "step": 33910
+    },
+    {
+      "epoch": 48.74,
+      "learning_rate": 7.801749271137026e-06,
+      "loss": 0.0163,
+      "step": 33920
+    },
+    {
+      "epoch": 48.75,
+      "learning_rate": 7.714285714285714e-06,
+      "loss": 0.0146,
+      "step": 33930
+    },
+    {
+      "epoch": 48.76,
+      "learning_rate": 7.626822157434402e-06,
+      "loss": 0.0151,
+      "step": 33940
+    },
+    {
+      "epoch": 48.78,
+      "learning_rate": 7.53935860058309e-06,
+      "loss": 0.0229,
+      "step": 33950
+    },
+    {
+      "epoch": 48.79,
+      "learning_rate": 7.451895043731778e-06,
+      "loss": 0.0155,
+      "step": 33960
+    },
+    {
+      "epoch": 48.81,
+      "learning_rate": 7.364431486880466e-06,
+      "loss": 0.0226,
+      "step": 33970
+    },
+    {
+      "epoch": 48.82,
+      "learning_rate": 7.2769679300291535e-06,
+      "loss": 0.0172,
+      "step": 33980
+    },
+    {
+      "epoch": 48.84,
+      "learning_rate": 7.189504373177842e-06,
+      "loss": 0.0254,
+      "step": 33990
+    },
+    {
+      "epoch": 48.85,
+      "learning_rate": 7.10204081632653e-06,
+      "loss": 0.0194,
+      "step": 34000
+    },
+    {
+      "epoch": 48.85,
+      "eval_loss": 0.12407871335744858,
+      "eval_runtime": 40.7097,
+      "eval_samples_per_second": 11.226,
+      "eval_steps_per_second": 1.425,
+      "eval_wer": 0.1170004517391959,
+      "step": 34000
+    },
+    {
+      "epoch": 48.86,
+      "learning_rate": 7.014577259475219e-06,
+      "loss": 0.0179,
+      "step": 34010
+    },
+    {
+      "epoch": 48.88,
+      "learning_rate": 6.927113702623906e-06,
+      "loss": 0.0244,
+      "step": 34020
+    },
+    {
+      "epoch": 48.89,
+      "learning_rate": 6.839650145772594e-06,
+      "loss": 0.0152,
+      "step": 34030
+    },
+    {
+      "epoch": 48.91,
+      "learning_rate": 6.7521865889212826e-06,
+      "loss": 0.0178,
+      "step": 34040
+    },
+    {
+      "epoch": 48.92,
+      "learning_rate": 6.66472303206997e-06,
+      "loss": 0.0164,
+      "step": 34050
+    },
+    {
+      "epoch": 48.94,
+      "learning_rate": 6.577259475218658e-06,
+      "loss": 0.0145,
+      "step": 34060
+    },
+    {
+      "epoch": 48.95,
+      "learning_rate": 6.489795918367346e-06,
+      "loss": 0.015,
+      "step": 34070
+    },
+    {
+      "epoch": 48.97,
+      "learning_rate": 6.402332361516035e-06,
+      "loss": 0.0096,
+      "step": 34080
+    },
+    {
+      "epoch": 48.98,
+      "learning_rate": 6.314868804664722e-06,
+      "loss": 0.0191,
+      "step": 34090
+    },
+    {
+      "epoch": 48.99,
+      "learning_rate": 6.227405247813411e-06,
+      "loss": 0.0164,
+      "step": 34100
+    },
+    {
+      "epoch": 48.99,
+      "eval_loss": 0.12447045743465424,
+      "eval_runtime": 40.8776,
+      "eval_samples_per_second": 11.18,
+      "eval_steps_per_second": 1.419,
+      "eval_wer": 0.11745219093510013,
+      "step": 34100
+    },
+    {
+      "epoch": 49.01,
+      "learning_rate": 6.139941690962099e-06,
+      "loss": 0.0251,
+      "step": 34110
+    },
+    {
+      "epoch": 49.02,
+      "learning_rate": 6.0524781341107866e-06,
+      "loss": 0.0203,
+      "step": 34120
+    },
+    {
+      "epoch": 49.04,
+      "learning_rate": 5.9650145772594745e-06,
+      "loss": 0.0193,
+      "step": 34130
+    },
+    {
+      "epoch": 49.05,
+      "learning_rate": 5.877551020408162e-06,
+      "loss": 0.0195,
+      "step": 34140
+    },
+    {
+      "epoch": 49.07,
+      "learning_rate": 5.790087463556851e-06,
+      "loss": 0.0194,
+      "step": 34150
+    },
+    {
+      "epoch": 49.08,
+      "learning_rate": 5.702623906705539e-06,
+      "loss": 0.0212,
+      "step": 34160
+    },
+    {
+      "epoch": 49.09,
+      "learning_rate": 5.615160349854228e-06,
+      "loss": 0.0183,
+      "step": 34170
+    },
+    {
+      "epoch": 49.11,
+      "learning_rate": 5.527696793002915e-06,
+      "loss": 0.0129,
+      "step": 34180
+    },
+    {
+      "epoch": 49.12,
+      "learning_rate": 5.440233236151603e-06,
+      "loss": 0.018,
+      "step": 34190
+    },
+    {
+      "epoch": 49.14,
+      "learning_rate": 5.3527696793002914e-06,
+      "loss": 0.0144,
+      "step": 34200
+    },
+    {
+      "epoch": 49.14,
+      "eval_loss": 0.1245112419128418,
+      "eval_runtime": 41.1029,
+      "eval_samples_per_second": 11.118,
+      "eval_steps_per_second": 1.411,
+      "eval_wer": 0.11609697334738744,
+      "step": 34200
+    },
+    {
+      "epoch": 49.15,
+      "learning_rate": 5.2653061224489785e-06,
+      "loss": 0.0193,
+      "step": 34210
+    },
+    {
+      "epoch": 49.17,
+      "learning_rate": 5.177842565597667e-06,
+      "loss": 0.0215,
+      "step": 34220
+    },
+    {
+      "epoch": 49.18,
+      "learning_rate": 5.090379008746355e-06,
+      "loss": 0.0131,
+      "step": 34230
+    },
+    {
+      "epoch": 49.2,
+      "learning_rate": 5.002915451895044e-06,
+      "loss": 0.0222,
+      "step": 34240
+    },
+    {
+      "epoch": 49.21,
+      "learning_rate": 4.915451895043731e-06,
+      "loss": 0.0166,
+      "step": 34250
+    },
+    {
+      "epoch": 49.22,
+      "learning_rate": 4.82798833819242e-06,
+      "loss": 0.0215,
+      "step": 34260
+    },
+    {
+      "epoch": 49.24,
+      "learning_rate": 4.7405247813411076e-06,
+      "loss": 0.014,
+      "step": 34270
+    },
+    {
+      "epoch": 49.25,
+      "learning_rate": 4.6530612244897954e-06,
+      "loss": 0.0125,
+      "step": 34280
+    },
+    {
+      "epoch": 49.27,
+      "learning_rate": 4.565597667638483e-06,
+      "loss": 0.0168,
+      "step": 34290
+    },
+    {
+      "epoch": 49.28,
+      "learning_rate": 4.478134110787172e-06,
+      "loss": 0.0209,
+      "step": 34300
+    },
+    {
+      "epoch": 49.28,
+      "eval_loss": 0.12437942624092102,
+      "eval_runtime": 40.865,
+      "eval_samples_per_second": 11.183,
+      "eval_steps_per_second": 1.419,
+      "eval_wer": 0.11609697334738744,
+      "step": 34300
+    },
+    {
+      "epoch": 49.3,
+      "learning_rate": 4.390670553935859e-06,
+      "loss": 0.0158,
+      "step": 34310
+    },
+    {
+      "epoch": 49.31,
+      "learning_rate": 4.303206997084548e-06,
+      "loss": 0.0141,
+      "step": 34320
+    },
+    {
+      "epoch": 49.32,
+      "learning_rate": 4.215743440233236e-06,
+      "loss": 0.0186,
+      "step": 34330
+    },
+    {
+      "epoch": 49.34,
+      "learning_rate": 4.128279883381924e-06,
+      "loss": 0.0244,
+      "step": 34340
+    },
+    {
+      "epoch": 49.35,
+      "learning_rate": 4.0408163265306116e-06,
+      "loss": 0.0122,
+      "step": 34350
+    },
+    {
+      "epoch": 49.37,
+      "learning_rate": 3.9533527696793e-06,
+      "loss": 0.0155,
+      "step": 34360
+    },
+    {
+      "epoch": 49.38,
+      "learning_rate": 3.865889212827988e-06,
+      "loss": 0.0133,
+      "step": 34370
+    },
+    {
+      "epoch": 49.4,
+      "learning_rate": 3.778425655976676e-06,
+      "loss": 0.0131,
+      "step": 34380
+    },
+    {
+      "epoch": 49.41,
+      "learning_rate": 3.690962099125364e-06,
+      "loss": 0.0167,
+      "step": 34390
+    },
+    {
+      "epoch": 49.43,
+      "learning_rate": 3.603498542274052e-06,
+      "loss": 0.0168,
+      "step": 34400
+    },
+    {
+      "epoch": 49.43,
+      "eval_loss": 0.12430938333272934,
+      "eval_runtime": 40.7925,
+      "eval_samples_per_second": 11.203,
+      "eval_steps_per_second": 1.422,
+      "eval_wer": 0.11639813281132359,
+      "step": 34400
+    },
+    {
+      "epoch": 49.44,
+      "learning_rate": 3.5160349854227402e-06,
+      "loss": 0.0211,
+      "step": 34410
+    },
+    {
+      "epoch": 49.45,
+      "learning_rate": 3.428571428571428e-06,
+      "loss": 0.0123,
+      "step": 34420
+    },
+    {
+      "epoch": 49.47,
+      "learning_rate": 3.3411078717201164e-06,
+      "loss": 0.0147,
+      "step": 34430
+    },
+    {
+      "epoch": 49.48,
+      "learning_rate": 3.2536443148688043e-06,
+      "loss": 0.0174,
+      "step": 34440
+    },
+    {
+      "epoch": 49.5,
+      "learning_rate": 3.1661807580174926e-06,
+      "loss": 0.0128,
+      "step": 34450
+    },
+    {
+      "epoch": 49.51,
+      "learning_rate": 3.07871720116618e-06,
+      "loss": 0.0208,
+      "step": 34460
+    },
+    {
+      "epoch": 49.53,
+      "learning_rate": 2.9912536443148684e-06,
+      "loss": 0.017,
+      "step": 34470
+    },
+    {
+      "epoch": 49.54,
+      "learning_rate": 2.9037900874635563e-06,
+      "loss": 0.0192,
+      "step": 34480
+    },
+    {
+      "epoch": 49.55,
+      "learning_rate": 2.8163265306122447e-06,
+      "loss": 0.0182,
+      "step": 34490
+    },
+    {
+      "epoch": 49.57,
+      "learning_rate": 2.7288629737609326e-06,
+      "loss": 0.0163,
+      "step": 34500
+    },
+    {
+      "epoch": 49.57,
+      "eval_loss": 0.12417034804821014,
+      "eval_runtime": 40.9643,
+      "eval_samples_per_second": 11.156,
+      "eval_steps_per_second": 1.416,
+      "eval_wer": 0.11654871254329167,
+      "step": 34500
+    },
+    {
+      "epoch": 49.58,
+      "learning_rate": 2.641399416909621e-06,
+      "loss": 0.0283,
+      "step": 34510
+    },
+    {
+      "epoch": 49.6,
+      "learning_rate": 2.5539358600583088e-06,
+      "loss": 0.0136,
+      "step": 34520
+    },
+    {
+      "epoch": 49.61,
+      "learning_rate": 2.466472303206997e-06,
+      "loss": 0.0163,
+      "step": 34530
+    },
+    {
+      "epoch": 49.63,
+      "learning_rate": 2.3790087463556846e-06,
+      "loss": 0.0204,
+      "step": 34540
+    },
+    {
+      "epoch": 49.64,
+      "learning_rate": 2.291545189504373e-06,
+      "loss": 0.0119,
+      "step": 34550
+    },
+    {
+      "epoch": 49.66,
+      "learning_rate": 2.2040816326530608e-06,
+      "loss": 0.0162,
+      "step": 34560
+    },
+    {
+      "epoch": 49.67,
+      "learning_rate": 2.116618075801749e-06,
+      "loss": 0.0153,
+      "step": 34570
+    },
+    {
+      "epoch": 49.68,
+      "learning_rate": 2.029154518950437e-06,
+      "loss": 0.0156,
+      "step": 34580
+    },
+    {
+      "epoch": 49.7,
+      "learning_rate": 1.9416909620991253e-06,
+      "loss": 0.0153,
+      "step": 34590
+    },
+    {
+      "epoch": 49.71,
+      "learning_rate": 1.8542274052478132e-06,
+      "loss": 0.0128,
+      "step": 34600
+    },
+    {
+      "epoch": 49.71,
+      "eval_loss": 0.12449096143245697,
+      "eval_runtime": 40.8869,
+      "eval_samples_per_second": 11.177,
+      "eval_steps_per_second": 1.419,
+      "eval_wer": 0.11624755307935553,
+      "step": 34600
+    },
+    {
+      "epoch": 49.73,
+      "learning_rate": 1.7667638483965013e-06,
+      "loss": 0.0219,
+      "step": 34610
+    },
+    {
+      "epoch": 49.74,
+      "learning_rate": 1.6793002915451894e-06,
+      "loss": 0.0164,
+      "step": 34620
+    },
+    {
+      "epoch": 49.76,
+      "learning_rate": 1.5918367346938775e-06,
+      "loss": 0.0118,
+      "step": 34630
+    },
+    {
+      "epoch": 49.77,
+      "learning_rate": 1.5043731778425654e-06,
+      "loss": 0.0197,
+      "step": 34640
+    },
+    {
+      "epoch": 49.78,
+      "learning_rate": 1.4169096209912535e-06,
+      "loss": 0.017,
+      "step": 34650
+    },
+    {
+      "epoch": 49.8,
+      "learning_rate": 1.3294460641399416e-06,
+      "loss": 0.0192,
+      "step": 34660
+    },
+    {
+      "epoch": 49.81,
+      "learning_rate": 1.2419825072886298e-06,
+      "loss": 0.0164,
+      "step": 34670
+    },
+    {
+      "epoch": 49.83,
+      "learning_rate": 1.1545189504373177e-06,
+      "loss": 0.0093,
+      "step": 34680
+    },
+    {
+      "epoch": 49.84,
+      "learning_rate": 1.0670553935860058e-06,
+      "loss": 0.018,
+      "step": 34690
+    },
+    {
+      "epoch": 49.86,
+      "learning_rate": 9.795918367346939e-07,
+      "loss": 0.0179,
+      "step": 34700
+    },
+    {
+      "epoch": 49.86,
+      "eval_loss": 0.12448325753211975,
+      "eval_runtime": 40.9585,
+      "eval_samples_per_second": 11.158,
+      "eval_steps_per_second": 1.416,
+      "eval_wer": 0.11639813281132359,
+      "step": 34700
+    },
+    {
+      "epoch": 49.87,
+      "learning_rate": 8.921282798833819e-07,
+      "loss": 0.0174,
+      "step": 34710
+    },
+    {
+      "epoch": 49.89,
+      "learning_rate": 8.0466472303207e-07,
+      "loss": 0.0132,
+      "step": 34720
+    },
+    {
+      "epoch": 49.9,
+      "learning_rate": 7.17201166180758e-07,
+      "loss": 0.014,
+      "step": 34730
+    },
+    {
+      "epoch": 49.91,
+      "learning_rate": 6.297376093294461e-07,
+      "loss": 0.0153,
+      "step": 34740
+    },
+    {
+      "epoch": 49.93,
+      "learning_rate": 5.422740524781341e-07,
+      "loss": 0.0108,
+      "step": 34750
+    },
+    {
+      "epoch": 49.94,
+      "learning_rate": 4.548104956268221e-07,
+      "loss": 0.015,
+      "step": 34760
+    },
+    {
+      "epoch": 49.96,
+      "learning_rate": 3.6734693877551015e-07,
+      "loss": 0.0158,
+      "step": 34770
+    },
+    {
+      "epoch": 49.97,
+      "learning_rate": 2.7988338192419826e-07,
+      "loss": 0.0174,
+      "step": 34780
+    },
+    {
+      "epoch": 49.99,
+      "learning_rate": 1.9241982507288626e-07,
+      "loss": 0.0182,
+      "step": 34790
+    },
+    {
+      "epoch": 50.0,
+      "learning_rate": 1.0495626822157434e-07,
+      "loss": 0.0137,
+      "step": 34800
+    },
+    {
+      "epoch": 50.0,
+      "eval_loss": 0.1244998648762703,
+      "eval_runtime": 41.0501,
+      "eval_samples_per_second": 11.133,
+      "eval_steps_per_second": 1.413,
+      "eval_wer": 0.11639813281132359,
+      "step": 34800
     }
   ],
-  "max_steps": 20880,
-  "num_train_epochs": 30,
-  "total_flos": 1.7201005515836912e+20,
+  "max_steps": 34800,
+  "num_train_epochs": 50,
+  "total_flos": 2.8774580960101165e+20,
   "trial_name": null,
   "trial_params": null
 }