diff --git "a/trainer_state.json" "b/trainer_state.json"
--- "a/trainer_state.json"
+++ "b/trainer_state.json"
@@ -1,6 +1,6 @@
 {
-  "best_metric": 0.4694591760635376,
-  "best_model_checkpoint": "ai-light-dance_singing3_ft_wav2vec2-large-xlsr-53-v1-5gram/checkpoint-4464",
+  "best_metric": 0.47944945096969604,
+  "best_model_checkpoint": "ai-light-dance_singing3_ft_wav2vec2-large-xlsr-53-v1-5gram/checkpoint-1512",
   "epoch": 99.99653979238754,
   "global_step": 7200,
   "is_hyper_param_search": false,
@@ -9,5232 +9,5232 @@
   "log_history": [
     {
       "epoch": 0.14,
-      "learning_rate": 5e-06,
-      "loss": 0.3608,
+      "learning_rate": 5.000000000000001e-07,
+      "loss": 0.3086,
       "step": 10
     },
     {
       "epoch": 0.28,
-      "learning_rate": 1e-05,
-      "loss": 0.3819,
+      "learning_rate": 1.0000000000000002e-06,
+      "loss": 0.3275,
       "step": 20
     },
     {
       "epoch": 0.42,
-      "learning_rate": 1.5e-05,
-      "loss": 0.3643,
+      "learning_rate": 1.5e-06,
+      "loss": 0.3051,
       "step": 30
     },
     {
       "epoch": 0.55,
-      "learning_rate": 2e-05,
-      "loss": 0.3776,
+      "learning_rate": 2.0000000000000003e-06,
+      "loss": 0.3072,
       "step": 40
     },
     {
       "epoch": 0.69,
-      "learning_rate": 2.5e-05,
-      "loss": 0.3628,
+      "learning_rate": 2.5e-06,
+      "loss": 0.2957,
       "step": 50
     },
     {
       "epoch": 0.83,
-      "learning_rate": 3e-05,
-      "loss": 0.3659,
+      "learning_rate": 3e-06,
+      "loss": 0.2856,
       "step": 60
     },
     {
       "epoch": 0.97,
-      "learning_rate": 3.5e-05,
-      "loss": 0.3993,
+      "learning_rate": 3.5000000000000004e-06,
+      "loss": 0.314,
       "step": 70
     },
     {
       "epoch": 1.0,
-      "eval_loss": 0.5761275887489319,
-      "eval_runtime": 129.3716,
-      "eval_samples_per_second": 4.112,
-      "eval_steps_per_second": 0.518,
-      "eval_wer": 0.26778941955389857,
+      "eval_loss": 0.4815235435962677,
+      "eval_runtime": 130.4571,
+      "eval_samples_per_second": 4.078,
+      "eval_steps_per_second": 0.514,
+      "eval_wer": 0.20768785755608407,
       "step": 72
     },
     {
       "epoch": 1.11,
-      "learning_rate": 4e-05,
-      "loss": 0.3983,
+      "learning_rate": 4.000000000000001e-06,
+      "loss": 0.2981,
       "step": 80
     },
     {
       "epoch": 1.25,
-      "learning_rate": 4.5e-05,
-      "loss": 0.3997,
+      "learning_rate": 4.5e-06,
+      "loss": 0.2972,
       "step": 90
     },
     {
       "epoch": 1.39,
-      "learning_rate": 5e-05,
-      "loss": 0.3924,
+      "learning_rate": 5e-06,
+      "loss": 0.2892,
       "step": 100
     },
     {
       "epoch": 1.53,
-      "learning_rate": 4.992957746478874e-05,
-      "loss": 0.4145,
+      "learning_rate": 5.500000000000001e-06,
+      "loss": 0.3014,
       "step": 110
     },
     {
       "epoch": 1.66,
-      "learning_rate": 4.9859154929577466e-05,
-      "loss": 0.4261,
+      "learning_rate": 6e-06,
+      "loss": 0.3013,
       "step": 120
     },
     {
       "epoch": 1.8,
-      "learning_rate": 4.97887323943662e-05,
-      "loss": 0.458,
+      "learning_rate": 6.5000000000000004e-06,
+      "loss": 0.315,
       "step": 130
     },
     {
       "epoch": 1.94,
-      "learning_rate": 4.972535211267606e-05,
-      "loss": 0.4252,
+      "learning_rate": 7.000000000000001e-06,
+      "loss": 0.3064,
       "step": 140
     },
     {
       "epoch": 2.0,
-      "eval_loss": 0.5669388175010681,
-      "eval_runtime": 129.6387,
-      "eval_samples_per_second": 4.104,
-      "eval_steps_per_second": 0.517,
-      "eval_wer": 0.2987079771164106,
+      "eval_loss": 0.5082846283912659,
+      "eval_runtime": 130.0863,
+      "eval_samples_per_second": 4.09,
+      "eval_steps_per_second": 0.515,
+      "eval_wer": 0.2101947676287202,
       "step": 144
     },
     {
       "epoch": 2.08,
-      "learning_rate": 4.965492957746479e-05,
-      "loss": 0.4345,
+      "learning_rate": 7.5e-06,
+      "loss": 0.3067,
       "step": 150
     },
     {
       "epoch": 2.22,
-      "learning_rate": 4.9584507042253526e-05,
-      "loss": 0.4184,
+      "learning_rate": 7.95e-06,
+      "loss": 0.304,
       "step": 160
     },
     {
       "epoch": 2.36,
-      "learning_rate": 4.951408450704226e-05,
-      "loss": 0.4452,
+      "learning_rate": 8.45e-06,
+      "loss": 0.3065,
       "step": 170
     },
     {
       "epoch": 2.5,
-      "learning_rate": 4.944366197183099e-05,
-      "loss": 0.4376,
+      "learning_rate": 8.95e-06,
+      "loss": 0.3014,
       "step": 180
     },
     {
       "epoch": 2.64,
-      "learning_rate": 4.9373239436619725e-05,
-      "loss": 0.4491,
+      "learning_rate": 9.450000000000001e-06,
+      "loss": 0.2938,
       "step": 190
     },
     {
       "epoch": 2.78,
-      "learning_rate": 4.930281690140845e-05,
-      "loss": 0.4416,
+      "learning_rate": 9.950000000000001e-06,
+      "loss": 0.3086,
       "step": 200
     },
     {
       "epoch": 2.91,
-      "learning_rate": 4.923239436619719e-05,
-      "loss": 0.4309,
+      "learning_rate": 1.045e-05,
+      "loss": 0.2951,
       "step": 210
     },
     {
       "epoch": 3.0,
-      "eval_loss": 0.553219199180603,
-      "eval_runtime": 128.929,
-      "eval_samples_per_second": 4.126,
-      "eval_steps_per_second": 0.52,
-      "eval_wer": 0.2913800861348589,
+      "eval_loss": 0.5146709680557251,
+      "eval_runtime": 129.1484,
+      "eval_samples_per_second": 4.119,
+      "eval_steps_per_second": 0.519,
+      "eval_wer": 0.2158513852285145,
       "step": 216
     },
     {
       "epoch": 3.06,
-      "learning_rate": 4.916197183098592e-05,
-      "loss": 0.4487,
+      "learning_rate": 1.095e-05,
+      "loss": 0.3079,
       "step": 220
     },
     {
       "epoch": 3.19,
-      "learning_rate": 4.909154929577465e-05,
-      "loss": 0.4235,
+      "learning_rate": 1.145e-05,
+      "loss": 0.2984,
       "step": 230
     },
     {
       "epoch": 3.33,
-      "learning_rate": 4.902112676056339e-05,
-      "loss": 0.4433,
+      "learning_rate": 1.195e-05,
+      "loss": 0.3067,
       "step": 240
     },
     {
       "epoch": 3.47,
-      "learning_rate": 4.8950704225352116e-05,
-      "loss": 0.4383,
+      "learning_rate": 1.2450000000000001e-05,
+      "loss": 0.3033,
       "step": 250
     },
     {
       "epoch": 3.61,
-      "learning_rate": 4.888028169014085e-05,
-      "loss": 0.4362,
+      "learning_rate": 1.2950000000000001e-05,
+      "loss": 0.2912,
       "step": 260
     },
     {
       "epoch": 3.75,
-      "learning_rate": 4.880985915492958e-05,
-      "loss": 0.4127,
+      "learning_rate": 1.3450000000000002e-05,
+      "loss": 0.2915,
       "step": 270
     },
     {
       "epoch": 3.89,
-      "learning_rate": 4.873943661971831e-05,
-      "loss": 0.4151,
+      "learning_rate": 1.3950000000000002e-05,
+      "loss": 0.2835,
       "step": 280
     },
     {
       "epoch": 4.0,
-      "eval_loss": 0.5906756520271301,
-      "eval_runtime": 130.7857,
-      "eval_samples_per_second": 4.068,
-      "eval_steps_per_second": 0.512,
-      "eval_wer": 0.27730282188082533,
+      "eval_loss": 0.515735387802124,
+      "eval_runtime": 128.8527,
+      "eval_samples_per_second": 4.129,
+      "eval_steps_per_second": 0.52,
+      "eval_wer": 0.20910201195603265,
       "step": 288
     },
     {
       "epoch": 4.03,
-      "learning_rate": 4.866901408450704e-05,
-      "loss": 0.4483,
+      "learning_rate": 1.4449999999999999e-05,
+      "loss": 0.3069,
       "step": 290
     },
     {
       "epoch": 4.17,
-      "learning_rate": 4.859859154929577e-05,
-      "loss": 0.4168,
+      "learning_rate": 1.4950000000000001e-05,
+      "loss": 0.2931,
       "step": 300
     },
     {
       "epoch": 4.3,
-      "learning_rate": 4.852816901408451e-05,
-      "loss": 0.4305,
+      "learning_rate": 1.545e-05,
+      "loss": 0.2916,
       "step": 310
     },
     {
       "epoch": 4.44,
-      "learning_rate": 4.845774647887324e-05,
-      "loss": 0.4207,
+      "learning_rate": 1.595e-05,
+      "loss": 0.304,
       "step": 320
     },
     {
       "epoch": 4.58,
-      "learning_rate": 4.838732394366197e-05,
-      "loss": 0.4307,
+      "learning_rate": 1.645e-05,
+      "loss": 0.301,
       "step": 330
     },
     {
       "epoch": 4.72,
-      "learning_rate": 4.8316901408450706e-05,
-      "loss": 0.4278,
+      "learning_rate": 1.6950000000000002e-05,
+      "loss": 0.3098,
       "step": 340
     },
     {
       "epoch": 4.86,
-      "learning_rate": 4.8246478873239435e-05,
-      "loss": 0.4152,
+      "learning_rate": 1.745e-05,
+      "loss": 0.2935,
       "step": 350
     },
     {
       "epoch": 5.0,
-      "learning_rate": 4.817605633802817e-05,
-      "loss": 0.4229,
+      "learning_rate": 1.795e-05,
+      "loss": 0.3011,
       "step": 360
     },
     {
       "epoch": 5.0,
-      "eval_loss": 0.5352545976638794,
-      "eval_runtime": 128.166,
-      "eval_samples_per_second": 4.151,
-      "eval_steps_per_second": 0.523,
-      "eval_wer": 0.2502410490454458,
+      "eval_loss": 0.495644748210907,
+      "eval_runtime": 129.107,
+      "eval_samples_per_second": 4.121,
+      "eval_steps_per_second": 0.519,
+      "eval_wer": 0.215080028283088,
       "step": 360
     },
     {
       "epoch": 5.14,
-      "learning_rate": 4.8105633802816905e-05,
-      "loss": 0.4146,
+      "learning_rate": 1.845e-05,
+      "loss": 0.302,
       "step": 370
     },
     {
       "epoch": 5.28,
-      "learning_rate": 4.8035211267605634e-05,
-      "loss": 0.4246,
+      "learning_rate": 1.895e-05,
+      "loss": 0.3081,
       "step": 380
     },
     {
       "epoch": 5.42,
-      "learning_rate": 4.796478873239437e-05,
-      "loss": 0.4175,
+      "learning_rate": 1.9450000000000002e-05,
+      "loss": 0.3041,
       "step": 390
     },
     {
       "epoch": 5.55,
-      "learning_rate": 4.78943661971831e-05,
-      "loss": 0.4417,
+      "learning_rate": 1.995e-05,
+      "loss": 0.3061,
       "step": 400
     },
     {
       "epoch": 5.69,
-      "learning_rate": 4.782394366197183e-05,
-      "loss": 0.4303,
+      "learning_rate": 2.045e-05,
+      "loss": 0.3081,
       "step": 410
     },
     {
       "epoch": 5.83,
-      "learning_rate": 4.775352112676056e-05,
-      "loss": 0.4343,
+      "learning_rate": 2.09e-05,
+      "loss": 0.2989,
       "step": 420
     },
     {
       "epoch": 5.97,
-      "learning_rate": 4.7683098591549296e-05,
-      "loss": 0.4229,
+      "learning_rate": 2.1400000000000002e-05,
+      "loss": 0.3097,
       "step": 430
     },
     {
       "epoch": 6.0,
-      "eval_loss": 0.5441553592681885,
-      "eval_runtime": 128.9208,
-      "eval_samples_per_second": 4.127,
-      "eval_steps_per_second": 0.52,
-      "eval_wer": 0.2606543678087035,
+      "eval_loss": 0.5190498232841492,
+      "eval_runtime": 129.2076,
+      "eval_samples_per_second": 4.117,
+      "eval_steps_per_second": 0.519,
+      "eval_wer": 0.21533714726489683,
       "step": 432
     },
     {
       "epoch": 6.11,
-      "learning_rate": 4.761267605633803e-05,
-      "loss": 0.4585,
+      "learning_rate": 2.19e-05,
+      "loss": 0.3235,
       "step": 440
     },
     {
       "epoch": 6.25,
-      "learning_rate": 4.754225352112676e-05,
-      "loss": 0.4034,
+      "learning_rate": 2.2400000000000002e-05,
+      "loss": 0.3015,
       "step": 450
     },
     {
       "epoch": 6.39,
-      "learning_rate": 4.7471830985915495e-05,
-      "loss": 0.4069,
+      "learning_rate": 2.29e-05,
+      "loss": 0.2979,
       "step": 460
     },
     {
       "epoch": 6.53,
-      "learning_rate": 4.7401408450704224e-05,
-      "loss": 0.4411,
+      "learning_rate": 2.3400000000000003e-05,
+      "loss": 0.3303,
       "step": 470
     },
     {
       "epoch": 6.66,
-      "learning_rate": 4.733098591549296e-05,
-      "loss": 0.4241,
+      "learning_rate": 2.39e-05,
+      "loss": 0.318,
       "step": 480
     },
     {
       "epoch": 6.8,
-      "learning_rate": 4.7260563380281694e-05,
-      "loss": 0.427,
+      "learning_rate": 2.44e-05,
+      "loss": 0.3111,
       "step": 490
     },
     {
       "epoch": 6.94,
-      "learning_rate": 4.719014084507042e-05,
-      "loss": 0.4106,
+      "learning_rate": 2.4900000000000002e-05,
+      "loss": 0.2998,
       "step": 500
     },
     {
       "epoch": 7.0,
-      "eval_loss": 0.5387436747550964,
-      "eval_runtime": 128.5459,
-      "eval_samples_per_second": 4.139,
-      "eval_steps_per_second": 0.521,
-      "eval_wer": 0.2513338047181333,
+      "eval_loss": 0.5349323153495789,
+      "eval_runtime": 129.2108,
+      "eval_samples_per_second": 4.117,
+      "eval_steps_per_second": 0.519,
+      "eval_wer": 0.22015812817381245,
       "step": 504
     },
     {
       "epoch": 7.08,
-      "learning_rate": 4.711971830985916e-05,
-      "loss": 0.4533,
+      "learning_rate": 2.4970149253731346e-05,
+      "loss": 0.3216,
       "step": 510
     },
     {
       "epoch": 7.22,
-      "learning_rate": 4.704929577464789e-05,
-      "loss": 0.4188,
+      "learning_rate": 2.493283582089552e-05,
+      "loss": 0.3147,
       "step": 520
     },
     {
       "epoch": 7.36,
-      "learning_rate": 4.697887323943662e-05,
-      "loss": 0.4313,
+      "learning_rate": 2.4895522388059704e-05,
+      "loss": 0.3128,
       "step": 530
     },
     {
       "epoch": 7.5,
-      "learning_rate": 4.690845070422536e-05,
-      "loss": 0.4281,
+      "learning_rate": 2.4858208955223883e-05,
+      "loss": 0.3079,
       "step": 540
     },
     {
       "epoch": 7.64,
-      "learning_rate": 4.6838028169014086e-05,
-      "loss": 0.4031,
+      "learning_rate": 2.482089552238806e-05,
+      "loss": 0.2967,
       "step": 550
     },
     {
       "epoch": 7.78,
-      "learning_rate": 4.676760563380282e-05,
-      "loss": 0.4223,
+      "learning_rate": 2.478358208955224e-05,
+      "loss": 0.3098,
       "step": 560
     },
     {
       "epoch": 7.91,
-      "learning_rate": 4.669718309859155e-05,
-      "loss": 0.4161,
+      "learning_rate": 2.474626865671642e-05,
+      "loss": 0.3086,
       "step": 570
     },
     {
       "epoch": 8.0,
-      "eval_loss": 0.5407251715660095,
-      "eval_runtime": 127.4676,
-      "eval_samples_per_second": 4.174,
-      "eval_steps_per_second": 0.526,
-      "eval_wer": 0.2591759336633027,
+      "eval_loss": 0.5055466890335083,
+      "eval_runtime": 128.9787,
+      "eval_samples_per_second": 4.125,
+      "eval_steps_per_second": 0.519,
+      "eval_wer": 0.21450151057401812,
       "step": 576
     },
     {
       "epoch": 8.06,
-      "learning_rate": 4.6626760563380285e-05,
-      "loss": 0.4399,
+      "learning_rate": 2.4708955223880597e-05,
+      "loss": 0.3243,
       "step": 580
     },
     {
       "epoch": 8.19,
-      "learning_rate": 4.655633802816901e-05,
-      "loss": 0.4198,
+      "learning_rate": 2.467164179104478e-05,
+      "loss": 0.3088,
       "step": 590
     },
     {
       "epoch": 8.33,
-      "learning_rate": 4.648591549295775e-05,
-      "loss": 0.4121,
+      "learning_rate": 2.4634328358208955e-05,
+      "loss": 0.3071,
       "step": 600
     },
     {
       "epoch": 8.47,
-      "learning_rate": 4.6415492957746484e-05,
-      "loss": 0.4238,
+      "learning_rate": 2.4597014925373134e-05,
+      "loss": 0.3094,
       "step": 610
     },
     {
       "epoch": 8.61,
-      "learning_rate": 4.634507042253521e-05,
-      "loss": 0.4315,
+      "learning_rate": 2.4559701492537317e-05,
+      "loss": 0.3049,
       "step": 620
     },
     {
       "epoch": 8.75,
-      "learning_rate": 4.627464788732395e-05,
-      "loss": 0.4255,
+      "learning_rate": 2.4522388059701492e-05,
+      "loss": 0.3061,
       "step": 630
     },
     {
       "epoch": 8.89,
-      "learning_rate": 4.6204225352112676e-05,
-      "loss": 0.4072,
+      "learning_rate": 2.448507462686567e-05,
+      "loss": 0.2964,
       "step": 640
     },
     {
       "epoch": 9.0,
-      "eval_loss": 0.5012030601501465,
-      "eval_runtime": 128.1858,
-      "eval_samples_per_second": 4.15,
-      "eval_steps_per_second": 0.523,
-      "eval_wer": 0.24934113260911486,
+      "eval_loss": 0.5091516971588135,
+      "eval_runtime": 129.4051,
+      "eval_samples_per_second": 4.111,
+      "eval_steps_per_second": 0.518,
+      "eval_wer": 0.2183582953011506,
       "step": 648
     },
     {
       "epoch": 9.03,
-      "learning_rate": 4.613380281690141e-05,
-      "loss": 0.4213,
+      "learning_rate": 2.4447761194029854e-05,
+      "loss": 0.3212,
       "step": 650
     },
     {
       "epoch": 9.17,
-      "learning_rate": 4.6063380281690146e-05,
-      "loss": 0.4047,
+      "learning_rate": 2.441044776119403e-05,
+      "loss": 0.302,
       "step": 660
     },
     {
       "epoch": 9.3,
-      "learning_rate": 4.5992957746478875e-05,
-      "loss": 0.437,
+      "learning_rate": 2.437313432835821e-05,
+      "loss": 0.3103,
       "step": 670
     },
     {
       "epoch": 9.44,
-      "learning_rate": 4.592253521126761e-05,
-      "loss": 0.4155,
+      "learning_rate": 2.433582089552239e-05,
+      "loss": 0.3102,
       "step": 680
     },
     {
       "epoch": 9.58,
-      "learning_rate": 4.585211267605634e-05,
-      "loss": 0.4113,
+      "learning_rate": 2.4298507462686567e-05,
+      "loss": 0.3045,
       "step": 690
     },
     {
       "epoch": 9.72,
-      "learning_rate": 4.5781690140845074e-05,
-      "loss": 0.4107,
+      "learning_rate": 2.4261194029850746e-05,
+      "loss": 0.3162,
       "step": 700
     },
     {
       "epoch": 9.86,
-      "learning_rate": 4.571126760563381e-05,
-      "loss": 0.4069,
+      "learning_rate": 2.4223880597014926e-05,
+      "loss": 0.3186,
       "step": 710
     },
     {
       "epoch": 10.0,
-      "learning_rate": 4.564084507042254e-05,
-      "loss": 0.4301,
+      "learning_rate": 2.4186567164179105e-05,
+      "loss": 0.3288,
       "step": 720
     },
     {
       "epoch": 10.0,
-      "eval_loss": 0.5448527932167053,
-      "eval_runtime": 128.8981,
-      "eval_samples_per_second": 4.127,
-      "eval_steps_per_second": 0.52,
-      "eval_wer": 0.2541621135180305,
+      "eval_loss": 0.5076370239257812,
+      "eval_runtime": 129.7675,
+      "eval_samples_per_second": 4.1,
+      "eval_steps_per_second": 0.516,
+      "eval_wer": 0.2190653725011249,
       "step": 720
     },
     {
       "epoch": 10.14,
-      "learning_rate": 4.557042253521127e-05,
-      "loss": 0.402,
+      "learning_rate": 2.4149253731343284e-05,
+      "loss": 0.3037,
       "step": 730
     },
     {
       "epoch": 10.28,
-      "learning_rate": 4.55e-05,
-      "loss": 0.4327,
+      "learning_rate": 2.4111940298507463e-05,
+      "loss": 0.3256,
       "step": 740
     },
     {
       "epoch": 10.42,
-      "learning_rate": 4.542957746478874e-05,
-      "loss": 0.4174,
+      "learning_rate": 2.4074626865671642e-05,
+      "loss": 0.3209,
       "step": 750
     },
     {
       "epoch": 10.55,
-      "learning_rate": 4.535915492957747e-05,
-      "loss": 0.4304,
+      "learning_rate": 2.403731343283582e-05,
+      "loss": 0.3259,
       "step": 760
     },
     {
       "epoch": 10.69,
-      "learning_rate": 4.52887323943662e-05,
-      "loss": 0.4234,
+      "learning_rate": 2.4e-05,
+      "loss": 0.3081,
       "step": 770
     },
     {
       "epoch": 10.83,
-      "learning_rate": 4.5218309859154936e-05,
-      "loss": 0.4194,
+      "learning_rate": 2.396268656716418e-05,
+      "loss": 0.3153,
       "step": 780
     },
     {
       "epoch": 10.97,
-      "learning_rate": 4.5147887323943664e-05,
-      "loss": 0.419,
+      "learning_rate": 2.392537313432836e-05,
+      "loss": 0.3019,
       "step": 790
     },
     {
       "epoch": 11.0,
-      "eval_loss": 0.5024985671043396,
-      "eval_runtime": 127.8059,
-      "eval_samples_per_second": 4.163,
-      "eval_steps_per_second": 0.524,
-      "eval_wer": 0.25576910715433565,
+      "eval_loss": 0.5125810503959656,
+      "eval_runtime": 128.8657,
+      "eval_samples_per_second": 4.128,
+      "eval_steps_per_second": 0.52,
+      "eval_wer": 0.20916629170148487,
       "step": 792
     },
     {
       "epoch": 11.11,
-      "learning_rate": 4.50774647887324e-05,
-      "loss": 0.4251,
+      "learning_rate": 2.3888059701492538e-05,
+      "loss": 0.3151,
       "step": 800
     },
     {
       "epoch": 11.25,
-      "learning_rate": 4.500704225352113e-05,
-      "loss": 0.4161,
+      "learning_rate": 2.3850746268656717e-05,
+      "loss": 0.3154,
       "step": 810
     },
     {
       "epoch": 11.39,
-      "learning_rate": 4.493661971830986e-05,
-      "loss": 0.4053,
+      "learning_rate": 2.3813432835820896e-05,
+      "loss": 0.2949,
       "step": 820
     },
     {
       "epoch": 11.53,
-      "learning_rate": 4.48661971830986e-05,
-      "loss": 0.444,
+      "learning_rate": 2.3776119402985076e-05,
+      "loss": 0.3214,
       "step": 830
     },
     {
       "epoch": 11.66,
-      "learning_rate": 4.479577464788733e-05,
-      "loss": 0.3947,
+      "learning_rate": 2.3738805970149255e-05,
+      "loss": 0.2946,
       "step": 840
     },
     {
       "epoch": 11.8,
-      "learning_rate": 4.472535211267606e-05,
-      "loss": 0.4309,
+      "learning_rate": 2.3701492537313434e-05,
+      "loss": 0.3154,
       "step": 850
     },
     {
       "epoch": 11.94,
-      "learning_rate": 4.465492957746479e-05,
-      "loss": 0.4336,
+      "learning_rate": 2.3664179104477613e-05,
+      "loss": 0.3047,
       "step": 860
     },
     {
       "epoch": 12.0,
-      "eval_loss": 0.5266738533973694,
-      "eval_runtime": 128.1272,
-      "eval_samples_per_second": 4.152,
-      "eval_steps_per_second": 0.523,
-      "eval_wer": 0.24509866940926914,
+      "eval_loss": 0.5249019861221313,
+      "eval_runtime": 129.2154,
+      "eval_samples_per_second": 4.117,
+      "eval_steps_per_second": 0.519,
+      "eval_wer": 0.2117374815195732,
       "step": 864
     },
     {
       "epoch": 12.08,
-      "learning_rate": 4.4584507042253526e-05,
-      "loss": 0.4158,
+      "learning_rate": 2.3626865671641792e-05,
+      "loss": 0.2959,
       "step": 870
     },
     {
       "epoch": 12.22,
-      "learning_rate": 4.4514084507042254e-05,
-      "loss": 0.3992,
+      "learning_rate": 2.358955223880597e-05,
+      "loss": 0.295,
       "step": 880
     },
     {
       "epoch": 12.36,
-      "learning_rate": 4.444366197183098e-05,
-      "loss": 0.4215,
+      "learning_rate": 2.355223880597015e-05,
+      "loss": 0.3122,
       "step": 890
     },
     {
       "epoch": 12.5,
-      "learning_rate": 4.437323943661972e-05,
-      "loss": 0.4284,
+      "learning_rate": 2.351492537313433e-05,
+      "loss": 0.3028,
       "step": 900
     },
     {
       "epoch": 12.64,
-      "learning_rate": 4.4302816901408453e-05,
-      "loss": 0.3886,
+      "learning_rate": 2.347761194029851e-05,
+      "loss": 0.2777,
       "step": 910
     },
     {
       "epoch": 12.78,
-      "learning_rate": 4.423239436619718e-05,
-      "loss": 0.4216,
+      "learning_rate": 2.3440298507462688e-05,
+      "loss": 0.3025,
       "step": 920
     },
     {
       "epoch": 12.91,
-      "learning_rate": 4.416197183098592e-05,
-      "loss": 0.4188,
+      "learning_rate": 2.3402985074626867e-05,
+      "loss": 0.2998,
       "step": 930
     },
     {
       "epoch": 13.0,
-      "eval_loss": 0.5068269371986389,
-      "eval_runtime": 127.9442,
-      "eval_samples_per_second": 4.158,
-      "eval_steps_per_second": 0.524,
-      "eval_wer": 0.2582117374815196,
+      "eval_loss": 0.531330406665802,
+      "eval_runtime": 129.1703,
+      "eval_samples_per_second": 4.119,
+      "eval_steps_per_second": 0.519,
+      "eval_wer": 0.21405155235585266,
       "step": 936
     },
     {
       "epoch": 13.06,
-      "learning_rate": 4.4091549295774646e-05,
-      "loss": 0.4384,
+      "learning_rate": 2.3365671641791046e-05,
+      "loss": 0.3192,
       "step": 940
     },
     {
       "epoch": 13.19,
-      "learning_rate": 4.402112676056338e-05,
-      "loss": 0.4001,
+      "learning_rate": 2.3328358208955225e-05,
+      "loss": 0.2997,
       "step": 950
     },
     {
       "epoch": 13.33,
-      "learning_rate": 4.395070422535211e-05,
-      "loss": 0.4263,
+      "learning_rate": 2.32910447761194e-05,
+      "loss": 0.3163,
       "step": 960
     },
     {
       "epoch": 13.47,
-      "learning_rate": 4.3880281690140845e-05,
-      "loss": 0.4052,
+      "learning_rate": 2.3253731343283584e-05,
+      "loss": 0.3046,
       "step": 970
     },
     {
       "epoch": 13.61,
-      "learning_rate": 4.380985915492958e-05,
-      "loss": 0.3927,
+      "learning_rate": 2.3216417910447763e-05,
+      "loss": 0.2924,
       "step": 980
     },
     {
       "epoch": 13.75,
-      "learning_rate": 4.373943661971831e-05,
-      "loss": 0.428,
+      "learning_rate": 2.317910447761194e-05,
+      "loss": 0.3161,
       "step": 990
     },
     {
       "epoch": 13.89,
-      "learning_rate": 4.3669014084507044e-05,
-      "loss": 0.4005,
+      "learning_rate": 2.314179104477612e-05,
+      "loss": 0.3043,
       "step": 1000
     },
     {
       "epoch": 14.0,
-      "eval_loss": 0.5392433404922485,
-      "eval_runtime": 128.4034,
-      "eval_samples_per_second": 4.143,
-      "eval_steps_per_second": 0.522,
-      "eval_wer": 0.2528765186089863,
+      "eval_loss": 0.5343348979949951,
+      "eval_runtime": 129.5843,
+      "eval_samples_per_second": 4.105,
+      "eval_steps_per_second": 0.517,
+      "eval_wer": 0.21861541428295944,
       "step": 1008
     },
     {
       "epoch": 14.03,
-      "learning_rate": 4.359859154929577e-05,
-      "loss": 0.4283,
+      "learning_rate": 2.31044776119403e-05,
+      "loss": 0.3099,
       "step": 1010
     },
     {
       "epoch": 14.17,
-      "learning_rate": 4.352816901408451e-05,
-      "loss": 0.4031,
+      "learning_rate": 2.306716417910448e-05,
+      "loss": 0.2953,
       "step": 1020
     },
     {
       "epoch": 14.3,
-      "learning_rate": 4.345774647887324e-05,
-      "loss": 0.4168,
+      "learning_rate": 2.302985074626866e-05,
+      "loss": 0.3134,
       "step": 1030
     },
     {
       "epoch": 14.44,
-      "learning_rate": 4.338732394366197e-05,
-      "loss": 0.3987,
+      "learning_rate": 2.2992537313432838e-05,
+      "loss": 0.2917,
       "step": 1040
     },
     {
       "epoch": 14.58,
-      "learning_rate": 4.3316901408450707e-05,
-      "loss": 0.4338,
+      "learning_rate": 2.2955223880597017e-05,
+      "loss": 0.3108,
       "step": 1050
     },
     {
       "epoch": 14.72,
-      "learning_rate": 4.3246478873239435e-05,
-      "loss": 0.4093,
+      "learning_rate": 2.2917910447761196e-05,
+      "loss": 0.3085,
       "step": 1060
     },
     {
       "epoch": 14.86,
-      "learning_rate": 4.317605633802817e-05,
-      "loss": 0.4004,
+      "learning_rate": 2.2880597014925372e-05,
+      "loss": 0.2921,
       "step": 1070
     },
     {
       "epoch": 15.0,
-      "learning_rate": 4.3105633802816906e-05,
-      "loss": 0.4213,
+      "learning_rate": 2.2843283582089555e-05,
+      "loss": 0.3147,
       "step": 1080
     },
     {
       "epoch": 15.0,
-      "eval_loss": 0.5082538723945618,
-      "eval_runtime": 127.3218,
-      "eval_samples_per_second": 4.178,
-      "eval_steps_per_second": 0.526,
-      "eval_wer": 0.24921257311821046,
+      "eval_loss": 0.506893515586853,
+      "eval_runtime": 129.0401,
+      "eval_samples_per_second": 4.123,
+      "eval_steps_per_second": 0.519,
+      "eval_wer": 0.21713698013755867,
       "step": 1080
     },
     {
       "epoch": 15.14,
-      "learning_rate": 4.3035211267605634e-05,
-      "loss": 0.4228,
+      "learning_rate": 2.2805970149253734e-05,
+      "loss": 0.307,
       "step": 1090
     },
     {
       "epoch": 15.28,
-      "learning_rate": 4.296478873239437e-05,
-      "loss": 0.4163,
+      "learning_rate": 2.276865671641791e-05,
+      "loss": 0.3152,
       "step": 1100
     },
     {
       "epoch": 15.42,
-      "learning_rate": 4.28943661971831e-05,
-      "loss": 0.3912,
+      "learning_rate": 2.2731343283582092e-05,
+      "loss": 0.3002,
       "step": 1110
     },
     {
       "epoch": 15.55,
-      "learning_rate": 4.282394366197183e-05,
-      "loss": 0.4152,
+      "learning_rate": 2.269402985074627e-05,
+      "loss": 0.2992,
       "step": 1120
     },
     {
       "epoch": 15.69,
-      "learning_rate": 4.275352112676057e-05,
-      "loss": 0.3893,
+      "learning_rate": 2.2656716417910447e-05,
+      "loss": 0.2897,
       "step": 1130
     },
     {
       "epoch": 15.83,
-      "learning_rate": 4.26830985915493e-05,
-      "loss": 0.4163,
+      "learning_rate": 2.261940298507463e-05,
+      "loss": 0.3069,
       "step": 1140
     },
     {
       "epoch": 15.97,
-      "learning_rate": 4.261267605633803e-05,
-      "loss": 0.4123,
+      "learning_rate": 2.258208955223881e-05,
+      "loss": 0.3049,
       "step": 1150
     },
     {
       "epoch": 16.0,
-      "eval_loss": 0.4909058213233948,
-      "eval_runtime": 127.0485,
-      "eval_samples_per_second": 4.187,
-      "eval_steps_per_second": 0.527,
-      "eval_wer": 0.24811981744552292,
+      "eval_loss": 0.5032855272293091,
+      "eval_runtime": 129.2941,
+      "eval_samples_per_second": 4.115,
+      "eval_steps_per_second": 0.518,
+      "eval_wer": 0.20736645882882304,
       "step": 1152
     },
     {
       "epoch": 16.11,
-      "learning_rate": 4.254225352112676e-05,
-      "loss": 0.4027,
+      "learning_rate": 2.2544776119402984e-05,
+      "loss": 0.3014,
       "step": 1160
     },
     {
       "epoch": 16.25,
-      "learning_rate": 4.2471830985915496e-05,
-      "loss": 0.4021,
+      "learning_rate": 2.2507462686567167e-05,
+      "loss": 0.3078,
       "step": 1170
     },
     {
       "epoch": 16.39,
-      "learning_rate": 4.2401408450704224e-05,
-      "loss": 0.3876,
+      "learning_rate": 2.2470149253731346e-05,
+      "loss": 0.2942,
       "step": 1180
     },
     {
       "epoch": 16.53,
-      "learning_rate": 4.233098591549296e-05,
-      "loss": 0.4161,
+      "learning_rate": 2.2432835820895522e-05,
+      "loss": 0.3168,
       "step": 1190
     },
     {
       "epoch": 16.66,
-      "learning_rate": 4.2260563380281695e-05,
-      "loss": 0.3975,
+      "learning_rate": 2.2395522388059704e-05,
+      "loss": 0.3124,
       "step": 1200
     },
     {
       "epoch": 16.8,
-      "learning_rate": 4.219014084507042e-05,
-      "loss": 0.3785,
+      "learning_rate": 2.235820895522388e-05,
+      "loss": 0.2951,
       "step": 1210
     },
     {
       "epoch": 16.94,
-      "learning_rate": 4.211971830985916e-05,
-      "loss": 0.3854,
+      "learning_rate": 2.232089552238806e-05,
+      "loss": 0.298,
       "step": 1220
     },
     {
       "epoch": 17.0,
-      "eval_loss": 0.5269584655761719,
-      "eval_runtime": 128.6883,
-      "eval_samples_per_second": 4.134,
-      "eval_steps_per_second": 0.521,
-      "eval_wer": 0.2593044931542071,
+      "eval_loss": 0.5108718276023865,
+      "eval_runtime": 129.096,
+      "eval_samples_per_second": 4.121,
+      "eval_steps_per_second": 0.519,
+      "eval_wer": 0.20498810824709135,
       "step": 1224
     },
     {
       "epoch": 17.08,
-      "learning_rate": 4.204929577464789e-05,
-      "loss": 0.4326,
+      "learning_rate": 2.2283582089552242e-05,
+      "loss": 0.3255,
       "step": 1230
     },
     {
       "epoch": 17.22,
-      "learning_rate": 4.197887323943662e-05,
-      "loss": 0.3852,
+      "learning_rate": 2.2246268656716418e-05,
+      "loss": 0.2867,
       "step": 1240
     },
     {
       "epoch": 17.36,
-      "learning_rate": 4.190845070422536e-05,
-      "loss": 0.3947,
+      "learning_rate": 2.2208955223880597e-05,
+      "loss": 0.2983,
       "step": 1250
     },
     {
       "epoch": 17.5,
-      "learning_rate": 4.1838028169014086e-05,
-      "loss": 0.4118,
+      "learning_rate": 2.217164179104478e-05,
+      "loss": 0.3105,
       "step": 1260
     },
     {
       "epoch": 17.64,
-      "learning_rate": 4.176760563380282e-05,
-      "loss": 0.3857,
+      "learning_rate": 2.2134328358208955e-05,
+      "loss": 0.2818,
       "step": 1270
     },
     {
       "epoch": 17.78,
-      "learning_rate": 4.169718309859155e-05,
-      "loss": 0.4201,
+      "learning_rate": 2.2097014925373134e-05,
+      "loss": 0.3107,
       "step": 1280
     },
     {
       "epoch": 17.91,
-      "learning_rate": 4.1626760563380285e-05,
-      "loss": 0.3887,
+      "learning_rate": 2.2059701492537317e-05,
+      "loss": 0.2948,
       "step": 1290
     },
     {
       "epoch": 18.0,
-      "eval_loss": 0.50589519739151,
-      "eval_runtime": 126.713,
-      "eval_samples_per_second": 4.198,
-      "eval_steps_per_second": 0.529,
-      "eval_wer": 0.250305328790898,
+      "eval_loss": 0.5252682566642761,
+      "eval_runtime": 129.1692,
+      "eval_samples_per_second": 4.119,
+      "eval_steps_per_second": 0.519,
+      "eval_wer": 0.20993764864691136,
       "step": 1296
     },
     {
       "epoch": 18.06,
-      "learning_rate": 4.155633802816902e-05,
-      "loss": 0.4153,
+      "learning_rate": 2.2022388059701493e-05,
+      "loss": 0.316,
       "step": 1300
     },
     {
       "epoch": 18.19,
-      "learning_rate": 4.148591549295775e-05,
-      "loss": 0.4241,
+      "learning_rate": 2.1985074626865672e-05,
+      "loss": 0.3066,
       "step": 1310
     },
     {
       "epoch": 18.33,
-      "learning_rate": 4.1415492957746484e-05,
-      "loss": 0.4298,
+      "learning_rate": 2.194776119402985e-05,
+      "loss": 0.3031,
       "step": 1320
     },
     {
       "epoch": 18.47,
-      "learning_rate": 4.134507042253521e-05,
-      "loss": 0.4065,
+      "learning_rate": 2.191044776119403e-05,
+      "loss": 0.3037,
       "step": 1330
     },
     {
       "epoch": 18.61,
-      "learning_rate": 4.127464788732395e-05,
-      "loss": 0.4064,
+      "learning_rate": 2.187313432835821e-05,
+      "loss": 0.2986,
       "step": 1340
     },
     {
       "epoch": 18.75,
-      "learning_rate": 4.120422535211268e-05,
-      "loss": 0.398,
+      "learning_rate": 2.183582089552239e-05,
+      "loss": 0.2955,
       "step": 1350
     },
     {
       "epoch": 18.89,
-      "learning_rate": 4.113380281690141e-05,
-      "loss": 0.3934,
+      "learning_rate": 2.1798507462686568e-05,
+      "loss": 0.2986,
       "step": 1360
     },
     {
       "epoch": 19.0,
-      "eval_loss": 0.4928908944129944,
-      "eval_runtime": 126.8145,
-      "eval_samples_per_second": 4.195,
-      "eval_steps_per_second": 0.528,
-      "eval_wer": 0.2352638683550813,
+      "eval_loss": 0.4976494014263153,
+      "eval_runtime": 129.378,
+      "eval_samples_per_second": 4.112,
+      "eval_steps_per_second": 0.518,
+      "eval_wer": 0.20209551970174197,
       "step": 1368
     },
     {
       "epoch": 19.03,
-      "learning_rate": 4.106338028169015e-05,
-      "loss": 0.4086,
+      "learning_rate": 2.1761194029850747e-05,
+      "loss": 0.3021,
       "step": 1370
     },
     {
       "epoch": 19.17,
-      "learning_rate": 4.0992957746478875e-05,
-      "loss": 0.3826,
+      "learning_rate": 2.1723880597014926e-05,
+      "loss": 0.2887,
       "step": 1380
     },
     {
       "epoch": 19.3,
-      "learning_rate": 4.092253521126761e-05,
-      "loss": 0.4094,
+      "learning_rate": 2.1686567164179105e-05,
+      "loss": 0.3059,
       "step": 1390
     },
     {
       "epoch": 19.44,
-      "learning_rate": 4.085211267605634e-05,
-      "loss": 0.3973,
+      "learning_rate": 2.1649253731343284e-05,
+      "loss": 0.2945,
       "step": 1400
     },
     {
       "epoch": 19.58,
-      "learning_rate": 4.0781690140845074e-05,
-      "loss": 0.4066,
+      "learning_rate": 2.1611940298507463e-05,
+      "loss": 0.3006,
       "step": 1410
     },
     {
       "epoch": 19.72,
-      "learning_rate": 4.071126760563381e-05,
-      "loss": 0.3961,
+      "learning_rate": 2.1574626865671643e-05,
+      "loss": 0.3033,
       "step": 1420
     },
     {
       "epoch": 19.86,
-      "learning_rate": 4.064084507042254e-05,
-      "loss": 0.402,
+      "learning_rate": 2.153731343283582e-05,
+      "loss": 0.3002,
       "step": 1430
     },
     {
       "epoch": 20.0,
-      "learning_rate": 4.057042253521127e-05,
-      "loss": 0.4005,
+      "learning_rate": 2.15e-05,
+      "loss": 0.2958,
       "step": 1440
     },
     {
       "epoch": 20.0,
-      "eval_loss": 0.5170095562934875,
-      "eval_runtime": 127.4875,
-      "eval_samples_per_second": 4.173,
-      "eval_steps_per_second": 0.526,
-      "eval_wer": 0.24439159220929485,
+      "eval_loss": 0.5233781933784485,
+      "eval_runtime": 129.422,
+      "eval_samples_per_second": 4.111,
+      "eval_steps_per_second": 0.518,
+      "eval_wer": 0.20935913093784148,
       "step": 1440
     },
     {
       "epoch": 20.14,
-      "learning_rate": 4.05e-05,
-      "loss": 0.3946,
+      "learning_rate": 2.146268656716418e-05,
+      "loss": 0.2982,
       "step": 1450
     },
     {
       "epoch": 20.28,
-      "learning_rate": 4.042957746478874e-05,
-      "loss": 0.3916,
+      "learning_rate": 2.142537313432836e-05,
+      "loss": 0.2861,
       "step": 1460
     },
     {
       "epoch": 20.42,
-      "learning_rate": 4.0359154929577466e-05,
-      "loss": 0.382,
+      "learning_rate": 2.1388059701492538e-05,
+      "loss": 0.293,
       "step": 1470
     },
     {
       "epoch": 20.55,
-      "learning_rate": 4.0288732394366194e-05,
-      "loss": 0.4081,
+      "learning_rate": 2.1350746268656717e-05,
+      "loss": 0.304,
       "step": 1480
     },
     {
       "epoch": 20.69,
-      "learning_rate": 4.021830985915493e-05,
-      "loss": 0.3917,
+      "learning_rate": 2.1313432835820897e-05,
+      "loss": 0.2926,
       "step": 1490
     },
     {
       "epoch": 20.83,
-      "learning_rate": 4.0147887323943665e-05,
-      "loss": 0.4083,
+      "learning_rate": 2.1276119402985076e-05,
+      "loss": 0.3092,
       "step": 1500
     },
     {
       "epoch": 20.97,
-      "learning_rate": 4.007746478873239e-05,
-      "loss": 0.4018,
+      "learning_rate": 2.1238805970149255e-05,
+      "loss": 0.2931,
       "step": 1510
     },
     {
       "epoch": 21.0,
-      "eval_loss": 0.4931071102619171,
-      "eval_runtime": 127.824,
-      "eval_samples_per_second": 4.162,
-      "eval_steps_per_second": 0.524,
-      "eval_wer": 0.24059908722761458,
+      "eval_loss": 0.47944945096969604,
+      "eval_runtime": 130.2981,
+      "eval_samples_per_second": 4.083,
+      "eval_steps_per_second": 0.514,
+      "eval_wer": 0.20595230442887447,
       "step": 1512
     },
     {
       "epoch": 21.11,
-      "learning_rate": 4.000704225352113e-05,
-      "loss": 0.3987,
+      "learning_rate": 2.1201492537313434e-05,
+      "loss": 0.2956,
       "step": 1520
     },
     {
       "epoch": 21.25,
-      "learning_rate": 3.993661971830986e-05,
-      "loss": 0.4043,
+      "learning_rate": 2.1164179104477613e-05,
+      "loss": 0.3007,
       "step": 1530
     },
     {
       "epoch": 21.39,
-      "learning_rate": 3.986619718309859e-05,
-      "loss": 0.398,
+      "learning_rate": 2.1126865671641792e-05,
+      "loss": 0.3045,
       "step": 1540
     },
     {
       "epoch": 21.53,
-      "learning_rate": 3.979577464788732e-05,
-      "loss": 0.406,
+      "learning_rate": 2.108955223880597e-05,
+      "loss": 0.3002,
       "step": 1550
     },
     {
       "epoch": 21.66,
-      "learning_rate": 3.9725352112676056e-05,
-      "loss": 0.3786,
+      "learning_rate": 2.105223880597015e-05,
+      "loss": 0.2848,
       "step": 1560
     },
     {
       "epoch": 21.8,
-      "learning_rate": 3.965492957746479e-05,
-      "loss": 0.3908,
+      "learning_rate": 2.1014925373134326e-05,
+      "loss": 0.286,
       "step": 1570
     },
     {
       "epoch": 21.94,
-      "learning_rate": 3.958450704225352e-05,
-      "loss": 0.3981,
+      "learning_rate": 2.097761194029851e-05,
+      "loss": 0.305,
       "step": 1580
     },
     {
       "epoch": 22.0,
-      "eval_loss": 0.4988018870353699,
-      "eval_runtime": 129.0602,
-      "eval_samples_per_second": 4.122,
+      "eval_loss": 0.4916737377643585,
+      "eval_runtime": 129.0137,
+      "eval_samples_per_second": 4.124,
       "eval_steps_per_second": 0.519,
-      "eval_wer": 0.23449251140965482,
+      "eval_wer": 0.20492382850163912,
       "step": 1584
     },
     {
       "epoch": 22.08,
-      "learning_rate": 3.9514084507042255e-05,
-      "loss": 0.4121,
+      "learning_rate": 2.0940298507462688e-05,
+      "loss": 0.3092,
       "step": 1590
     },
     {
       "epoch": 22.22,
-      "learning_rate": 3.9443661971830983e-05,
-      "loss": 0.3837,
+      "learning_rate": 2.0902985074626867e-05,
+      "loss": 0.2979,
       "step": 1600
     },
     {
       "epoch": 22.36,
-      "learning_rate": 3.937323943661972e-05,
-      "loss": 0.3951,
+      "learning_rate": 2.0865671641791047e-05,
+      "loss": 0.2921,
       "step": 1610
     },
     {
       "epoch": 22.5,
-      "learning_rate": 3.9302816901408454e-05,
-      "loss": 0.4119,
+      "learning_rate": 2.0828358208955226e-05,
+      "loss": 0.3052,
       "step": 1620
     },
     {
       "epoch": 22.64,
-      "learning_rate": 3.923239436619718e-05,
-      "loss": 0.3912,
+      "learning_rate": 2.0791044776119405e-05,
+      "loss": 0.2977,
       "step": 1630
     },
     {
       "epoch": 22.78,
-      "learning_rate": 3.916197183098592e-05,
-      "loss": 0.3787,
+      "learning_rate": 2.0753731343283584e-05,
+      "loss": 0.2938,
       "step": 1640
     },
     {
       "epoch": 22.91,
-      "learning_rate": 3.9091549295774646e-05,
-      "loss": 0.3645,
+      "learning_rate": 2.0716417910447763e-05,
+      "loss": 0.2847,
       "step": 1650
     },
     {
       "epoch": 23.0,
-      "eval_loss": 0.5220637321472168,
-      "eval_runtime": 127.4377,
-      "eval_samples_per_second": 4.175,
-      "eval_steps_per_second": 0.526,
-      "eval_wer": 0.2494696921000193,
+      "eval_loss": 0.5089883804321289,
+      "eval_runtime": 129.655,
+      "eval_samples_per_second": 4.103,
+      "eval_steps_per_second": 0.517,
+      "eval_wer": 0.21000192839236356,
       "step": 1656
     },
     {
       "epoch": 23.06,
-      "learning_rate": 3.902112676056338e-05,
-      "loss": 0.403,
+      "learning_rate": 2.0679104477611942e-05,
+      "loss": 0.2993,
       "step": 1660
     },
     {
       "epoch": 23.19,
-      "learning_rate": 3.895070422535212e-05,
-      "loss": 0.3992,
+      "learning_rate": 2.064179104477612e-05,
+      "loss": 0.3081,
       "step": 1670
     },
     {
       "epoch": 23.33,
-      "learning_rate": 3.8880281690140845e-05,
-      "loss": 0.3958,
+      "learning_rate": 2.0604477611940297e-05,
+      "loss": 0.2998,
       "step": 1680
     },
     {
       "epoch": 23.47,
-      "learning_rate": 3.880985915492958e-05,
-      "loss": 0.4091,
+      "learning_rate": 2.056716417910448e-05,
+      "loss": 0.3125,
       "step": 1690
     },
     {
       "epoch": 23.61,
-      "learning_rate": 3.873943661971831e-05,
-      "loss": 0.4006,
+      "learning_rate": 2.052985074626866e-05,
+      "loss": 0.3135,
       "step": 1700
     },
     {
       "epoch": 23.75,
-      "learning_rate": 3.8669014084507044e-05,
-      "loss": 0.3856,
+      "learning_rate": 2.0492537313432835e-05,
+      "loss": 0.3039,
       "step": 1710
     },
     {
       "epoch": 23.89,
-      "learning_rate": 3.859859154929578e-05,
-      "loss": 0.3727,
+      "learning_rate": 2.0455223880597017e-05,
+      "loss": 0.2869,
       "step": 1720
     },
     {
       "epoch": 24.0,
-      "eval_loss": 0.5192346572875977,
-      "eval_runtime": 128.6258,
-      "eval_samples_per_second": 4.136,
-      "eval_steps_per_second": 0.521,
-      "eval_wer": 0.23609950504596003,
+      "eval_loss": 0.5296673774719238,
+      "eval_runtime": 132.3841,
+      "eval_samples_per_second": 4.019,
+      "eval_steps_per_second": 0.506,
+      "eval_wer": 0.2101947676287202,
       "step": 1728
     },
     {
       "epoch": 24.03,
-      "learning_rate": 3.852816901408451e-05,
-      "loss": 0.3787,
+      "learning_rate": 2.0417910447761196e-05,
+      "loss": 0.2958,
       "step": 1730
     },
     {
       "epoch": 24.17,
-      "learning_rate": 3.845774647887324e-05,
-      "loss": 0.3807,
+      "learning_rate": 2.0380597014925372e-05,
+      "loss": 0.2985,
       "step": 1740
     },
     {
       "epoch": 24.3,
-      "learning_rate": 3.838732394366197e-05,
-      "loss": 0.3858,
+      "learning_rate": 2.0343283582089555e-05,
+      "loss": 0.2914,
       "step": 1750
     },
     {
       "epoch": 24.44,
-      "learning_rate": 3.831690140845071e-05,
-      "loss": 0.3864,
+      "learning_rate": 2.0305970149253734e-05,
+      "loss": 0.2975,
       "step": 1760
     },
     {
       "epoch": 24.58,
-      "learning_rate": 3.8246478873239435e-05,
-      "loss": 0.3966,
+      "learning_rate": 2.026865671641791e-05,
+      "loss": 0.3037,
       "step": 1770
     },
     {
       "epoch": 24.72,
-      "learning_rate": 3.817605633802817e-05,
-      "loss": 0.3775,
+      "learning_rate": 2.0231343283582092e-05,
+      "loss": 0.294,
       "step": 1780
     },
     {
       "epoch": 24.86,
-      "learning_rate": 3.8105633802816906e-05,
-      "loss": 0.3813,
+      "learning_rate": 2.0194029850746268e-05,
+      "loss": 0.3024,
       "step": 1790
     },
     {
       "epoch": 25.0,
-      "learning_rate": 3.8035211267605634e-05,
-      "loss": 0.3751,
+      "learning_rate": 2.0156716417910447e-05,
+      "loss": 0.2997,
       "step": 1800
     },
     {
       "epoch": 25.0,
-      "eval_loss": 0.4995923638343811,
-      "eval_runtime": 127.8241,
-      "eval_samples_per_second": 4.162,
-      "eval_steps_per_second": 0.524,
-      "eval_wer": 0.24027768850035353,
+      "eval_loss": 0.5185533165931702,
+      "eval_runtime": 132.6367,
+      "eval_samples_per_second": 4.011,
+      "eval_steps_per_second": 0.505,
+      "eval_wer": 0.20878061322877162,
       "step": 1800
     },
     {
       "epoch": 25.14,
-      "learning_rate": 3.796478873239437e-05,
-      "loss": 0.3957,
+      "learning_rate": 2.011940298507463e-05,
+      "loss": 0.3077,
       "step": 1810
     },
     {
       "epoch": 25.28,
-      "learning_rate": 3.78943661971831e-05,
-      "loss": 0.3909,
+      "learning_rate": 2.0082089552238805e-05,
+      "loss": 0.2926,
       "step": 1820
     },
     {
       "epoch": 25.42,
-      "learning_rate": 3.7823943661971833e-05,
-      "loss": 0.3882,
+      "learning_rate": 2.0044776119402985e-05,
+      "loss": 0.2924,
       "step": 1830
     },
     {
       "epoch": 25.55,
-      "learning_rate": 3.775352112676057e-05,
-      "loss": 0.391,
+      "learning_rate": 2.0007462686567167e-05,
+      "loss": 0.2969,
       "step": 1840
     },
     {
       "epoch": 25.69,
-      "learning_rate": 3.76830985915493e-05,
-      "loss": 0.3897,
+      "learning_rate": 1.9970149253731343e-05,
+      "loss": 0.299,
       "step": 1850
     },
     {
       "epoch": 25.83,
-      "learning_rate": 3.761267605633803e-05,
-      "loss": 0.3926,
+      "learning_rate": 1.9932835820895522e-05,
+      "loss": 0.2963,
       "step": 1860
     },
     {
       "epoch": 25.97,
-      "learning_rate": 3.754225352112676e-05,
-      "loss": 0.3839,
+      "learning_rate": 1.9895522388059705e-05,
+      "loss": 0.2996,
       "step": 1870
     },
     {
       "epoch": 26.0,
-      "eval_loss": 0.4840199053287506,
-      "eval_runtime": 127.8495,
-      "eval_samples_per_second": 4.161,
-      "eval_steps_per_second": 0.524,
-      "eval_wer": 0.22272931799190074,
+      "eval_loss": 0.5218982100486755,
+      "eval_runtime": 131.3811,
+      "eval_samples_per_second": 4.049,
+      "eval_steps_per_second": 0.51,
+      "eval_wer": 0.20980908915600693,
       "step": 1872
     },
     {
       "epoch": 26.11,
-      "learning_rate": 3.7471830985915496e-05,
-      "loss": 0.3878,
+      "learning_rate": 1.985820895522388e-05,
+      "loss": 0.2972,
       "step": 1880
     },
     {
       "epoch": 26.25,
-      "learning_rate": 3.740140845070423e-05,
-      "loss": 0.4001,
+      "learning_rate": 1.982089552238806e-05,
+      "loss": 0.2993,
       "step": 1890
     },
     {
       "epoch": 26.39,
-      "learning_rate": 3.733098591549296e-05,
-      "loss": 0.3706,
+      "learning_rate": 1.9783582089552242e-05,
+      "loss": 0.2877,
       "step": 1900
     },
     {
       "epoch": 26.53,
-      "learning_rate": 3.7260563380281695e-05,
-      "loss": 0.3816,
+      "learning_rate": 1.9746268656716418e-05,
+      "loss": 0.2864,
       "step": 1910
     },
     {
       "epoch": 26.66,
-      "learning_rate": 3.7190140845070424e-05,
-      "loss": 0.3735,
+      "learning_rate": 1.9708955223880597e-05,
+      "loss": 0.2869,
       "step": 1920
     },
     {
       "epoch": 26.8,
-      "learning_rate": 3.711971830985916e-05,
-      "loss": 0.3979,
+      "learning_rate": 1.9671641791044776e-05,
+      "loss": 0.3113,
       "step": 1930
     },
     {
       "epoch": 26.94,
-      "learning_rate": 3.704929577464789e-05,
-      "loss": 0.3739,
+      "learning_rate": 1.9634328358208955e-05,
+      "loss": 0.298,
       "step": 1940
     },
     {
       "epoch": 27.0,
-      "eval_loss": 0.4790037274360657,
-      "eval_runtime": 129.2383,
-      "eval_samples_per_second": 4.116,
-      "eval_steps_per_second": 0.518,
-      "eval_wer": 0.2264575432281288,
+      "eval_loss": 0.506175696849823,
+      "eval_runtime": 132.0981,
+      "eval_samples_per_second": 4.027,
+      "eval_steps_per_second": 0.507,
+      "eval_wer": 0.21077328533779005,
       "step": 1944
     },
     {
       "epoch": 27.08,
-      "learning_rate": 3.697887323943662e-05,
-      "loss": 0.3946,
+      "learning_rate": 1.9597014925373135e-05,
+      "loss": 0.3057,
       "step": 1950
     },
     {
       "epoch": 27.22,
-      "learning_rate": 3.690845070422536e-05,
-      "loss": 0.3866,
+      "learning_rate": 1.9559701492537314e-05,
+      "loss": 0.3026,
       "step": 1960
     },
     {
       "epoch": 27.36,
-      "learning_rate": 3.6838028169014087e-05,
-      "loss": 0.3747,
+      "learning_rate": 1.9522388059701493e-05,
+      "loss": 0.298,
       "step": 1970
     },
     {
       "epoch": 27.5,
-      "learning_rate": 3.676760563380282e-05,
-      "loss": 0.3778,
+      "learning_rate": 1.9485074626865672e-05,
+      "loss": 0.3077,
       "step": 1980
     },
     {
       "epoch": 27.64,
-      "learning_rate": 3.669718309859155e-05,
-      "loss": 0.3772,
+      "learning_rate": 1.944776119402985e-05,
+      "loss": 0.3073,
       "step": 1990
     },
     {
       "epoch": 27.78,
-      "learning_rate": 3.6626760563380286e-05,
-      "loss": 0.3872,
+      "learning_rate": 1.941044776119403e-05,
+      "loss": 0.3163,
       "step": 2000
     },
     {
       "epoch": 27.91,
-      "learning_rate": 3.655633802816902e-05,
-      "loss": 0.3781,
+      "learning_rate": 1.9373134328358213e-05,
+      "loss": 0.3057,
       "step": 2010
     },
     {
       "epoch": 28.0,
-      "eval_loss": 0.5081353783607483,
-      "eval_runtime": 127.0954,
-      "eval_samples_per_second": 4.186,
-      "eval_steps_per_second": 0.527,
-      "eval_wer": 0.2247219901009192,
+      "eval_loss": 0.5299437642097473,
+      "eval_runtime": 132.6119,
+      "eval_samples_per_second": 4.012,
+      "eval_steps_per_second": 0.505,
+      "eval_wer": 0.21366587388313943,
       "step": 2016
     },
     {
       "epoch": 28.06,
-      "learning_rate": 3.648591549295775e-05,
-      "loss": 0.389,
+      "learning_rate": 1.933582089552239e-05,
+      "loss": 0.3058,
       "step": 2020
     },
     {
       "epoch": 28.19,
-      "learning_rate": 3.6415492957746485e-05,
-      "loss": 0.3737,
+      "learning_rate": 1.9298507462686568e-05,
+      "loss": 0.2959,
       "step": 2030
     },
     {
       "epoch": 28.33,
-      "learning_rate": 3.634507042253521e-05,
-      "loss": 0.4003,
+      "learning_rate": 1.9261194029850747e-05,
+      "loss": 0.3052,
       "step": 2040
     },
     {
       "epoch": 28.47,
-      "learning_rate": 3.627464788732395e-05,
-      "loss": 0.3863,
+      "learning_rate": 1.9223880597014926e-05,
+      "loss": 0.2961,
       "step": 2050
     },
     {
       "epoch": 28.61,
-      "learning_rate": 3.620422535211268e-05,
-      "loss": 0.395,
+      "learning_rate": 1.9186567164179105e-05,
+      "loss": 0.3006,
       "step": 2060
     },
     {
       "epoch": 28.75,
-      "learning_rate": 3.6133802816901405e-05,
-      "loss": 0.3946,
+      "learning_rate": 1.9149253731343284e-05,
+      "loss": 0.3096,
       "step": 2070
     },
     {
       "epoch": 28.89,
-      "learning_rate": 3.606338028169014e-05,
-      "loss": 0.3655,
+      "learning_rate": 1.9111940298507464e-05,
+      "loss": 0.2937,
       "step": 2080
     },
     {
       "epoch": 29.0,
-      "eval_loss": 0.4903033375740051,
-      "eval_runtime": 127.2191,
-      "eval_samples_per_second": 4.182,
-      "eval_steps_per_second": 0.527,
-      "eval_wer": 0.2262647039917722,
+      "eval_loss": 0.5345147848129272,
+      "eval_runtime": 131.6737,
+      "eval_samples_per_second": 4.04,
+      "eval_steps_per_second": 0.509,
+      "eval_wer": 0.21540142701034903,
       "step": 2088
     },
     {
       "epoch": 29.03,
-      "learning_rate": 3.5992957746478876e-05,
-      "loss": 0.3724,
+      "learning_rate": 1.9074626865671643e-05,
+      "loss": 0.3,
       "step": 2090
     },
     {
       "epoch": 29.17,
-      "learning_rate": 3.5922535211267604e-05,
-      "loss": 0.3667,
+      "learning_rate": 1.9037313432835822e-05,
+      "loss": 0.2895,
       "step": 2100
     },
     {
       "epoch": 29.3,
-      "learning_rate": 3.585211267605634e-05,
-      "loss": 0.3724,
+      "learning_rate": 1.9e-05,
+      "loss": 0.2907,
       "step": 2110
     },
     {
       "epoch": 29.44,
-      "learning_rate": 3.578169014084507e-05,
-      "loss": 0.3749,
+      "learning_rate": 1.896268656716418e-05,
+      "loss": 0.2975,
       "step": 2120
     },
     {
       "epoch": 29.58,
-      "learning_rate": 3.57112676056338e-05,
-      "loss": 0.3839,
+      "learning_rate": 1.892537313432836e-05,
+      "loss": 0.2945,
       "step": 2130
     },
     {
       "epoch": 29.72,
-      "learning_rate": 3.564084507042253e-05,
-      "loss": 0.3756,
+      "learning_rate": 1.888805970149254e-05,
+      "loss": 0.3019,
       "step": 2140
     },
     {
       "epoch": 29.86,
-      "learning_rate": 3.557042253521127e-05,
-      "loss": 0.3799,
+      "learning_rate": 1.8850746268656718e-05,
+      "loss": 0.2987,
       "step": 2150
     },
     {
       "epoch": 30.0,
-      "learning_rate": 3.55e-05,
-      "loss": 0.3832,
+      "learning_rate": 1.8813432835820897e-05,
+      "loss": 0.3064,
       "step": 2160
     },
     {
       "epoch": 30.0,
-      "eval_loss": 0.5195701718330383,
-      "eval_runtime": 128.2447,
-      "eval_samples_per_second": 4.148,
-      "eval_steps_per_second": 0.522,
-      "eval_wer": 0.2382850163913351,
+      "eval_loss": 0.5374366044998169,
+      "eval_runtime": 131.625,
+      "eval_samples_per_second": 4.042,
+      "eval_steps_per_second": 0.509,
+      "eval_wer": 0.21160892202866877,
       "step": 2160
     },
     {
       "epoch": 30.14,
-      "learning_rate": 3.543661971830986e-05,
-      "loss": 0.3806,
+      "learning_rate": 1.8776119402985076e-05,
+      "loss": 0.2945,
       "step": 2170
     },
     {
       "epoch": 30.28,
-      "learning_rate": 3.536619718309859e-05,
-      "loss": 0.379,
+      "learning_rate": 1.8738805970149255e-05,
+      "loss": 0.2925,
       "step": 2180
     },
     {
       "epoch": 30.42,
-      "learning_rate": 3.5295774647887326e-05,
-      "loss": 0.369,
+      "learning_rate": 1.8701492537313434e-05,
+      "loss": 0.2875,
       "step": 2190
     },
     {
       "epoch": 30.55,
-      "learning_rate": 3.5225352112676055e-05,
-      "loss": 0.3781,
+      "learning_rate": 1.8664179104477613e-05,
+      "loss": 0.2936,
       "step": 2200
     },
     {
       "epoch": 30.69,
-      "learning_rate": 3.515492957746479e-05,
-      "loss": 0.3635,
+      "learning_rate": 1.8626865671641793e-05,
+      "loss": 0.2937,
       "step": 2210
     },
     {
       "epoch": 30.83,
-      "learning_rate": 3.5084507042253525e-05,
-      "loss": 0.3843,
+      "learning_rate": 1.8589552238805972e-05,
+      "loss": 0.2952,
       "step": 2220
     },
     {
       "epoch": 30.97,
-      "learning_rate": 3.5014084507042254e-05,
-      "loss": 0.3662,
+      "learning_rate": 1.855223880597015e-05,
+      "loss": 0.2915,
       "step": 2230
     },
     {
       "epoch": 31.0,
-      "eval_loss": 0.5286476612091064,
-      "eval_runtime": 127.7591,
-      "eval_samples_per_second": 4.164,
-      "eval_steps_per_second": 0.524,
-      "eval_wer": 0.23629234428231663,
+      "eval_loss": 0.5184913873672485,
+      "eval_runtime": 131.6654,
+      "eval_samples_per_second": 4.041,
+      "eval_steps_per_second": 0.509,
+      "eval_wer": 0.20736645882882304,
       "step": 2232
     },
     {
       "epoch": 31.11,
-      "learning_rate": 3.494366197183099e-05,
-      "loss": 0.3803,
+      "learning_rate": 1.851492537313433e-05,
+      "loss": 0.3009,
       "step": 2240
     },
     {
       "epoch": 31.25,
-      "learning_rate": 3.487323943661972e-05,
-      "loss": 0.3676,
+      "learning_rate": 1.847761194029851e-05,
+      "loss": 0.3002,
       "step": 2250
     },
     {
       "epoch": 31.39,
-      "learning_rate": 3.480281690140845e-05,
-      "loss": 0.3443,
+      "learning_rate": 1.844029850746269e-05,
+      "loss": 0.2676,
       "step": 2260
     },
     {
       "epoch": 31.53,
-      "learning_rate": 3.473239436619719e-05,
-      "loss": 0.3662,
+      "learning_rate": 1.8402985074626868e-05,
+      "loss": 0.2795,
       "step": 2270
     },
     {
       "epoch": 31.66,
-      "learning_rate": 3.466197183098592e-05,
-      "loss": 0.3751,
+      "learning_rate": 1.8365671641791047e-05,
+      "loss": 0.2855,
       "step": 2280
     },
     {
       "epoch": 31.8,
-      "learning_rate": 3.459154929577465e-05,
-      "loss": 0.3838,
+      "learning_rate": 1.8328358208955223e-05,
+      "loss": 0.2967,
       "step": 2290
     },
     {
       "epoch": 31.94,
-      "learning_rate": 3.452112676056338e-05,
-      "loss": 0.3692,
+      "learning_rate": 1.8291044776119405e-05,
+      "loss": 0.2942,
       "step": 2300
     },
     {
       "epoch": 32.0,
-      "eval_loss": 0.5283333659172058,
-      "eval_runtime": 128.3957,
-      "eval_samples_per_second": 4.143,
-      "eval_steps_per_second": 0.522,
-      "eval_wer": 0.24599858584560005,
+      "eval_loss": 0.5221661925315857,
+      "eval_runtime": 131.7358,
+      "eval_samples_per_second": 4.038,
+      "eval_steps_per_second": 0.509,
+      "eval_wer": 0.20935913093784148,
       "step": 2304
     },
     {
       "epoch": 32.08,
-      "learning_rate": 3.4450704225352116e-05,
-      "loss": 0.3885,
+      "learning_rate": 1.8253731343283584e-05,
+      "loss": 0.3027,
       "step": 2310
     },
     {
       "epoch": 32.22,
-      "learning_rate": 3.4380281690140844e-05,
-      "loss": 0.3639,
+      "learning_rate": 1.821641791044776e-05,
+      "loss": 0.2919,
       "step": 2320
     },
     {
       "epoch": 32.36,
-      "learning_rate": 3.430985915492958e-05,
-      "loss": 0.3875,
+      "learning_rate": 1.8179104477611943e-05,
+      "loss": 0.2963,
       "step": 2330
     },
     {
       "epoch": 32.5,
-      "learning_rate": 3.4239436619718315e-05,
-      "loss": 0.3779,
+      "learning_rate": 1.814179104477612e-05,
+      "loss": 0.2917,
       "step": 2340
     },
     {
       "epoch": 32.64,
-      "learning_rate": 3.416901408450704e-05,
-      "loss": 0.3504,
+      "learning_rate": 1.8104477611940297e-05,
+      "loss": 0.2753,
       "step": 2350
     },
     {
       "epoch": 32.78,
-      "learning_rate": 3.409859154929578e-05,
-      "loss": 0.3638,
+      "learning_rate": 1.806716417910448e-05,
+      "loss": 0.288,
       "step": 2360
     },
     {
       "epoch": 32.91,
-      "learning_rate": 3.402816901408451e-05,
-      "loss": 0.3484,
+      "learning_rate": 1.802985074626866e-05,
+      "loss": 0.2777,
       "step": 2370
     },
     {
       "epoch": 33.0,
-      "eval_loss": 0.49058952927589417,
-      "eval_runtime": 127.7529,
-      "eval_samples_per_second": 4.164,
-      "eval_steps_per_second": 0.524,
-      "eval_wer": 0.22665038246448543,
+      "eval_loss": 0.5229014158248901,
+      "eval_runtime": 131.9614,
+      "eval_samples_per_second": 4.031,
+      "eval_steps_per_second": 0.508,
+      "eval_wer": 0.21077328533779005,
       "step": 2376
     },
     {
       "epoch": 33.06,
-      "learning_rate": 3.395774647887324e-05,
-      "loss": 0.387,
+      "learning_rate": 1.7992537313432835e-05,
+      "loss": 0.3004,
       "step": 2380
     },
     {
       "epoch": 33.19,
-      "learning_rate": 3.388732394366198e-05,
-      "loss": 0.3559,
+      "learning_rate": 1.7955223880597017e-05,
+      "loss": 0.2916,
       "step": 2390
     },
     {
       "epoch": 33.33,
-      "learning_rate": 3.3816901408450706e-05,
-      "loss": 0.3691,
+      "learning_rate": 1.7917910447761193e-05,
+      "loss": 0.2933,
       "step": 2400
     },
     {
       "epoch": 33.47,
-      "learning_rate": 3.374647887323944e-05,
-      "loss": 0.371,
+      "learning_rate": 1.7880597014925372e-05,
+      "loss": 0.296,
       "step": 2410
     },
     {
       "epoch": 33.61,
-      "learning_rate": 3.367605633802817e-05,
-      "loss": 0.3688,
+      "learning_rate": 1.7843283582089555e-05,
+      "loss": 0.2959,
       "step": 2420
     },
     {
       "epoch": 33.75,
-      "learning_rate": 3.3605633802816905e-05,
-      "loss": 0.3602,
+      "learning_rate": 1.780597014925373e-05,
+      "loss": 0.2846,
       "step": 2430
     },
     {
       "epoch": 33.89,
-      "learning_rate": 3.353521126760564e-05,
-      "loss": 0.3493,
+      "learning_rate": 1.776865671641791e-05,
+      "loss": 0.2821,
       "step": 2440
     },
     {
       "epoch": 34.0,
-      "eval_loss": 0.4982738196849823,
-      "eval_runtime": 129.337,
-      "eval_samples_per_second": 4.113,
-      "eval_steps_per_second": 0.518,
-      "eval_wer": 0.2214437230828566,
+      "eval_loss": 0.5242453813552856,
+      "eval_runtime": 132.7946,
+      "eval_samples_per_second": 4.006,
+      "eval_steps_per_second": 0.505,
+      "eval_wer": 0.2101947676287202,
       "step": 2448
     },
     {
       "epoch": 34.03,
-      "learning_rate": 3.346478873239437e-05,
-      "loss": 0.3825,
+      "learning_rate": 1.7731343283582092e-05,
+      "loss": 0.2997,
       "step": 2450
     },
     {
       "epoch": 34.17,
-      "learning_rate": 3.3394366197183104e-05,
-      "loss": 0.3668,
+      "learning_rate": 1.7694029850746268e-05,
+      "loss": 0.2872,
       "step": 2460
     },
     {
       "epoch": 34.3,
-      "learning_rate": 3.332394366197183e-05,
-      "loss": 0.3679,
+      "learning_rate": 1.7656716417910447e-05,
+      "loss": 0.2826,
       "step": 2470
     },
     {
       "epoch": 34.44,
-      "learning_rate": 3.325352112676057e-05,
-      "loss": 0.3697,
+      "learning_rate": 1.761940298507463e-05,
+      "loss": 0.2965,
       "step": 2480
     },
     {
       "epoch": 34.58,
-      "learning_rate": 3.31830985915493e-05,
-      "loss": 0.3699,
+      "learning_rate": 1.7582089552238806e-05,
+      "loss": 0.2812,
       "step": 2490
     },
     {
       "epoch": 34.72,
-      "learning_rate": 3.311267605633803e-05,
-      "loss": 0.3794,
+      "learning_rate": 1.7548507462686566e-05,
+      "loss": 0.2893,
       "step": 2500
     },
     {
       "epoch": 34.86,
-      "learning_rate": 3.304225352112677e-05,
-      "loss": 0.3838,
+      "learning_rate": 1.751119402985075e-05,
+      "loss": 0.2799,
       "step": 2510
     },
     {
       "epoch": 35.0,
-      "learning_rate": 3.2971830985915495e-05,
-      "loss": 0.3783,
+      "learning_rate": 1.7473880597014928e-05,
+      "loss": 0.2859,
       "step": 2520
     },
     {
       "epoch": 35.0,
-      "eval_loss": 0.5358861088752747,
-      "eval_runtime": 128.4601,
-      "eval_samples_per_second": 4.141,
-      "eval_steps_per_second": 0.522,
-      "eval_wer": 0.2319213215915665,
+      "eval_loss": 0.5292270183563232,
+      "eval_runtime": 131.3549,
+      "eval_samples_per_second": 4.05,
+      "eval_steps_per_second": 0.51,
+      "eval_wer": 0.2096162499196503,
       "step": 2520
     },
     {
       "epoch": 35.14,
-      "learning_rate": 3.2901408450704224e-05,
-      "loss": 0.3732,
+      "learning_rate": 1.7436567164179103e-05,
+      "loss": 0.2936,
       "step": 2530
     },
     {
       "epoch": 35.28,
-      "learning_rate": 3.283098591549296e-05,
-      "loss": 0.3765,
+      "learning_rate": 1.7399253731343286e-05,
+      "loss": 0.2941,
       "step": 2540
     },
     {
       "epoch": 35.42,
-      "learning_rate": 3.276056338028169e-05,
-      "loss": 0.3705,
+      "learning_rate": 1.7361940298507465e-05,
+      "loss": 0.2887,
       "step": 2550
     },
     {
       "epoch": 35.55,
-      "learning_rate": 3.269014084507042e-05,
-      "loss": 0.3824,
+      "learning_rate": 1.732462686567164e-05,
+      "loss": 0.296,
       "step": 2560
     },
     {
       "epoch": 35.69,
-      "learning_rate": 3.261971830985915e-05,
-      "loss": 0.368,
+      "learning_rate": 1.7287313432835823e-05,
+      "loss": 0.2905,
       "step": 2570
     },
     {
       "epoch": 35.83,
-      "learning_rate": 3.2549295774647887e-05,
-      "loss": 0.3926,
+      "learning_rate": 1.725e-05,
+      "loss": 0.2872,
       "step": 2580
     },
     {
       "epoch": 35.97,
-      "learning_rate": 3.247887323943662e-05,
-      "loss": 0.3604,
+      "learning_rate": 1.721268656716418e-05,
+      "loss": 0.283,
       "step": 2590
     },
     {
       "epoch": 36.0,
-      "eval_loss": 0.5129276514053345,
-      "eval_runtime": 127.2766,
-      "eval_samples_per_second": 4.18,
-      "eval_steps_per_second": 0.526,
-      "eval_wer": 0.23706370122774315,
+      "eval_loss": 0.5041770339012146,
+      "eval_runtime": 131.4728,
+      "eval_samples_per_second": 4.046,
+      "eval_steps_per_second": 0.51,
+      "eval_wer": 0.20665938162884875,
       "step": 2592
     },
     {
       "epoch": 36.11,
-      "learning_rate": 3.240845070422535e-05,
-      "loss": 0.379,
+      "learning_rate": 1.717537313432836e-05,
+      "loss": 0.2914,
       "step": 2600
     },
     {
       "epoch": 36.25,
-      "learning_rate": 3.2338028169014086e-05,
-      "loss": 0.3653,
+      "learning_rate": 1.7138059701492537e-05,
+      "loss": 0.2829,
       "step": 2610
     },
     {
       "epoch": 36.39,
-      "learning_rate": 3.2267605633802814e-05,
-      "loss": 0.3473,
+      "learning_rate": 1.7100746268656716e-05,
+      "loss": 0.2682,
       "step": 2620
     },
     {
       "epoch": 36.53,
-      "learning_rate": 3.219718309859155e-05,
-      "loss": 0.3891,
+      "learning_rate": 1.70634328358209e-05,
+      "loss": 0.2989,
       "step": 2630
     },
     {
       "epoch": 36.66,
-      "learning_rate": 3.2126760563380285e-05,
-      "loss": 0.3686,
+      "learning_rate": 1.7026119402985074e-05,
+      "loss": 0.2873,
       "step": 2640
     },
     {
       "epoch": 36.8,
-      "learning_rate": 3.205633802816901e-05,
-      "loss": 0.3736,
+      "learning_rate": 1.6992537313432834e-05,
+      "loss": 0.29,
       "step": 2650
     },
     {
       "epoch": 36.94,
-      "learning_rate": 3.198591549295775e-05,
-      "loss": 0.3796,
+      "learning_rate": 1.6955223880597017e-05,
+      "loss": 0.2895,
       "step": 2660
     },
     {
       "epoch": 37.0,
-      "eval_loss": 0.4993954300880432,
-      "eval_runtime": 129.0338,
-      "eval_samples_per_second": 4.123,
-      "eval_steps_per_second": 0.519,
-      "eval_wer": 0.2320498810824709,
+      "eval_loss": 0.5222975015640259,
+      "eval_runtime": 131.841,
+      "eval_samples_per_second": 4.035,
+      "eval_steps_per_second": 0.508,
+      "eval_wer": 0.2045381500289259,
       "step": 2664
     },
     {
       "epoch": 37.08,
-      "learning_rate": 3.191549295774648e-05,
-      "loss": 0.3864,
+      "learning_rate": 1.6917910447761196e-05,
+      "loss": 0.2939,
       "step": 2670
     },
     {
       "epoch": 37.22,
-      "learning_rate": 3.184507042253521e-05,
-      "loss": 0.3758,
+      "learning_rate": 1.6880597014925372e-05,
+      "loss": 0.295,
       "step": 2680
     },
     {
       "epoch": 37.36,
-      "learning_rate": 3.177464788732394e-05,
-      "loss": 0.3665,
+      "learning_rate": 1.6843283582089554e-05,
+      "loss": 0.2793,
       "step": 2690
     },
     {
       "epoch": 37.5,
-      "learning_rate": 3.1704225352112676e-05,
-      "loss": 0.3618,
+      "learning_rate": 1.6805970149253734e-05,
+      "loss": 0.2842,
       "step": 2700
     },
     {
       "epoch": 37.64,
-      "learning_rate": 3.163380281690141e-05,
-      "loss": 0.3579,
+      "learning_rate": 1.676865671641791e-05,
+      "loss": 0.2678,
       "step": 2710
     },
     {
       "epoch": 37.78,
-      "learning_rate": 3.156338028169014e-05,
-      "loss": 0.3909,
+      "learning_rate": 1.6731343283582092e-05,
+      "loss": 0.2868,
       "step": 2720
     },
     {
       "epoch": 37.91,
-      "learning_rate": 3.1492957746478875e-05,
-      "loss": 0.3629,
+      "learning_rate": 1.669402985074627e-05,
+      "loss": 0.2822,
       "step": 2730
     },
     {
       "epoch": 38.0,
-      "eval_loss": 0.47256624698638916,
-      "eval_runtime": 126.9577,
-      "eval_samples_per_second": 4.19,
-      "eval_steps_per_second": 0.528,
-      "eval_wer": 0.22260075850099634,
+      "eval_loss": 0.5093514919281006,
+      "eval_runtime": 132.4112,
+      "eval_samples_per_second": 4.018,
+      "eval_steps_per_second": 0.506,
+      "eval_wer": 0.2071093398470142,
       "step": 2736
     },
     {
       "epoch": 38.06,
-      "learning_rate": 3.14225352112676e-05,
-      "loss": 0.38,
+      "learning_rate": 1.6656716417910447e-05,
+      "loss": 0.2914,
       "step": 2740
     },
     {
       "epoch": 38.19,
-      "learning_rate": 3.135211267605634e-05,
-      "loss": 0.3617,
+      "learning_rate": 1.661940298507463e-05,
+      "loss": 0.2828,
       "step": 2750
     },
     {
       "epoch": 38.33,
-      "learning_rate": 3.1281690140845074e-05,
-      "loss": 0.3469,
+      "learning_rate": 1.658208955223881e-05,
+      "loss": 0.2637,
       "step": 2760
     },
     {
       "epoch": 38.47,
-      "learning_rate": 3.12112676056338e-05,
-      "loss": 0.3528,
+      "learning_rate": 1.6544776119402984e-05,
+      "loss": 0.288,
       "step": 2770
     },
     {
       "epoch": 38.61,
-      "learning_rate": 3.114084507042254e-05,
-      "loss": 0.3533,
+      "learning_rate": 1.6507462686567167e-05,
+      "loss": 0.2814,
       "step": 2780
     },
     {
       "epoch": 38.75,
-      "learning_rate": 3.1070422535211266e-05,
-      "loss": 0.3649,
+      "learning_rate": 1.6470149253731343e-05,
+      "loss": 0.2902,
       "step": 2790
     },
     {
       "epoch": 38.89,
-      "learning_rate": 3.1e-05,
-      "loss": 0.3428,
+      "learning_rate": 1.6432835820895522e-05,
+      "loss": 0.2672,
       "step": 2800
     },
     {
       "epoch": 39.0,
-      "eval_loss": 0.5130040645599365,
-      "eval_runtime": 127.749,
-      "eval_samples_per_second": 4.164,
-      "eval_steps_per_second": 0.524,
-      "eval_wer": 0.22253647875554414,
+      "eval_loss": 0.5250556468963623,
+      "eval_runtime": 131.822,
+      "eval_samples_per_second": 4.036,
+      "eval_steps_per_second": 0.508,
+      "eval_wer": 0.20415247155621263,
       "step": 2808
     },
     {
       "epoch": 39.03,
-      "learning_rate": 3.092957746478874e-05,
-      "loss": 0.3717,
+      "learning_rate": 1.6395522388059704e-05,
+      "loss": 0.2954,
       "step": 2810
     },
     {
       "epoch": 39.17,
-      "learning_rate": 3.0859154929577465e-05,
-      "loss": 0.36,
+      "learning_rate": 1.635820895522388e-05,
+      "loss": 0.2854,
       "step": 2820
     },
     {
       "epoch": 39.3,
-      "learning_rate": 3.07887323943662e-05,
-      "loss": 0.3595,
+      "learning_rate": 1.632089552238806e-05,
+      "loss": 0.2874,
       "step": 2830
     },
     {
       "epoch": 39.44,
-      "learning_rate": 3.071830985915493e-05,
-      "loss": 0.3378,
+      "learning_rate": 1.6283582089552242e-05,
+      "loss": 0.2671,
       "step": 2840
     },
     {
       "epoch": 39.58,
-      "learning_rate": 3.0647887323943664e-05,
-      "loss": 0.3543,
+      "learning_rate": 1.6246268656716418e-05,
+      "loss": 0.2761,
       "step": 2850
     },
     {
       "epoch": 39.72,
-      "learning_rate": 3.05774647887324e-05,
-      "loss": 0.3628,
+      "learning_rate": 1.6208955223880597e-05,
+      "loss": 0.2904,
       "step": 2860
     },
     {
       "epoch": 39.86,
-      "learning_rate": 3.0507042253521128e-05,
-      "loss": 0.3473,
+      "learning_rate": 1.617164179104478e-05,
+      "loss": 0.2783,
       "step": 2870
     },
     {
       "epoch": 40.0,
-      "learning_rate": 3.043661971830986e-05,
-      "loss": 0.375,
+      "learning_rate": 1.6134328358208955e-05,
+      "loss": 0.2925,
       "step": 2880
     },
     {
       "epoch": 40.0,
-      "eval_loss": 0.5154946446418762,
-      "eval_runtime": 127.9888,
-      "eval_samples_per_second": 4.157,
-      "eval_steps_per_second": 0.523,
-      "eval_wer": 0.22851449508259947,
+      "eval_loss": 0.5071823000907898,
+      "eval_runtime": 132.8431,
+      "eval_samples_per_second": 4.005,
+      "eval_steps_per_second": 0.504,
+      "eval_wer": 0.20511666773799575,
       "step": 2880
     },
     {
       "epoch": 40.14,
-      "learning_rate": 3.0366197183098595e-05,
-      "loss": 0.375,
+      "learning_rate": 1.6097014925373134e-05,
+      "loss": 0.2936,
       "step": 2890
     },
     {
       "epoch": 40.28,
-      "learning_rate": 3.0295774647887327e-05,
-      "loss": 0.3668,
+      "learning_rate": 1.6059701492537313e-05,
+      "loss": 0.2848,
       "step": 2900
     },
     {
       "epoch": 40.42,
-      "learning_rate": 3.022535211267606e-05,
-      "loss": 0.3508,
+      "learning_rate": 1.6022388059701492e-05,
+      "loss": 0.2866,
       "step": 2910
     },
     {
       "epoch": 40.55,
-      "learning_rate": 3.015492957746479e-05,
-      "loss": 0.3602,
+      "learning_rate": 1.598507462686567e-05,
+      "loss": 0.289,
       "step": 2920
     },
     {
       "epoch": 40.69,
-      "learning_rate": 3.0084507042253523e-05,
-      "loss": 0.3435,
+      "learning_rate": 1.594776119402985e-05,
+      "loss": 0.2817,
       "step": 2930
     },
     {
       "epoch": 40.83,
-      "learning_rate": 3.0014084507042254e-05,
-      "loss": 0.3787,
+      "learning_rate": 1.591044776119403e-05,
+      "loss": 0.293,
       "step": 2940
     },
     {
       "epoch": 40.97,
-      "learning_rate": 2.994366197183099e-05,
-      "loss": 0.3581,
+      "learning_rate": 1.587313432835821e-05,
+      "loss": 0.295,
       "step": 2950
     },
     {
       "epoch": 41.0,
-      "eval_loss": 0.4994366765022278,
-      "eval_runtime": 127.9492,
-      "eval_samples_per_second": 4.158,
-      "eval_steps_per_second": 0.524,
-      "eval_wer": 0.22665038246448543,
+      "eval_loss": 0.4903075695037842,
+      "eval_runtime": 131.5707,
+      "eval_samples_per_second": 4.043,
+      "eval_steps_per_second": 0.509,
+      "eval_wer": 0.2071093398470142,
       "step": 2952
     },
     {
       "epoch": 41.11,
-      "learning_rate": 2.988028169014085e-05,
-      "loss": 0.3779,
+      "learning_rate": 1.5835820895522388e-05,
+      "loss": 0.2957,
       "step": 2960
     },
     {
       "epoch": 41.25,
-      "learning_rate": 2.9809859154929582e-05,
-      "loss": 0.3643,
+      "learning_rate": 1.5798507462686567e-05,
+      "loss": 0.294,
       "step": 2970
     },
     {
       "epoch": 41.39,
-      "learning_rate": 2.9739436619718314e-05,
-      "loss": 0.3476,
+      "learning_rate": 1.5761194029850747e-05,
+      "loss": 0.2779,
       "step": 2980
     },
     {
       "epoch": 41.53,
-      "learning_rate": 2.9669014084507046e-05,
-      "loss": 0.3483,
+      "learning_rate": 1.5723880597014926e-05,
+      "loss": 0.2808,
       "step": 2990
     },
     {
       "epoch": 41.66,
-      "learning_rate": 2.9598591549295778e-05,
-      "loss": 0.3596,
+      "learning_rate": 1.5686567164179105e-05,
+      "loss": 0.2924,
       "step": 3000
     },
     {
       "epoch": 41.8,
-      "learning_rate": 2.9528169014084506e-05,
-      "loss": 0.3649,
+      "learning_rate": 1.5649253731343284e-05,
+      "loss": 0.2917,
       "step": 3010
     },
     {
       "epoch": 41.94,
-      "learning_rate": 2.9457746478873238e-05,
-      "loss": 0.3394,
+      "learning_rate": 1.5611940298507463e-05,
+      "loss": 0.2725,
       "step": 3020
     },
     {
       "epoch": 42.0,
-      "eval_loss": 0.4989403784275055,
-      "eval_runtime": 129.1484,
-      "eval_samples_per_second": 4.119,
-      "eval_steps_per_second": 0.519,
-      "eval_wer": 0.21765121810117632,
+      "eval_loss": 0.49732398986816406,
+      "eval_runtime": 131.4816,
+      "eval_samples_per_second": 4.046,
+      "eval_steps_per_second": 0.51,
+      "eval_wer": 0.20485954875618692,
       "step": 3024
     },
     {
       "epoch": 42.08,
-      "learning_rate": 2.938732394366197e-05,
-      "loss": 0.3666,
+      "learning_rate": 1.5574626865671642e-05,
+      "loss": 0.2972,
       "step": 3030
     },
     {
       "epoch": 42.22,
-      "learning_rate": 2.93169014084507e-05,
-      "loss": 0.3416,
+      "learning_rate": 1.553731343283582e-05,
+      "loss": 0.2854,
       "step": 3040
     },
     {
       "epoch": 42.36,
-      "learning_rate": 2.9246478873239437e-05,
-      "loss": 0.3542,
+      "learning_rate": 1.55e-05,
+      "loss": 0.2874,
       "step": 3050
     },
     {
       "epoch": 42.5,
-      "learning_rate": 2.917605633802817e-05,
-      "loss": 0.3639,
+      "learning_rate": 1.546268656716418e-05,
+      "loss": 0.2928,
       "step": 3060
     },
     {
       "epoch": 42.64,
-      "learning_rate": 2.91056338028169e-05,
-      "loss": 0.3514,
+      "learning_rate": 1.542537313432836e-05,
+      "loss": 0.2773,
       "step": 3070
     },
     {
       "epoch": 42.78,
-      "learning_rate": 2.9035211267605633e-05,
-      "loss": 0.3658,
+      "learning_rate": 1.5388059701492538e-05,
+      "loss": 0.2954,
       "step": 3080
     },
     {
       "epoch": 42.91,
-      "learning_rate": 2.8964788732394364e-05,
-      "loss": 0.3421,
+      "learning_rate": 1.5350746268656717e-05,
+      "loss": 0.2758,
       "step": 3090
     },
     {
       "epoch": 43.0,
-      "eval_loss": 0.4820369780063629,
-      "eval_runtime": 128.7733,
-      "eval_samples_per_second": 4.131,
-      "eval_steps_per_second": 0.52,
-      "eval_wer": 0.22735745966445972,
+      "eval_loss": 0.5097233057022095,
+      "eval_runtime": 131.4822,
+      "eval_samples_per_second": 4.046,
+      "eval_steps_per_second": 0.51,
+      "eval_wer": 0.2015812817381243,
       "step": 3096
     },
     {
       "epoch": 43.06,
-      "learning_rate": 2.88943661971831e-05,
-      "loss": 0.3618,
+      "learning_rate": 1.5313432835820896e-05,
+      "loss": 0.2904,
       "step": 3100
     },
     {
       "epoch": 43.19,
-      "learning_rate": 2.882394366197183e-05,
-      "loss": 0.3449,
+      "learning_rate": 1.5276119402985076e-05,
+      "loss": 0.2831,
       "step": 3110
     },
     {
       "epoch": 43.33,
-      "learning_rate": 2.8753521126760563e-05,
-      "loss": 0.3446,
+      "learning_rate": 1.5238805970149256e-05,
+      "loss": 0.2779,
       "step": 3120
     },
     {
       "epoch": 43.47,
-      "learning_rate": 2.8683098591549295e-05,
-      "loss": 0.3456,
+      "learning_rate": 1.5201492537313434e-05,
+      "loss": 0.2844,
       "step": 3130
     },
     {
       "epoch": 43.61,
-      "learning_rate": 2.8612676056338027e-05,
-      "loss": 0.3545,
+      "learning_rate": 1.5164179104477613e-05,
+      "loss": 0.2842,
       "step": 3140
     },
     {
       "epoch": 43.75,
-      "learning_rate": 2.854225352112676e-05,
-      "loss": 0.3493,
+      "learning_rate": 1.512686567164179e-05,
+      "loss": 0.2856,
       "step": 3150
     },
     {
       "epoch": 43.89,
-      "learning_rate": 2.8471830985915494e-05,
-      "loss": 0.342,
+      "learning_rate": 1.5089552238805971e-05,
+      "loss": 0.2809,
       "step": 3160
     },
     {
       "epoch": 44.0,
-      "eval_loss": 0.48394179344177246,
-      "eval_runtime": 129.9101,
-      "eval_samples_per_second": 4.095,
-      "eval_steps_per_second": 0.516,
-      "eval_wer": 0.21887253326476827,
+      "eval_loss": 0.49982935190200806,
+      "eval_runtime": 131.8613,
+      "eval_samples_per_second": 4.035,
+      "eval_steps_per_second": 0.508,
+      "eval_wer": 0.20280259690171626,
       "step": 3168
     },
     {
       "epoch": 44.03,
-      "learning_rate": 2.8401408450704226e-05,
-      "loss": 0.3561,
+      "learning_rate": 1.505223880597015e-05,
+      "loss": 0.2936,
       "step": 3170
     },
     {
       "epoch": 44.17,
-      "learning_rate": 2.8330985915492958e-05,
-      "loss": 0.3402,
+      "learning_rate": 1.5014925373134328e-05,
+      "loss": 0.285,
       "step": 3180
     },
     {
       "epoch": 44.3,
-      "learning_rate": 2.826056338028169e-05,
-      "loss": 0.3596,
+      "learning_rate": 1.4977611940298509e-05,
+      "loss": 0.2833,
       "step": 3190
     },
     {
       "epoch": 44.44,
-      "learning_rate": 2.8190140845070422e-05,
-      "loss": 0.3485,
+      "learning_rate": 1.4940298507462688e-05,
+      "loss": 0.282,
       "step": 3200
     },
     {
       "epoch": 44.58,
-      "learning_rate": 2.8119718309859157e-05,
-      "loss": 0.3435,
+      "learning_rate": 1.4902985074626866e-05,
+      "loss": 0.2816,
       "step": 3210
     },
     {
       "epoch": 44.72,
-      "learning_rate": 2.804929577464789e-05,
-      "loss": 0.332,
+      "learning_rate": 1.4865671641791046e-05,
+      "loss": 0.2711,
       "step": 3220
     },
     {
       "epoch": 44.86,
-      "learning_rate": 2.797887323943662e-05,
-      "loss": 0.3339,
+      "learning_rate": 1.4828358208955226e-05,
+      "loss": 0.2702,
       "step": 3230
     },
     {
       "epoch": 45.0,
-      "learning_rate": 2.7908450704225353e-05,
-      "loss": 0.3482,
+      "learning_rate": 1.4791044776119403e-05,
+      "loss": 0.2804,
       "step": 3240
     },
     {
       "epoch": 45.0,
-      "eval_loss": 0.4978794753551483,
-      "eval_runtime": 129.1828,
-      "eval_samples_per_second": 4.118,
-      "eval_steps_per_second": 0.519,
-      "eval_wer": 0.21199460050138202,
+      "eval_loss": 0.5103247761726379,
+      "eval_runtime": 131.3748,
+      "eval_samples_per_second": 4.049,
+      "eval_steps_per_second": 0.51,
+      "eval_wer": 0.20421675130166483,
       "step": 3240
     },
     {
       "epoch": 45.14,
-      "learning_rate": 2.7838028169014085e-05,
-      "loss": 0.3325,
+      "learning_rate": 1.4753731343283584e-05,
+      "loss": 0.2664,
       "step": 3250
     },
     {
       "epoch": 45.28,
-      "learning_rate": 2.7767605633802816e-05,
-      "loss": 0.3409,
+      "learning_rate": 1.4716417910447761e-05,
+      "loss": 0.2755,
       "step": 3260
     },
     {
       "epoch": 45.42,
-      "learning_rate": 2.7697183098591552e-05,
-      "loss": 0.3368,
+      "learning_rate": 1.467910447761194e-05,
+      "loss": 0.2725,
       "step": 3270
     },
     {
       "epoch": 45.55,
-      "learning_rate": 2.7626760563380284e-05,
-      "loss": 0.35,
+      "learning_rate": 1.4641791044776121e-05,
+      "loss": 0.2878,
       "step": 3280
     },
     {
       "epoch": 45.69,
-      "learning_rate": 2.7556338028169015e-05,
-      "loss": 0.3433,
+      "learning_rate": 1.4604477611940299e-05,
+      "loss": 0.2847,
       "step": 3290
     },
     {
       "epoch": 45.83,
-      "learning_rate": 2.7485915492957747e-05,
-      "loss": 0.3441,
+      "learning_rate": 1.4567164179104478e-05,
+      "loss": 0.2826,
       "step": 3300
     },
     {
       "epoch": 45.97,
-      "learning_rate": 2.741549295774648e-05,
-      "loss": 0.3494,
+      "learning_rate": 1.4529850746268659e-05,
+      "loss": 0.2842,
       "step": 3310
     },
     {
       "epoch": 46.0,
-      "eval_loss": 0.5068593621253967,
-      "eval_runtime": 128.5981,
-      "eval_samples_per_second": 4.137,
-      "eval_steps_per_second": 0.521,
-      "eval_wer": 0.21861541428295944,
+      "eval_loss": 0.48608192801475525,
+      "eval_runtime": 131.8901,
+      "eval_samples_per_second": 4.034,
+      "eval_steps_per_second": 0.508,
+      "eval_wer": 0.2021597994471942,
       "step": 3312
     },
     {
       "epoch": 46.11,
-      "learning_rate": 2.734507042253521e-05,
-      "loss": 0.3578,
+      "learning_rate": 1.4492537313432836e-05,
+      "loss": 0.2921,
       "step": 3320
     },
     {
       "epoch": 46.25,
-      "learning_rate": 2.7274647887323946e-05,
-      "loss": 0.3451,
+      "learning_rate": 1.4455223880597015e-05,
+      "loss": 0.2833,
       "step": 3330
     },
     {
       "epoch": 46.39,
-      "learning_rate": 2.7204225352112678e-05,
-      "loss": 0.3417,
+      "learning_rate": 1.4417910447761196e-05,
+      "loss": 0.2787,
       "step": 3340
     },
     {
       "epoch": 46.53,
-      "learning_rate": 2.713380281690141e-05,
-      "loss": 0.3444,
+      "learning_rate": 1.4380597014925374e-05,
+      "loss": 0.2772,
       "step": 3350
     },
     {
       "epoch": 46.66,
-      "learning_rate": 2.7063380281690142e-05,
-      "loss": 0.3409,
+      "learning_rate": 1.4343283582089553e-05,
+      "loss": 0.2756,
       "step": 3360
     },
     {
       "epoch": 46.8,
-      "learning_rate": 2.6992957746478874e-05,
-      "loss": 0.3477,
+      "learning_rate": 1.430597014925373e-05,
+      "loss": 0.2745,
       "step": 3370
     },
     {
       "epoch": 46.94,
-      "learning_rate": 2.692253521126761e-05,
-      "loss": 0.345,
+      "learning_rate": 1.4268656716417911e-05,
+      "loss": 0.2721,
       "step": 3380
     },
     {
       "epoch": 47.0,
-      "eval_loss": 0.49980834126472473,
-      "eval_runtime": 128.798,
-      "eval_samples_per_second": 4.13,
-      "eval_steps_per_second": 0.52,
-      "eval_wer": 0.23211416082792313,
+      "eval_loss": 0.5053983926773071,
+      "eval_runtime": 131.5158,
+      "eval_samples_per_second": 4.045,
+      "eval_steps_per_second": 0.509,
+      "eval_wer": 0.20595230442887447,
       "step": 3384
     },
     {
       "epoch": 47.08,
-      "learning_rate": 2.685211267605634e-05,
-      "loss": 0.3616,
+      "learning_rate": 1.423134328358209e-05,
+      "loss": 0.2809,
       "step": 3390
     },
     {
       "epoch": 47.22,
-      "learning_rate": 2.6781690140845073e-05,
-      "loss": 0.3455,
+      "learning_rate": 1.4194029850746268e-05,
+      "loss": 0.2767,
       "step": 3400
     },
     {
       "epoch": 47.36,
-      "learning_rate": 2.6711267605633805e-05,
-      "loss": 0.3323,
+      "learning_rate": 1.4156716417910449e-05,
+      "loss": 0.267,
       "step": 3410
     },
     {
       "epoch": 47.5,
-      "learning_rate": 2.6640845070422537e-05,
-      "loss": 0.3476,
+      "learning_rate": 1.4119402985074628e-05,
+      "loss": 0.2758,
       "step": 3420
     },
     {
       "epoch": 47.64,
-      "learning_rate": 2.657042253521127e-05,
-      "loss": 0.3346,
+      "learning_rate": 1.4082089552238805e-05,
+      "loss": 0.2654,
       "step": 3430
     },
     {
       "epoch": 47.78,
-      "learning_rate": 2.6500000000000004e-05,
-      "loss": 0.3439,
+      "learning_rate": 1.4044776119402986e-05,
+      "loss": 0.2781,
       "step": 3440
     },
     {
       "epoch": 47.91,
-      "learning_rate": 2.6429577464788736e-05,
-      "loss": 0.327,
+      "learning_rate": 1.4007462686567165e-05,
+      "loss": 0.2673,
       "step": 3450
     },
     {
       "epoch": 48.0,
-      "eval_loss": 0.49777674674987793,
-      "eval_runtime": 127.6593,
-      "eval_samples_per_second": 4.167,
-      "eval_steps_per_second": 0.525,
-      "eval_wer": 0.21893681301022047,
+      "eval_loss": 0.5044223070144653,
+      "eval_runtime": 131.7824,
+      "eval_samples_per_second": 4.037,
+      "eval_steps_per_second": 0.508,
+      "eval_wer": 0.1999742881018191,
       "step": 3456
     },
     {
       "epoch": 48.06,
-      "learning_rate": 2.6359154929577468e-05,
-      "loss": 0.3601,
+      "learning_rate": 1.3970149253731343e-05,
+      "loss": 0.2927,
       "step": 3460
     },
     {
       "epoch": 48.19,
-      "learning_rate": 2.62887323943662e-05,
-      "loss": 0.3493,
+      "learning_rate": 1.3932835820895524e-05,
+      "loss": 0.2854,
       "step": 3470
     },
     {
       "epoch": 48.33,
-      "learning_rate": 2.621830985915493e-05,
-      "loss": 0.3506,
+      "learning_rate": 1.3895522388059703e-05,
+      "loss": 0.284,
       "step": 3480
     },
     {
       "epoch": 48.47,
-      "learning_rate": 2.6147887323943667e-05,
-      "loss": 0.3293,
+      "learning_rate": 1.385820895522388e-05,
+      "loss": 0.2675,
       "step": 3490
     },
     {
       "epoch": 48.61,
-      "learning_rate": 2.60774647887324e-05,
-      "loss": 0.3398,
+      "learning_rate": 1.3820895522388061e-05,
+      "loss": 0.2719,
       "step": 3500
     },
     {
       "epoch": 48.75,
-      "learning_rate": 2.600704225352113e-05,
-      "loss": 0.3399,
+      "learning_rate": 1.3783582089552239e-05,
+      "loss": 0.2748,
       "step": 3510
     },
     {
       "epoch": 48.89,
-      "learning_rate": 2.5936619718309862e-05,
-      "loss": 0.3266,
+      "learning_rate": 1.3746268656716418e-05,
+      "loss": 0.2696,
       "step": 3520
     },
     {
       "epoch": 49.0,
-      "eval_loss": 0.5078500509262085,
-      "eval_runtime": 127.4612,
-      "eval_samples_per_second": 4.174,
-      "eval_steps_per_second": 0.526,
-      "eval_wer": 0.21874397377386384,
+      "eval_loss": 0.5187488794326782,
+      "eval_runtime": 131.1067,
+      "eval_samples_per_second": 4.058,
+      "eval_steps_per_second": 0.511,
+      "eval_wer": 0.1983030147200617,
       "step": 3528
     },
     {
       "epoch": 49.03,
-      "learning_rate": 2.5866197183098594e-05,
-      "loss": 0.3485,
+      "learning_rate": 1.3708955223880599e-05,
+      "loss": 0.2806,
       "step": 3530
     },
     {
       "epoch": 49.17,
-      "learning_rate": 2.5795774647887326e-05,
-      "loss": 0.3352,
+      "learning_rate": 1.3671641791044776e-05,
+      "loss": 0.2766,
       "step": 3540
     },
     {
       "epoch": 49.3,
-      "learning_rate": 2.572535211267606e-05,
-      "loss": 0.3424,
+      "learning_rate": 1.3634328358208955e-05,
+      "loss": 0.2821,
       "step": 3550
     },
     {
       "epoch": 49.44,
-      "learning_rate": 2.5654929577464793e-05,
-      "loss": 0.3218,
+      "learning_rate": 1.3597014925373136e-05,
+      "loss": 0.2673,
       "step": 3560
     },
     {
       "epoch": 49.58,
-      "learning_rate": 2.5584507042253525e-05,
-      "loss": 0.3407,
+      "learning_rate": 1.3559701492537314e-05,
+      "loss": 0.2794,
       "step": 3570
     },
     {
       "epoch": 49.72,
-      "learning_rate": 2.5514084507042257e-05,
-      "loss": 0.3429,
+      "learning_rate": 1.3522388059701493e-05,
+      "loss": 0.2745,
       "step": 3580
     },
     {
       "epoch": 49.86,
-      "learning_rate": 2.544366197183099e-05,
-      "loss": 0.333,
+      "learning_rate": 1.3485074626865674e-05,
+      "loss": 0.2664,
       "step": 3590
     },
     {
       "epoch": 50.0,
-      "learning_rate": 2.5373239436619717e-05,
-      "loss": 0.3353,
+      "learning_rate": 1.3447761194029851e-05,
+      "loss": 0.2724,
       "step": 3600
     },
     {
       "epoch": 50.0,
-      "eval_loss": 0.5302192568778992,
-      "eval_runtime": 129.7397,
-      "eval_samples_per_second": 4.101,
-      "eval_steps_per_second": 0.516,
-      "eval_wer": 0.22350067493732725,
+      "eval_loss": 0.5234389901161194,
+      "eval_runtime": 131.879,
+      "eval_samples_per_second": 4.034,
+      "eval_steps_per_second": 0.508,
+      "eval_wer": 0.20350967410169055,
       "step": 3600
     },
     {
       "epoch": 50.14,
-      "learning_rate": 2.530281690140845e-05,
-      "loss": 0.3433,
+      "learning_rate": 1.3410447761194032e-05,
+      "loss": 0.2814,
       "step": 3610
     },
     {
       "epoch": 50.28,
-      "learning_rate": 2.523239436619718e-05,
-      "loss": 0.3373,
+      "learning_rate": 1.3373134328358208e-05,
+      "loss": 0.267,
       "step": 3620
     },
     {
       "epoch": 50.42,
-      "learning_rate": 2.5161971830985913e-05,
-      "loss": 0.3393,
+      "learning_rate": 1.3335820895522388e-05,
+      "loss": 0.2681,
       "step": 3630
     },
     {
       "epoch": 50.55,
-      "learning_rate": 2.5091549295774648e-05,
-      "loss": 0.3473,
+      "learning_rate": 1.329850746268657e-05,
+      "loss": 0.2812,
       "step": 3640
     },
     {
       "epoch": 50.69,
-      "learning_rate": 2.502112676056338e-05,
-      "loss": 0.3399,
+      "learning_rate": 1.3261194029850745e-05,
+      "loss": 0.2709,
       "step": 3650
     },
     {
       "epoch": 50.83,
-      "learning_rate": 2.4950704225352115e-05,
-      "loss": 0.3421,
+      "learning_rate": 1.3223880597014926e-05,
+      "loss": 0.2719,
       "step": 3660
     },
     {
       "epoch": 50.97,
-      "learning_rate": 2.4880281690140847e-05,
-      "loss": 0.3384,
+      "learning_rate": 1.3186567164179107e-05,
+      "loss": 0.2736,
       "step": 3670
     },
     {
       "epoch": 51.0,
-      "eval_loss": 0.4837527275085449,
-      "eval_runtime": 128.8469,
-      "eval_samples_per_second": 4.129,
-      "eval_steps_per_second": 0.52,
-      "eval_wer": 0.21167320177412097,
+      "eval_loss": 0.5106495022773743,
+      "eval_runtime": 131.9308,
+      "eval_samples_per_second": 4.032,
+      "eval_steps_per_second": 0.508,
+      "eval_wer": 0.2016455614835765,
       "step": 3672
     },
     {
       "epoch": 51.11,
-      "learning_rate": 2.480985915492958e-05,
-      "loss": 0.3385,
+      "learning_rate": 1.3149253731343284e-05,
+      "loss": 0.2766,
       "step": 3680
     },
     {
       "epoch": 51.25,
-      "learning_rate": 2.473943661971831e-05,
-      "loss": 0.3374,
+      "learning_rate": 1.3111940298507463e-05,
+      "loss": 0.2771,
       "step": 3690
     },
     {
       "epoch": 51.39,
-      "learning_rate": 2.4669014084507046e-05,
-      "loss": 0.3305,
+      "learning_rate": 1.3074626865671644e-05,
+      "loss": 0.2754,
       "step": 3700
     },
     {
       "epoch": 51.53,
-      "learning_rate": 2.4598591549295778e-05,
-      "loss": 0.3462,
+      "learning_rate": 1.3037313432835822e-05,
+      "loss": 0.2849,
       "step": 3710
     },
     {
       "epoch": 51.66,
-      "learning_rate": 2.452816901408451e-05,
-      "loss": 0.3221,
+      "learning_rate": 1.3000000000000001e-05,
+      "loss": 0.2669,
       "step": 3720
     },
     {
       "epoch": 51.8,
-      "learning_rate": 2.4457746478873242e-05,
-      "loss": 0.3276,
+      "learning_rate": 1.2962686567164178e-05,
+      "loss": 0.2623,
       "step": 3730
     },
     {
       "epoch": 51.94,
-      "learning_rate": 2.4387323943661974e-05,
-      "loss": 0.3259,
+      "learning_rate": 1.292537313432836e-05,
+      "loss": 0.2662,
       "step": 3740
     },
     {
       "epoch": 52.0,
-      "eval_loss": 0.5022213459014893,
-      "eval_runtime": 129.1853,
-      "eval_samples_per_second": 4.118,
-      "eval_steps_per_second": 0.519,
-      "eval_wer": 0.21694414090120204,
+      "eval_loss": 0.5071024298667908,
+      "eval_runtime": 131.882,
+      "eval_samples_per_second": 4.034,
+      "eval_steps_per_second": 0.508,
+      "eval_wer": 0.20537378671980458,
       "step": 3744
     },
     {
       "epoch": 52.08,
-      "learning_rate": 2.4316901408450705e-05,
-      "loss": 0.3393,
+      "learning_rate": 1.2888059701492538e-05,
+      "loss": 0.2814,
       "step": 3750
     },
     {
       "epoch": 52.22,
-      "learning_rate": 2.4246478873239437e-05,
-      "loss": 0.3306,
+      "learning_rate": 1.2850746268656716e-05,
+      "loss": 0.2733,
       "step": 3760
     },
     {
       "epoch": 52.36,
-      "learning_rate": 2.417605633802817e-05,
-      "loss": 0.3342,
+      "learning_rate": 1.2813432835820897e-05,
+      "loss": 0.2679,
       "step": 3770
     },
     {
       "epoch": 52.5,
-      "learning_rate": 2.41056338028169e-05,
-      "loss": 0.3341,
+      "learning_rate": 1.2776119402985076e-05,
+      "loss": 0.2717,
       "step": 3780
     },
     {
       "epoch": 52.64,
-      "learning_rate": 2.4035211267605633e-05,
-      "loss": 0.3149,
+      "learning_rate": 1.2738805970149253e-05,
+      "loss": 0.2563,
       "step": 3790
     },
     {
       "epoch": 52.78,
-      "learning_rate": 2.3964788732394365e-05,
-      "loss": 0.3452,
+      "learning_rate": 1.2701492537313434e-05,
+      "loss": 0.2791,
       "step": 3800
     },
     {
       "epoch": 52.91,
-      "learning_rate": 2.38943661971831e-05,
-      "loss": 0.336,
+      "learning_rate": 1.2664179104477613e-05,
+      "loss": 0.2701,
       "step": 3810
     },
     {
       "epoch": 53.0,
-      "eval_loss": 0.4834235608577728,
-      "eval_runtime": 128.5245,
-      "eval_samples_per_second": 4.139,
-      "eval_steps_per_second": 0.521,
-      "eval_wer": 0.2214437230828566,
+      "eval_loss": 0.5253528356552124,
+      "eval_runtime": 131.2156,
+      "eval_samples_per_second": 4.054,
+      "eval_steps_per_second": 0.511,
+      "eval_wer": 0.2016455614835765,
       "step": 3816
     },
     {
       "epoch": 53.06,
-      "learning_rate": 2.3823943661971832e-05,
-      "loss": 0.3509,
+      "learning_rate": 1.262686567164179e-05,
+      "loss": 0.2935,
       "step": 3820
     },
     {
       "epoch": 53.19,
-      "learning_rate": 2.3753521126760564e-05,
-      "loss": 0.326,
+      "learning_rate": 1.2589552238805972e-05,
+      "loss": 0.2654,
       "step": 3830
     },
     {
       "epoch": 53.33,
-      "learning_rate": 2.3683098591549296e-05,
-      "loss": 0.3393,
+      "learning_rate": 1.255223880597015e-05,
+      "loss": 0.2768,
       "step": 3840
     },
     {
       "epoch": 53.47,
-      "learning_rate": 2.3612676056338028e-05,
-      "loss": 0.3369,
+      "learning_rate": 1.2514925373134328e-05,
+      "loss": 0.2779,
       "step": 3850
     },
     {
       "epoch": 53.61,
-      "learning_rate": 2.3542253521126763e-05,
-      "loss": 0.3317,
+      "learning_rate": 1.2477611940298509e-05,
+      "loss": 0.2743,
       "step": 3860
     },
     {
       "epoch": 53.75,
-      "learning_rate": 2.3471830985915495e-05,
-      "loss": 0.3275,
+      "learning_rate": 1.2440298507462687e-05,
+      "loss": 0.2719,
       "step": 3870
     },
     {
       "epoch": 53.89,
-      "learning_rate": 2.3401408450704227e-05,
-      "loss": 0.3123,
+      "learning_rate": 1.2402985074626866e-05,
+      "loss": 0.261,
       "step": 3880
     },
     {
       "epoch": 54.0,
-      "eval_loss": 0.5113204121589661,
-      "eval_runtime": 130.2117,
-      "eval_samples_per_second": 4.086,
-      "eval_steps_per_second": 0.515,
-      "eval_wer": 0.21642990293758438,
+      "eval_loss": 0.4967404007911682,
+      "eval_runtime": 131.9132,
+      "eval_samples_per_second": 4.033,
+      "eval_steps_per_second": 0.508,
+      "eval_wer": 0.19939577039274925,
       "step": 3888
     },
     {
       "epoch": 54.03,
-      "learning_rate": 2.333098591549296e-05,
-      "loss": 0.3572,
+      "learning_rate": 1.2365671641791045e-05,
+      "loss": 0.2909,
       "step": 3890
     },
     {
       "epoch": 54.17,
-      "learning_rate": 2.326056338028169e-05,
-      "loss": 0.3197,
+      "learning_rate": 1.2328358208955226e-05,
+      "loss": 0.2642,
       "step": 3900
     },
     {
       "epoch": 54.3,
-      "learning_rate": 2.3190140845070422e-05,
-      "loss": 0.3348,
+      "learning_rate": 1.2291044776119403e-05,
+      "loss": 0.2746,
       "step": 3910
     },
     {
       "epoch": 54.44,
-      "learning_rate": 2.3119718309859158e-05,
-      "loss": 0.3166,
+      "learning_rate": 1.2253731343283582e-05,
+      "loss": 0.2655,
       "step": 3920
     },
     {
       "epoch": 54.58,
-      "learning_rate": 2.304929577464789e-05,
-      "loss": 0.3311,
+      "learning_rate": 1.2216417910447763e-05,
+      "loss": 0.2722,
       "step": 3930
     },
     {
       "epoch": 54.72,
-      "learning_rate": 2.297887323943662e-05,
-      "loss": 0.3262,
+      "learning_rate": 1.217910447761194e-05,
+      "loss": 0.2721,
       "step": 3940
     },
     {
       "epoch": 54.86,
-      "learning_rate": 2.2908450704225353e-05,
-      "loss": 0.3358,
+      "learning_rate": 1.214179104477612e-05,
+      "loss": 0.2728,
       "step": 3950
     },
     {
       "epoch": 55.0,
-      "learning_rate": 2.2838028169014085e-05,
-      "loss": 0.3306,
+      "learning_rate": 1.2104477611940299e-05,
+      "loss": 0.2708,
       "step": 3960
     },
     {
       "epoch": 55.0,
-      "eval_loss": 0.47700902819633484,
-      "eval_runtime": 129.6952,
-      "eval_samples_per_second": 4.102,
-      "eval_steps_per_second": 0.517,
-      "eval_wer": 0.2111589638105033,
+      "eval_loss": 0.5263388156890869,
+      "eval_runtime": 131.5494,
+      "eval_samples_per_second": 4.044,
+      "eval_steps_per_second": 0.509,
+      "eval_wer": 0.19688886032011313,
       "step": 3960
     },
     {
       "epoch": 55.14,
-      "learning_rate": 2.276760563380282e-05,
-      "loss": 0.3319,
+      "learning_rate": 1.2067164179104478e-05,
+      "loss": 0.2684,
       "step": 3970
     },
     {
       "epoch": 55.28,
-      "learning_rate": 2.2697183098591552e-05,
-      "loss": 0.3313,
+      "learning_rate": 1.2029850746268657e-05,
+      "loss": 0.2699,
       "step": 3980
     },
     {
       "epoch": 55.42,
-      "learning_rate": 2.2626760563380284e-05,
-      "loss": 0.3325,
+      "learning_rate": 1.1992537313432836e-05,
+      "loss": 0.2796,
       "step": 3990
     },
     {
       "epoch": 55.55,
-      "learning_rate": 2.2556338028169016e-05,
-      "loss": 0.3255,
+      "learning_rate": 1.1955223880597016e-05,
+      "loss": 0.2666,
       "step": 4000
     },
     {
       "epoch": 55.69,
-      "learning_rate": 2.2485915492957748e-05,
-      "loss": 0.3299,
+      "learning_rate": 1.1917910447761195e-05,
+      "loss": 0.2702,
       "step": 4010
     },
     {
       "epoch": 55.83,
-      "learning_rate": 2.241549295774648e-05,
-      "loss": 0.3321,
+      "learning_rate": 1.1880597014925374e-05,
+      "loss": 0.2711,
       "step": 4020
     },
     {
       "epoch": 55.97,
-      "learning_rate": 2.2345070422535215e-05,
-      "loss": 0.3344,
+      "learning_rate": 1.1843283582089553e-05,
+      "loss": 0.2764,
       "step": 4030
     },
     {
       "epoch": 56.0,
-      "eval_loss": 0.4752146601676941,
-      "eval_runtime": 128.9657,
-      "eval_samples_per_second": 4.125,
-      "eval_steps_per_second": 0.52,
-      "eval_wer": 0.21302307642861734,
+      "eval_loss": 0.5201473236083984,
+      "eval_runtime": 131.8696,
+      "eval_samples_per_second": 4.034,
+      "eval_steps_per_second": 0.508,
+      "eval_wer": 0.19676030082920873,
       "step": 4032
     },
     {
       "epoch": 56.11,
-      "learning_rate": 2.2274647887323943e-05,
-      "loss": 0.3311,
+      "learning_rate": 1.1805970149253732e-05,
+      "loss": 0.272,
       "step": 4040
     },
     {
       "epoch": 56.25,
-      "learning_rate": 2.2204225352112675e-05,
-      "loss": 0.3314,
+      "learning_rate": 1.1768656716417911e-05,
+      "loss": 0.2624,
       "step": 4050
     },
     {
       "epoch": 56.39,
-      "learning_rate": 2.2133802816901407e-05,
-      "loss": 0.3128,
+      "learning_rate": 1.173134328358209e-05,
+      "loss": 0.2566,
       "step": 4060
     },
     {
       "epoch": 56.53,
-      "learning_rate": 2.2063380281690142e-05,
-      "loss": 0.3471,
+      "learning_rate": 1.1694029850746268e-05,
+      "loss": 0.2789,
       "step": 4070
     },
     {
       "epoch": 56.66,
-      "learning_rate": 2.1992957746478874e-05,
-      "loss": 0.3281,
+      "learning_rate": 1.1656716417910449e-05,
+      "loss": 0.267,
       "step": 4080
     },
     {
       "epoch": 56.8,
-      "learning_rate": 2.1922535211267606e-05,
-      "loss": 0.3371,
+      "learning_rate": 1.1619402985074628e-05,
+      "loss": 0.2711,
       "step": 4090
     },
     {
       "epoch": 56.94,
-      "learning_rate": 2.1852112676056338e-05,
-      "loss": 0.3279,
+      "learning_rate": 1.1582089552238806e-05,
+      "loss": 0.268,
       "step": 4100
     },
     {
       "epoch": 57.0,
-      "eval_loss": 0.4953095316886902,
-      "eval_runtime": 129.8414,
-      "eval_samples_per_second": 4.097,
-      "eval_steps_per_second": 0.516,
-      "eval_wer": 0.21231599922864305,
+      "eval_loss": 0.5023506879806519,
+      "eval_runtime": 131.2011,
+      "eval_samples_per_second": 4.055,
+      "eval_steps_per_second": 0.511,
+      "eval_wer": 0.19785305650189625,
       "step": 4104
     },
     {
       "epoch": 57.08,
-      "learning_rate": 2.178169014084507e-05,
-      "loss": 0.3253,
+      "learning_rate": 1.1544776119402986e-05,
+      "loss": 0.2639,
       "step": 4110
     },
     {
       "epoch": 57.22,
-      "learning_rate": 2.1711267605633802e-05,
-      "loss": 0.33,
+      "learning_rate": 1.1507462686567166e-05,
+      "loss": 0.2725,
       "step": 4120
     },
     {
       "epoch": 57.36,
-      "learning_rate": 2.1640845070422537e-05,
-      "loss": 0.3386,
+      "learning_rate": 1.1470149253731343e-05,
+      "loss": 0.271,
       "step": 4130
     },
     {
       "epoch": 57.5,
-      "learning_rate": 2.157042253521127e-05,
-      "loss": 0.3367,
+      "learning_rate": 1.1432835820895522e-05,
+      "loss": 0.2778,
       "step": 4140
     },
     {
       "epoch": 57.64,
-      "learning_rate": 2.15e-05,
-      "loss": 0.3266,
+      "learning_rate": 1.1395522388059703e-05,
+      "loss": 0.2673,
       "step": 4150
     },
     {
       "epoch": 57.78,
-      "learning_rate": 2.1429577464788733e-05,
-      "loss": 0.3329,
+      "learning_rate": 1.135820895522388e-05,
+      "loss": 0.2665,
       "step": 4160
     },
     {
       "epoch": 57.91,
-      "learning_rate": 2.1359154929577465e-05,
-      "loss": 0.3067,
+      "learning_rate": 1.132089552238806e-05,
+      "loss": 0.2536,
       "step": 4170
     },
     {
       "epoch": 58.0,
-      "eval_loss": 0.4822605550289154,
-      "eval_runtime": 127.9027,
-      "eval_samples_per_second": 4.159,
-      "eval_steps_per_second": 0.524,
-      "eval_wer": 0.2175226586102719,
+      "eval_loss": 0.5224213600158691,
+      "eval_runtime": 131.5544,
+      "eval_samples_per_second": 4.044,
+      "eval_steps_per_second": 0.509,
+      "eval_wer": 0.19701741981101756,
       "step": 4176
     },
     {
       "epoch": 58.06,
-      "learning_rate": 2.12887323943662e-05,
-      "loss": 0.3345,
+      "learning_rate": 1.1283582089552239e-05,
+      "loss": 0.2727,
       "step": 4180
     },
     {
       "epoch": 58.19,
-      "learning_rate": 2.1218309859154932e-05,
-      "loss": 0.3205,
+      "learning_rate": 1.124626865671642e-05,
+      "loss": 0.2596,
       "step": 4190
     },
     {
       "epoch": 58.33,
-      "learning_rate": 2.1147887323943664e-05,
-      "loss": 0.3362,
+      "learning_rate": 1.1208955223880597e-05,
+      "loss": 0.2671,
       "step": 4200
     },
     {
       "epoch": 58.47,
-      "learning_rate": 2.1077464788732395e-05,
-      "loss": 0.323,
+      "learning_rate": 1.1171641791044776e-05,
+      "loss": 0.2667,
       "step": 4210
     },
     {
       "epoch": 58.61,
-      "learning_rate": 2.1007042253521127e-05,
-      "loss": 0.3101,
+      "learning_rate": 1.1134328358208957e-05,
+      "loss": 0.2648,
       "step": 4220
     },
     {
       "epoch": 58.75,
-      "learning_rate": 2.093661971830986e-05,
-      "loss": 0.3338,
+      "learning_rate": 1.1097014925373135e-05,
+      "loss": 0.2828,
       "step": 4230
     },
     {
       "epoch": 58.89,
-      "learning_rate": 2.0866197183098594e-05,
-      "loss": 0.3131,
+      "learning_rate": 1.1059701492537314e-05,
+      "loss": 0.2625,
       "step": 4240
     },
     {
       "epoch": 59.0,
-      "eval_loss": 0.48572954535484314,
-      "eval_runtime": 127.6235,
-      "eval_samples_per_second": 4.169,
-      "eval_steps_per_second": 0.525,
-      "eval_wer": 0.21366587388313943,
+      "eval_loss": 0.5316326022148132,
+      "eval_runtime": 131.6958,
+      "eval_samples_per_second": 4.04,
+      "eval_steps_per_second": 0.509,
+      "eval_wer": 0.1999742881018191,
       "step": 4248
     },
     {
       "epoch": 59.03,
-      "learning_rate": 2.0795774647887326e-05,
-      "loss": 0.3319,
+      "learning_rate": 1.1022388059701493e-05,
+      "loss": 0.2725,
       "step": 4250
     },
     {
       "epoch": 59.17,
-      "learning_rate": 2.0725352112676058e-05,
-      "loss": 0.3056,
+      "learning_rate": 1.0985074626865672e-05,
+      "loss": 0.2541,
       "step": 4260
     },
     {
       "epoch": 59.3,
-      "learning_rate": 2.065492957746479e-05,
-      "loss": 0.3301,
+      "learning_rate": 1.0947761194029851e-05,
+      "loss": 0.2703,
       "step": 4270
     },
     {
       "epoch": 59.44,
-      "learning_rate": 2.0584507042253522e-05,
-      "loss": 0.333,
+      "learning_rate": 1.091044776119403e-05,
+      "loss": 0.2725,
       "step": 4280
     },
     {
       "epoch": 59.58,
-      "learning_rate": 2.0514084507042254e-05,
-      "loss": 0.3269,
+      "learning_rate": 1.087313432835821e-05,
+      "loss": 0.273,
       "step": 4290
     },
     {
       "epoch": 59.72,
-      "learning_rate": 2.044366197183099e-05,
-      "loss": 0.3316,
+      "learning_rate": 1.0835820895522389e-05,
+      "loss": 0.2793,
       "step": 4300
     },
     {
       "epoch": 59.86,
-      "learning_rate": 2.037323943661972e-05,
-      "loss": 0.3096,
+      "learning_rate": 1.0798507462686568e-05,
+      "loss": 0.257,
       "step": 4310
     },
     {
       "epoch": 60.0,
-      "learning_rate": 2.0302816901408453e-05,
-      "loss": 0.3178,
+      "learning_rate": 1.0761194029850747e-05,
+      "loss": 0.2621,
       "step": 4320
     },
     {
       "epoch": 60.0,
-      "eval_loss": 0.49693578481674194,
-      "eval_runtime": 128.0482,
-      "eval_samples_per_second": 4.155,
-      "eval_steps_per_second": 0.523,
-      "eval_wer": 0.21540142701034903,
+      "eval_loss": 0.5173215270042419,
+      "eval_runtime": 131.579,
+      "eval_samples_per_second": 4.043,
+      "eval_steps_per_second": 0.509,
+      "eval_wer": 0.19965288937455808,
       "step": 4320
     },
     {
       "epoch": 60.14,
-      "learning_rate": 2.0232394366197185e-05,
-      "loss": 0.3303,
+      "learning_rate": 1.0723880597014926e-05,
+      "loss": 0.272,
       "step": 4330
     },
     {
       "epoch": 60.28,
-      "learning_rate": 2.0161971830985917e-05,
-      "loss": 0.3226,
+      "learning_rate": 1.0686567164179105e-05,
+      "loss": 0.2669,
       "step": 4340
     },
     {
       "epoch": 60.42,
-      "learning_rate": 2.009154929577465e-05,
-      "loss": 0.3207,
+      "learning_rate": 1.0649253731343284e-05,
+      "loss": 0.2703,
       "step": 4350
     },
     {
       "epoch": 60.55,
-      "learning_rate": 2.002112676056338e-05,
-      "loss": 0.3228,
+      "learning_rate": 1.0611940298507462e-05,
+      "loss": 0.2633,
       "step": 4360
     },
     {
       "epoch": 60.69,
-      "learning_rate": 1.9950704225352112e-05,
-      "loss": 0.3258,
+      "learning_rate": 1.0574626865671643e-05,
+      "loss": 0.2657,
       "step": 4370
     },
     {
       "epoch": 60.83,
-      "learning_rate": 1.9880281690140844e-05,
-      "loss": 0.3223,
+      "learning_rate": 1.0537313432835822e-05,
+      "loss": 0.2654,
       "step": 4380
     },
     {
       "epoch": 60.97,
-      "learning_rate": 1.9809859154929576e-05,
-      "loss": 0.3306,
+      "learning_rate": 1.05e-05,
+      "loss": 0.2673,
       "step": 4390
     },
     {
       "epoch": 61.0,
-      "eval_loss": 0.4816165566444397,
-      "eval_runtime": 128.2312,
-      "eval_samples_per_second": 4.149,
-      "eval_steps_per_second": 0.522,
-      "eval_wer": 0.21655846242848878,
+      "eval_loss": 0.5116785764694214,
+      "eval_runtime": 131.1081,
+      "eval_samples_per_second": 4.058,
+      "eval_steps_per_second": 0.511,
+      "eval_wer": 0.20280259690171626,
       "step": 4392
     },
     {
       "epoch": 61.11,
-      "learning_rate": 1.973943661971831e-05,
-      "loss": 0.3323,
+      "learning_rate": 1.046268656716418e-05,
+      "loss": 0.2678,
       "step": 4400
     },
     {
       "epoch": 61.25,
-      "learning_rate": 1.9669014084507043e-05,
-      "loss": 0.3327,
+      "learning_rate": 1.042537313432836e-05,
+      "loss": 0.2706,
       "step": 4410
     },
     {
       "epoch": 61.39,
-      "learning_rate": 1.9598591549295775e-05,
-      "loss": 0.3154,
+      "learning_rate": 1.0388059701492537e-05,
+      "loss": 0.2641,
       "step": 4420
     },
     {
       "epoch": 61.53,
-      "learning_rate": 1.9528169014084507e-05,
-      "loss": 0.3179,
+      "learning_rate": 1.0350746268656716e-05,
+      "loss": 0.2596,
       "step": 4430
     },
     {
       "epoch": 61.66,
-      "learning_rate": 1.945774647887324e-05,
-      "loss": 0.305,
+      "learning_rate": 1.0313432835820897e-05,
+      "loss": 0.2562,
       "step": 4440
     },
     {
       "epoch": 61.8,
-      "learning_rate": 1.9387323943661974e-05,
-      "loss": 0.331,
+      "learning_rate": 1.0276119402985074e-05,
+      "loss": 0.2802,
       "step": 4450
     },
     {
       "epoch": 61.94,
-      "learning_rate": 1.9316901408450706e-05,
-      "loss": 0.3152,
+      "learning_rate": 1.0238805970149254e-05,
+      "loss": 0.2723,
       "step": 4460
     },
     {
       "epoch": 62.0,
-      "eval_loss": 0.4694591760635376,
-      "eval_runtime": 128.0603,
-      "eval_samples_per_second": 4.154,
-      "eval_steps_per_second": 0.523,
-      "eval_wer": 0.21064472584688565,
+      "eval_loss": 0.5285342335700989,
+      "eval_runtime": 131.8603,
+      "eval_samples_per_second": 4.035,
+      "eval_steps_per_second": 0.508,
+      "eval_wer": 0.1999742881018191,
       "step": 4464
     },
     {
       "epoch": 62.08,
-      "learning_rate": 1.9246478873239438e-05,
-      "loss": 0.3376,
+      "learning_rate": 1.0201492537313434e-05,
+      "loss": 0.2846,
       "step": 4470
     },
     {
       "epoch": 62.22,
-      "learning_rate": 1.917605633802817e-05,
-      "loss": 0.3176,
+      "learning_rate": 1.0164179104477614e-05,
+      "loss": 0.2758,
       "step": 4480
     },
     {
       "epoch": 62.36,
-      "learning_rate": 1.91056338028169e-05,
-      "loss": 0.3249,
+      "learning_rate": 1.0126865671641791e-05,
+      "loss": 0.277,
       "step": 4490
     },
     {
       "epoch": 62.5,
-      "learning_rate": 1.9035211267605633e-05,
-      "loss": 0.3214,
+      "learning_rate": 1.008955223880597e-05,
+      "loss": 0.2761,
       "step": 4500
     },
     {
       "epoch": 62.64,
-      "learning_rate": 1.896478873239437e-05,
-      "loss": 0.3153,
+      "learning_rate": 1.0052238805970151e-05,
+      "loss": 0.2684,
       "step": 4510
     },
     {
       "epoch": 62.78,
-      "learning_rate": 1.88943661971831e-05,
-      "loss": 0.3308,
+      "learning_rate": 1.0014925373134328e-05,
+      "loss": 0.2817,
       "step": 4520
     },
     {
       "epoch": 62.91,
-      "learning_rate": 1.8823943661971832e-05,
-      "loss": 0.3119,
+      "learning_rate": 9.977611940298508e-06,
+      "loss": 0.2691,
       "step": 4530
     },
     {
       "epoch": 63.0,
-      "eval_loss": 0.5111212730407715,
-      "eval_runtime": 128.7341,
-      "eval_samples_per_second": 4.133,
-      "eval_steps_per_second": 0.52,
-      "eval_wer": 0.21058044610143345,
+      "eval_loss": 0.5328835248947144,
+      "eval_runtime": 131.3705,
+      "eval_samples_per_second": 4.05,
+      "eval_steps_per_second": 0.51,
+      "eval_wer": 0.19611750337468664,
       "step": 4536
     },
     {
       "epoch": 63.06,
-      "learning_rate": 1.8753521126760564e-05,
-      "loss": 0.3313,
+      "learning_rate": 9.940298507462687e-06,
+      "loss": 0.2824,
       "step": 4540
     },
     {
       "epoch": 63.19,
-      "learning_rate": 1.8683098591549296e-05,
-      "loss": 0.3123,
+      "learning_rate": 9.902985074626866e-06,
+      "loss": 0.2772,
       "step": 4550
     },
     {
       "epoch": 63.33,
-      "learning_rate": 1.861267605633803e-05,
-      "loss": 0.3206,
+      "learning_rate": 9.865671641791045e-06,
+      "loss": 0.2708,
       "step": 4560
     },
     {
       "epoch": 63.47,
-      "learning_rate": 1.8542253521126763e-05,
-      "loss": 0.3139,
+      "learning_rate": 9.828358208955224e-06,
+      "loss": 0.2679,
       "step": 4570
     },
     {
       "epoch": 63.61,
-      "learning_rate": 1.8471830985915495e-05,
-      "loss": 0.3169,
+      "learning_rate": 9.791044776119403e-06,
+      "loss": 0.268,
       "step": 4580
     },
     {
       "epoch": 63.75,
-      "learning_rate": 1.8401408450704227e-05,
-      "loss": 0.3155,
+      "learning_rate": 9.753731343283583e-06,
+      "loss": 0.2705,
       "step": 4590
     },
     {
       "epoch": 63.89,
-      "learning_rate": 1.833098591549296e-05,
-      "loss": 0.3124,
+      "learning_rate": 9.716417910447762e-06,
+      "loss": 0.2668,
       "step": 4600
     },
     {
       "epoch": 64.0,
-      "eval_loss": 0.501674473285675,
-      "eval_runtime": 127.9707,
-      "eval_samples_per_second": 4.157,
-      "eval_steps_per_second": 0.524,
-      "eval_wer": 0.2118660410104776,
+      "eval_loss": 0.5160889625549316,
+      "eval_runtime": 131.2903,
+      "eval_samples_per_second": 4.052,
+      "eval_steps_per_second": 0.51,
+      "eval_wer": 0.19643890210194767,
       "step": 4608
     },
     {
       "epoch": 64.03,
-      "learning_rate": 1.826056338028169e-05,
-      "loss": 0.3236,
+      "learning_rate": 9.679104477611941e-06,
+      "loss": 0.2774,
       "step": 4610
     },
     {
       "epoch": 64.17,
-      "learning_rate": 1.8190140845070426e-05,
-      "loss": 0.3077,
+      "learning_rate": 9.64179104477612e-06,
+      "loss": 0.2688,
       "step": 4620
     },
     {
       "epoch": 64.3,
-      "learning_rate": 1.8119718309859155e-05,
-      "loss": 0.321,
+      "learning_rate": 9.6044776119403e-06,
+      "loss": 0.2773,
       "step": 4630
     },
     {
       "epoch": 64.44,
-      "learning_rate": 1.8049295774647886e-05,
-      "loss": 0.3019,
+      "learning_rate": 9.567164179104478e-06,
+      "loss": 0.2633,
       "step": 4640
     },
     {
       "epoch": 64.58,
-      "learning_rate": 1.797887323943662e-05,
-      "loss": 0.3181,
+      "learning_rate": 9.529850746268656e-06,
+      "loss": 0.2735,
       "step": 4650
     },
     {
       "epoch": 64.72,
-      "learning_rate": 1.790845070422535e-05,
-      "loss": 0.3113,
+      "learning_rate": 9.492537313432837e-06,
+      "loss": 0.2683,
       "step": 4660
     },
     {
       "epoch": 64.86,
-      "learning_rate": 1.7838028169014085e-05,
-      "loss": 0.3191,
+      "learning_rate": 9.455223880597016e-06,
+      "loss": 0.2719,
       "step": 4670
     },
     {
       "epoch": 65.0,
-      "learning_rate": 1.7767605633802817e-05,
-      "loss": 0.317,
+      "learning_rate": 9.417910447761193e-06,
+      "loss": 0.27,
       "step": 4680
     },
     {
       "epoch": 65.0,
-      "eval_loss": 0.5071455836296082,
-      "eval_runtime": 129.2281,
-      "eval_samples_per_second": 4.117,
-      "eval_steps_per_second": 0.518,
-      "eval_wer": 0.21649418268303658,
+      "eval_loss": 0.5026705861091614,
+      "eval_runtime": 131.4228,
+      "eval_samples_per_second": 4.048,
+      "eval_steps_per_second": 0.51,
+      "eval_wer": 0.19695314006556533,
       "step": 4680
     },
     {
       "epoch": 65.14,
-      "learning_rate": 1.769718309859155e-05,
-      "loss": 0.3253,
+      "learning_rate": 9.380597014925374e-06,
+      "loss": 0.2786,
       "step": 4690
     },
     {
       "epoch": 65.28,
-      "learning_rate": 1.762676056338028e-05,
-      "loss": 0.3272,
+      "learning_rate": 9.343283582089553e-06,
+      "loss": 0.2731,
       "step": 4700
     },
     {
       "epoch": 65.42,
-      "learning_rate": 1.7556338028169013e-05,
-      "loss": 0.3143,
+      "learning_rate": 9.30597014925373e-06,
+      "loss": 0.2707,
       "step": 4710
     },
     {
       "epoch": 65.55,
-      "learning_rate": 1.7485915492957748e-05,
-      "loss": 0.3108,
+      "learning_rate": 9.26865671641791e-06,
+      "loss": 0.2638,
       "step": 4720
     },
     {
       "epoch": 65.69,
-      "learning_rate": 1.741549295774648e-05,
-      "loss": 0.3075,
+      "learning_rate": 9.23134328358209e-06,
+      "loss": 0.2605,
       "step": 4730
     },
     {
       "epoch": 65.83,
-      "learning_rate": 1.7345070422535212e-05,
-      "loss": 0.3167,
+      "learning_rate": 9.194029850746268e-06,
+      "loss": 0.2736,
       "step": 4740
     },
     {
       "epoch": 65.97,
-      "learning_rate": 1.7274647887323944e-05,
-      "loss": 0.2934,
+      "learning_rate": 9.156716417910447e-06,
+      "loss": 0.257,
       "step": 4750
     },
     {
       "epoch": 66.0,
-      "eval_loss": 0.4819434583187103,
-      "eval_runtime": 128.1216,
-      "eval_samples_per_second": 4.152,
-      "eval_steps_per_second": 0.523,
-      "eval_wer": 0.2168155814102976,
+      "eval_loss": 0.5016449689865112,
+      "eval_runtime": 132.0547,
+      "eval_samples_per_second": 4.029,
+      "eval_steps_per_second": 0.507,
+      "eval_wer": 0.19965288937455808,
       "step": 4752
     },
     {
       "epoch": 66.11,
-      "learning_rate": 1.7204225352112676e-05,
-      "loss": 0.3263,
+      "learning_rate": 9.119402985074628e-06,
+      "loss": 0.277,
       "step": 4760
     },
     {
       "epoch": 66.25,
-      "learning_rate": 1.7133802816901408e-05,
-      "loss": 0.3156,
+      "learning_rate": 9.082089552238807e-06,
+      "loss": 0.2734,
       "step": 4770
     },
     {
       "epoch": 66.39,
-      "learning_rate": 1.7063380281690143e-05,
-      "loss": 0.3101,
+      "learning_rate": 9.044776119402985e-06,
+      "loss": 0.2679,
       "step": 4780
     },
     {
       "epoch": 66.53,
-      "learning_rate": 1.6992957746478875e-05,
-      "loss": 0.3222,
+      "learning_rate": 9.007462686567164e-06,
+      "loss": 0.2755,
       "step": 4790
     },
     {
       "epoch": 66.66,
-      "learning_rate": 1.6922535211267607e-05,
-      "loss": 0.3104,
+      "learning_rate": 8.970149253731345e-06,
+      "loss": 0.2686,
       "step": 4800
     },
     {
       "epoch": 66.8,
-      "learning_rate": 1.685211267605634e-05,
-      "loss": 0.3104,
+      "learning_rate": 8.932835820895522e-06,
+      "loss": 0.2651,
       "step": 4810
     },
     {
       "epoch": 66.94,
-      "learning_rate": 1.678169014084507e-05,
-      "loss": 0.3103,
+      "learning_rate": 8.895522388059702e-06,
+      "loss": 0.2711,
       "step": 4820
     },
     {
       "epoch": 67.0,
-      "eval_loss": 0.4943528175354004,
-      "eval_runtime": 128.9828,
-      "eval_samples_per_second": 4.125,
-      "eval_steps_per_second": 0.519,
-      "eval_wer": 0.21430867133766152,
+      "eval_loss": 0.5059924721717834,
+      "eval_runtime": 131.6036,
+      "eval_samples_per_second": 4.042,
+      "eval_steps_per_second": 0.509,
+      "eval_wer": 0.19714597930192196,
       "step": 4824
     },
     {
       "epoch": 67.08,
-      "learning_rate": 1.6711267605633806e-05,
-      "loss": 0.3063,
+      "learning_rate": 8.85820895522388e-06,
+      "loss": 0.2678,
       "step": 4830
     },
     {
       "epoch": 67.22,
-      "learning_rate": 1.6640845070422538e-05,
-      "loss": 0.3118,
+      "learning_rate": 8.82089552238806e-06,
+      "loss": 0.2676,
       "step": 4840
     },
     {
       "epoch": 67.36,
-      "learning_rate": 1.657042253521127e-05,
-      "loss": 0.2975,
+      "learning_rate": 8.783582089552239e-06,
+      "loss": 0.2569,
       "step": 4850
     },
     {
       "epoch": 67.5,
-      "learning_rate": 1.65e-05,
-      "loss": 0.3126,
+      "learning_rate": 8.746268656716418e-06,
+      "loss": 0.2681,
       "step": 4860
     },
     {
       "epoch": 67.64,
-      "learning_rate": 1.6429577464788733e-05,
-      "loss": 0.3022,
+      "learning_rate": 8.708955223880597e-06,
+      "loss": 0.2664,
       "step": 4870
     },
     {
       "epoch": 67.78,
-      "learning_rate": 1.6359154929577465e-05,
-      "loss": 0.3155,
+      "learning_rate": 8.671641791044776e-06,
+      "loss": 0.2725,
       "step": 4880
     },
     {
       "epoch": 67.91,
-      "learning_rate": 1.62887323943662e-05,
-      "loss": 0.3078,
+      "learning_rate": 8.634328358208956e-06,
+      "loss": 0.269,
       "step": 4890
     },
     {
       "epoch": 68.0,
-      "eval_loss": 0.5003769993782043,
-      "eval_runtime": 127.2601,
-      "eval_samples_per_second": 4.18,
-      "eval_steps_per_second": 0.526,
-      "eval_wer": 0.2110304043195989,
+      "eval_loss": 0.5219683051109314,
+      "eval_runtime": 131.9262,
+      "eval_samples_per_second": 4.033,
+      "eval_steps_per_second": 0.508,
+      "eval_wer": 0.1983030147200617,
       "step": 4896
     },
     {
       "epoch": 68.06,
-      "learning_rate": 1.6218309859154932e-05,
-      "loss": 0.3161,
+      "learning_rate": 8.597014925373135e-06,
+      "loss": 0.2747,
       "step": 4900
     },
     {
       "epoch": 68.19,
-      "learning_rate": 1.6147887323943664e-05,
-      "loss": 0.3094,
+      "learning_rate": 8.559701492537314e-06,
+      "loss": 0.2724,
       "step": 4910
     },
     {
       "epoch": 68.33,
-      "learning_rate": 1.6077464788732396e-05,
-      "loss": 0.3191,
+      "learning_rate": 8.522388059701493e-06,
+      "loss": 0.2773,
       "step": 4920
     },
     {
       "epoch": 68.47,
-      "learning_rate": 1.6007042253521128e-05,
-      "loss": 0.3246,
+      "learning_rate": 8.485074626865672e-06,
+      "loss": 0.282,
       "step": 4930
     },
     {
       "epoch": 68.61,
-      "learning_rate": 1.593661971830986e-05,
-      "loss": 0.3078,
+      "learning_rate": 8.447761194029851e-06,
+      "loss": 0.2674,
       "step": 4940
     },
     {
       "epoch": 68.75,
-      "learning_rate": 1.586619718309859e-05,
-      "loss": 0.3044,
+      "learning_rate": 8.41044776119403e-06,
+      "loss": 0.2611,
       "step": 4950
     },
     {
       "epoch": 68.89,
-      "learning_rate": 1.5795774647887323e-05,
-      "loss": 0.3084,
+      "learning_rate": 8.37313432835821e-06,
+      "loss": 0.2659,
       "step": 4960
     },
     {
       "epoch": 69.0,
-      "eval_loss": 0.49981018900871277,
-      "eval_runtime": 127.4871,
-      "eval_samples_per_second": 4.173,
-      "eval_steps_per_second": 0.526,
-      "eval_wer": 0.21475862955582695,
+      "eval_loss": 0.498529314994812,
+      "eval_runtime": 132.3204,
+      "eval_samples_per_second": 4.021,
+      "eval_steps_per_second": 0.506,
+      "eval_wer": 0.19856013370187053,
       "step": 4968
     },
     {
       "epoch": 69.03,
-      "learning_rate": 1.5725352112676055e-05,
-      "loss": 0.3225,
+      "learning_rate": 8.335820895522387e-06,
+      "loss": 0.2752,
       "step": 4970
     },
     {
       "epoch": 69.17,
-      "learning_rate": 1.5654929577464787e-05,
-      "loss": 0.3125,
+      "learning_rate": 8.298507462686568e-06,
+      "loss": 0.2671,
       "step": 4980
     },
     {
       "epoch": 69.3,
-      "learning_rate": 1.5584507042253522e-05,
-      "loss": 0.3189,
+      "learning_rate": 8.261194029850747e-06,
+      "loss": 0.2704,
       "step": 4990
     },
     {
       "epoch": 69.44,
-      "learning_rate": 1.5514084507042254e-05,
-      "loss": 0.3132,
+      "learning_rate": 8.223880597014925e-06,
+      "loss": 0.265,
       "step": 5000
     },
     {
       "epoch": 69.58,
-      "learning_rate": 1.5443661971830986e-05,
-      "loss": 0.319,
+      "learning_rate": 8.186567164179104e-06,
+      "loss": 0.2757,
       "step": 5010
     },
     {
       "epoch": 69.72,
-      "learning_rate": 1.5373239436619718e-05,
-      "loss": 0.3092,
+      "learning_rate": 8.149253731343285e-06,
+      "loss": 0.2688,
       "step": 5020
     },
     {
       "epoch": 69.86,
-      "learning_rate": 1.530281690140845e-05,
-      "loss": 0.3108,
+      "learning_rate": 8.111940298507462e-06,
+      "loss": 0.2666,
       "step": 5030
     },
     {
       "epoch": 70.0,
-      "learning_rate": 1.5232394366197183e-05,
-      "loss": 0.3147,
+      "learning_rate": 8.074626865671641e-06,
+      "loss": 0.2669,
       "step": 5040
     },
     {
       "epoch": 70.0,
-      "eval_loss": 0.48848283290863037,
-      "eval_runtime": 127.875,
-      "eval_samples_per_second": 4.16,
-      "eval_steps_per_second": 0.524,
-      "eval_wer": 0.20987336890145916,
+      "eval_loss": 0.514457643032074,
+      "eval_runtime": 131.6706,
+      "eval_samples_per_second": 4.04,
+      "eval_steps_per_second": 0.509,
+      "eval_wer": 0.1958603843928778,
       "step": 5040
     },
     {
       "epoch": 70.14,
-      "learning_rate": 1.5161971830985915e-05,
-      "loss": 0.312,
+      "learning_rate": 8.037313432835822e-06,
+      "loss": 0.2716,
       "step": 5050
     },
     {
       "epoch": 70.28,
-      "learning_rate": 1.5091549295774649e-05,
-      "loss": 0.3199,
+      "learning_rate": 8.000000000000001e-06,
+      "loss": 0.2729,
       "step": 5060
     },
     {
       "epoch": 70.42,
-      "learning_rate": 1.502112676056338e-05,
-      "loss": 0.3058,
+      "learning_rate": 7.962686567164179e-06,
+      "loss": 0.259,
       "step": 5070
     },
     {
       "epoch": 70.55,
-      "learning_rate": 1.4950704225352113e-05,
-      "loss": 0.3022,
+      "learning_rate": 7.925373134328358e-06,
+      "loss": 0.2651,
       "step": 5080
     },
     {
       "epoch": 70.69,
-      "learning_rate": 1.4880281690140846e-05,
-      "loss": 0.3043,
+      "learning_rate": 7.888059701492539e-06,
+      "loss": 0.2657,
       "step": 5090
     },
     {
       "epoch": 70.83,
-      "learning_rate": 1.4809859154929578e-05,
-      "loss": 0.3183,
+      "learning_rate": 7.850746268656716e-06,
+      "loss": 0.2735,
       "step": 5100
     },
     {
       "epoch": 70.97,
-      "learning_rate": 1.4739436619718312e-05,
-      "loss": 0.3081,
+      "learning_rate": 7.813432835820895e-06,
+      "loss": 0.2685,
       "step": 5110
     },
     {
       "epoch": 71.0,
-      "eval_loss": 0.5062239766120911,
-      "eval_runtime": 127.6812,
-      "eval_samples_per_second": 4.167,
-      "eval_steps_per_second": 0.525,
-      "eval_wer": 0.2096162499196503,
+      "eval_loss": 0.5079246163368225,
+      "eval_runtime": 131.456,
+      "eval_samples_per_second": 4.047,
+      "eval_steps_per_second": 0.51,
+      "eval_wer": 0.19560326541106898,
       "step": 5112
     },
     {
       "epoch": 71.11,
-      "learning_rate": 1.4669014084507044e-05,
-      "loss": 0.3103,
+      "learning_rate": 7.776119402985076e-06,
+      "loss": 0.2732,
       "step": 5120
     },
     {
       "epoch": 71.25,
-      "learning_rate": 1.4598591549295775e-05,
-      "loss": 0.2963,
+      "learning_rate": 7.738805970149254e-06,
+      "loss": 0.2568,
       "step": 5130
     },
     {
       "epoch": 71.39,
-      "learning_rate": 1.4528169014084509e-05,
-      "loss": 0.2978,
+      "learning_rate": 7.701492537313433e-06,
+      "loss": 0.2552,
       "step": 5140
     },
     {
       "epoch": 71.53,
-      "learning_rate": 1.4457746478873241e-05,
-      "loss": 0.3107,
+      "learning_rate": 7.664179104477612e-06,
+      "loss": 0.2634,
       "step": 5150
     },
     {
       "epoch": 71.66,
-      "learning_rate": 1.4387323943661973e-05,
-      "loss": 0.3103,
+      "learning_rate": 7.626865671641792e-06,
+      "loss": 0.2663,
       "step": 5160
     },
     {
       "epoch": 71.8,
-      "learning_rate": 1.4316901408450706e-05,
-      "loss": 0.3113,
+      "learning_rate": 7.58955223880597e-06,
+      "loss": 0.2679,
       "step": 5170
     },
     {
       "epoch": 71.94,
-      "learning_rate": 1.4246478873239438e-05,
-      "loss": 0.3046,
+      "learning_rate": 7.5522388059701495e-06,
+      "loss": 0.2669,
       "step": 5180
     },
     {
       "epoch": 72.0,
-      "eval_loss": 0.47900620102882385,
-      "eval_runtime": 127.6914,
-      "eval_samples_per_second": 4.166,
-      "eval_steps_per_second": 0.525,
-      "eval_wer": 0.21244455871954748,
+      "eval_loss": 0.49233612418174744,
+      "eval_runtime": 131.3756,
+      "eval_samples_per_second": 4.049,
+      "eval_steps_per_second": 0.51,
+      "eval_wer": 0.19361059330205052,
       "step": 5184
     },
     {
       "epoch": 72.08,
-      "learning_rate": 1.417605633802817e-05,
-      "loss": 0.3169,
+      "learning_rate": 7.514925373134328e-06,
+      "loss": 0.2722,
       "step": 5190
     },
     {
       "epoch": 72.22,
-      "learning_rate": 1.4105633802816904e-05,
-      "loss": 0.3112,
+      "learning_rate": 7.477611940298508e-06,
+      "loss": 0.2698,
       "step": 5200
     },
     {
       "epoch": 72.36,
-      "learning_rate": 1.4035211267605636e-05,
-      "loss": 0.2977,
+      "learning_rate": 7.440298507462687e-06,
+      "loss": 0.2527,
       "step": 5210
     },
     {
       "epoch": 72.5,
-      "learning_rate": 1.3964788732394367e-05,
-      "loss": 0.312,
+      "learning_rate": 7.402985074626865e-06,
+      "loss": 0.2658,
       "step": 5220
     },
     {
       "epoch": 72.64,
-      "learning_rate": 1.3894366197183098e-05,
-      "loss": 0.2976,
+      "learning_rate": 7.365671641791045e-06,
+      "loss": 0.2623,
       "step": 5230
     },
     {
       "epoch": 72.78,
-      "learning_rate": 1.3823943661971831e-05,
-      "loss": 0.2993,
+      "learning_rate": 7.3283582089552245e-06,
+      "loss": 0.2574,
       "step": 5240
     },
     {
       "epoch": 72.91,
-      "learning_rate": 1.3753521126760563e-05,
-      "loss": 0.2993,
+      "learning_rate": 7.291044776119403e-06,
+      "loss": 0.2586,
       "step": 5250
     },
     {
       "epoch": 73.0,
-      "eval_loss": 0.5045033097267151,
-      "eval_runtime": 127.5139,
-      "eval_samples_per_second": 4.172,
-      "eval_steps_per_second": 0.525,
-      "eval_wer": 0.2096162499196503,
+      "eval_loss": 0.4957040250301361,
+      "eval_runtime": 131.9056,
+      "eval_samples_per_second": 4.033,
+      "eval_steps_per_second": 0.508,
+      "eval_wer": 0.19528186668380793,
       "step": 5256
     },
     {
       "epoch": 73.06,
-      "learning_rate": 1.3683098591549295e-05,
-      "loss": 0.3093,
+      "learning_rate": 7.253731343283582e-06,
+      "loss": 0.2703,
       "step": 5260
     },
     {
       "epoch": 73.19,
-      "learning_rate": 1.3612676056338028e-05,
-      "loss": 0.2961,
+      "learning_rate": 7.216417910447762e-06,
+      "loss": 0.2618,
       "step": 5270
     },
     {
       "epoch": 73.33,
-      "learning_rate": 1.354225352112676e-05,
-      "loss": 0.3014,
+      "learning_rate": 7.17910447761194e-06,
+      "loss": 0.2643,
       "step": 5280
     },
     {
       "epoch": 73.47,
-      "learning_rate": 1.3471830985915492e-05,
-      "loss": 0.3113,
+      "learning_rate": 7.1417910447761194e-06,
+      "loss": 0.2735,
       "step": 5290
     },
     {
       "epoch": 73.61,
-      "learning_rate": 1.3401408450704226e-05,
-      "loss": 0.2948,
+      "learning_rate": 7.1044776119402994e-06,
+      "loss": 0.2562,
       "step": 5300
     },
     {
       "epoch": 73.75,
-      "learning_rate": 1.3330985915492958e-05,
-      "loss": 0.3022,
+      "learning_rate": 7.067164179104479e-06,
+      "loss": 0.2597,
       "step": 5310
     },
     {
       "epoch": 73.89,
-      "learning_rate": 1.326056338028169e-05,
-      "loss": 0.2844,
+      "learning_rate": 7.029850746268657e-06,
+      "loss": 0.2486,
       "step": 5320
     },
     {
       "epoch": 74.0,
-      "eval_loss": 0.4865418076515198,
-      "eval_runtime": 127.6726,
-      "eval_samples_per_second": 4.167,
-      "eval_steps_per_second": 0.525,
-      "eval_wer": 0.20601658417432667,
+      "eval_loss": 0.4974006712436676,
+      "eval_runtime": 132.0022,
+      "eval_samples_per_second": 4.03,
+      "eval_steps_per_second": 0.508,
+      "eval_wer": 0.19316063508388506,
       "step": 5328
     },
     {
       "epoch": 74.03,
-      "learning_rate": 1.3190140845070423e-05,
-      "loss": 0.3137,
+      "learning_rate": 6.992537313432835e-06,
+      "loss": 0.2732,
       "step": 5330
     },
     {
       "epoch": 74.17,
-      "learning_rate": 1.3119718309859155e-05,
-      "loss": 0.2849,
+      "learning_rate": 6.955223880597016e-06,
+      "loss": 0.2455,
       "step": 5340
     },
     {
       "epoch": 74.3,
-      "learning_rate": 1.3049295774647889e-05,
-      "loss": 0.301,
+      "learning_rate": 6.917910447761194e-06,
+      "loss": 0.2517,
       "step": 5350
     },
     {
       "epoch": 74.44,
-      "learning_rate": 1.297887323943662e-05,
-      "loss": 0.3007,
+      "learning_rate": 6.8805970149253735e-06,
+      "loss": 0.2618,
       "step": 5360
     },
     {
       "epoch": 74.58,
-      "learning_rate": 1.2908450704225352e-05,
-      "loss": 0.3074,
+      "learning_rate": 6.843283582089552e-06,
+      "loss": 0.2667,
       "step": 5370
     },
     {
       "epoch": 74.72,
-      "learning_rate": 1.2838028169014086e-05,
-      "loss": 0.306,
+      "learning_rate": 6.805970149253732e-06,
+      "loss": 0.2635,
       "step": 5380
     },
     {
       "epoch": 74.86,
-      "learning_rate": 1.2767605633802818e-05,
-      "loss": 0.3103,
+      "learning_rate": 6.768656716417911e-06,
+      "loss": 0.2667,
       "step": 5390
     },
     {
       "epoch": 75.0,
-      "learning_rate": 1.269718309859155e-05,
-      "loss": 0.3061,
+      "learning_rate": 6.731343283582089e-06,
+      "loss": 0.2611,
       "step": 5400
     },
     {
       "epoch": 75.0,
-      "eval_loss": 0.49969518184661865,
-      "eval_runtime": 126.5881,
-      "eval_samples_per_second": 4.203,
-      "eval_steps_per_second": 0.529,
-      "eval_wer": 0.2102590473741724,
+      "eval_loss": 0.5044746398925781,
+      "eval_runtime": 131.9706,
+      "eval_samples_per_second": 4.031,
+      "eval_steps_per_second": 0.508,
+      "eval_wer": 0.1948961882110947,
       "step": 5400
     },
     {
       "epoch": 75.14,
-      "learning_rate": 1.2626760563380283e-05,
-      "loss": 0.3207,
+      "learning_rate": 6.694029850746269e-06,
+      "loss": 0.2813,
       "step": 5410
     },
     {
       "epoch": 75.28,
-      "learning_rate": 1.2556338028169015e-05,
-      "loss": 0.3112,
+      "learning_rate": 6.6567164179104485e-06,
+      "loss": 0.2708,
       "step": 5420
     },
     {
       "epoch": 75.42,
-      "learning_rate": 1.2485915492957747e-05,
-      "loss": 0.2981,
+      "learning_rate": 6.619402985074627e-06,
+      "loss": 0.2558,
       "step": 5430
     },
     {
       "epoch": 75.55,
-      "learning_rate": 1.241549295774648e-05,
-      "loss": 0.3007,
+      "learning_rate": 6.582089552238806e-06,
+      "loss": 0.2593,
       "step": 5440
     },
     {
       "epoch": 75.69,
-      "learning_rate": 1.234507042253521e-05,
-      "loss": 0.2967,
+      "learning_rate": 6.544776119402986e-06,
+      "loss": 0.2532,
       "step": 5450
     },
     {
       "epoch": 75.83,
-      "learning_rate": 1.2274647887323944e-05,
-      "loss": 0.2986,
+      "learning_rate": 6.507462686567164e-06,
+      "loss": 0.2601,
       "step": 5460
     },
     {
       "epoch": 75.97,
-      "learning_rate": 1.2204225352112676e-05,
-      "loss": 0.305,
+      "learning_rate": 6.4701492537313434e-06,
+      "loss": 0.264,
       "step": 5470
     },
     {
       "epoch": 76.0,
-      "eval_loss": 0.5118193626403809,
-      "eval_runtime": 125.4914,
-      "eval_samples_per_second": 4.239,
-      "eval_steps_per_second": 0.534,
-      "eval_wer": 0.21199460050138202,
+      "eval_loss": 0.5025977492332458,
+      "eval_runtime": 132.0676,
+      "eval_samples_per_second": 4.028,
+      "eval_steps_per_second": 0.507,
+      "eval_wer": 0.1965674615928521,
       "step": 5472
     },
     {
       "epoch": 76.11,
-      "learning_rate": 1.2133802816901408e-05,
-      "loss": 0.3029,
+      "learning_rate": 6.4328358208955234e-06,
+      "loss": 0.256,
       "step": 5480
     },
     {
       "epoch": 76.25,
-      "learning_rate": 1.2063380281690142e-05,
-      "loss": 0.3084,
+      "learning_rate": 6.395522388059702e-06,
+      "loss": 0.2661,
       "step": 5490
     },
     {
       "epoch": 76.39,
-      "learning_rate": 1.1992957746478873e-05,
-      "loss": 0.3031,
+      "learning_rate": 6.358208955223881e-06,
+      "loss": 0.2603,
       "step": 5500
     },
     {
       "epoch": 76.53,
-      "learning_rate": 1.1922535211267605e-05,
-      "loss": 0.3015,
+      "learning_rate": 6.320895522388059e-06,
+      "loss": 0.2574,
       "step": 5510
     },
     {
       "epoch": 76.66,
-      "learning_rate": 1.1852112676056339e-05,
-      "loss": 0.2959,
+      "learning_rate": 6.283582089552239e-06,
+      "loss": 0.2572,
       "step": 5520
     },
     {
       "epoch": 76.8,
-      "learning_rate": 1.178169014084507e-05,
-      "loss": 0.3022,
+      "learning_rate": 6.246268656716418e-06,
+      "loss": 0.2587,
       "step": 5530
     },
     {
       "epoch": 76.94,
-      "learning_rate": 1.1711267605633804e-05,
-      "loss": 0.3015,
+      "learning_rate": 6.208955223880597e-06,
+      "loss": 0.2653,
       "step": 5540
     },
     {
       "epoch": 77.0,
-      "eval_loss": 0.489970326423645,
-      "eval_runtime": 125.7413,
-      "eval_samples_per_second": 4.231,
-      "eval_steps_per_second": 0.533,
-      "eval_wer": 0.2095519701741981,
+      "eval_loss": 0.5044174790382385,
+      "eval_runtime": 133.5324,
+      "eval_samples_per_second": 3.984,
+      "eval_steps_per_second": 0.502,
+      "eval_wer": 0.19361059330205052,
       "step": 5544
     },
     {
       "epoch": 77.08,
-      "learning_rate": 1.1640845070422536e-05,
-      "loss": 0.2974,
+      "learning_rate": 6.171641791044777e-06,
+      "loss": 0.2586,
       "step": 5550
     },
     {
       "epoch": 77.22,
-      "learning_rate": 1.1570422535211268e-05,
-      "loss": 0.3004,
+      "learning_rate": 6.134328358208956e-06,
+      "loss": 0.2628,
       "step": 5560
     },
     {
       "epoch": 77.36,
-      "learning_rate": 1.1500000000000002e-05,
-      "loss": 0.2996,
+      "learning_rate": 6.097014925373134e-06,
+      "loss": 0.2637,
       "step": 5570
     },
     {
       "epoch": 77.5,
-      "learning_rate": 1.1429577464788734e-05,
-      "loss": 0.3014,
+      "learning_rate": 6.059701492537314e-06,
+      "loss": 0.2594,
       "step": 5580
     },
     {
       "epoch": 77.64,
-      "learning_rate": 1.1359154929577465e-05,
-      "loss": 0.2959,
+      "learning_rate": 6.0223880597014925e-06,
+      "loss": 0.2567,
       "step": 5590
     },
     {
       "epoch": 77.78,
-      "learning_rate": 1.1288732394366197e-05,
-      "loss": 0.3052,
+      "learning_rate": 5.9850746268656725e-06,
+      "loss": 0.2684,
       "step": 5600
     },
     {
       "epoch": 77.91,
-      "learning_rate": 1.121830985915493e-05,
-      "loss": 0.2969,
+      "learning_rate": 5.947761194029851e-06,
+      "loss": 0.2579,
       "step": 5610
     },
     {
       "epoch": 78.0,
-      "eval_loss": 0.49838951230049133,
-      "eval_runtime": 125.5517,
-      "eval_samples_per_second": 4.237,
-      "eval_steps_per_second": 0.534,
-      "eval_wer": 0.21064472584688565,
+      "eval_loss": 0.4963833689689636,
+      "eval_runtime": 131.9034,
+      "eval_samples_per_second": 4.033,
+      "eval_steps_per_second": 0.508,
+      "eval_wer": 0.1966960210837565,
       "step": 5616
     },
     {
       "epoch": 78.06,
-      "learning_rate": 1.1147887323943663e-05,
-      "loss": 0.3042,
+      "learning_rate": 5.91044776119403e-06,
+      "loss": 0.2654,
       "step": 5620
     },
     {
       "epoch": 78.19,
-      "learning_rate": 1.1077464788732395e-05,
-      "loss": 0.2955,
+      "learning_rate": 5.873134328358209e-06,
+      "loss": 0.2565,
       "step": 5630
     },
     {
       "epoch": 78.33,
-      "learning_rate": 1.1007042253521127e-05,
-      "loss": 0.3023,
+      "learning_rate": 5.835820895522388e-06,
+      "loss": 0.2647,
       "step": 5640
     },
     {
       "epoch": 78.47,
-      "learning_rate": 1.093661971830986e-05,
-      "loss": 0.2917,
+      "learning_rate": 5.7985074626865674e-06,
+      "loss": 0.2561,
       "step": 5650
     },
     {
       "epoch": 78.61,
-      "learning_rate": 1.0866197183098592e-05,
-      "loss": 0.2865,
+      "learning_rate": 5.761194029850747e-06,
+      "loss": 0.2495,
       "step": 5660
     },
     {
       "epoch": 78.75,
-      "learning_rate": 1.0795774647887324e-05,
-      "loss": 0.2969,
+      "learning_rate": 5.723880597014926e-06,
+      "loss": 0.2562,
       "step": 5670
     },
     {
       "epoch": 78.89,
-      "learning_rate": 1.0725352112676057e-05,
-      "loss": 0.2988,
+      "learning_rate": 5.686567164179105e-06,
+      "loss": 0.26,
       "step": 5680
     },
     {
       "epoch": 79.0,
-      "eval_loss": 0.494792103767395,
-      "eval_runtime": 125.4092,
-      "eval_samples_per_second": 4.242,
-      "eval_steps_per_second": 0.534,
-      "eval_wer": 0.20653082213794433,
+      "eval_loss": 0.4918227195739746,
+      "eval_runtime": 132.2622,
+      "eval_samples_per_second": 4.022,
+      "eval_steps_per_second": 0.507,
+      "eval_wer": 0.19373915279295495,
       "step": 5688
     },
     {
       "epoch": 79.03,
-      "learning_rate": 1.065492957746479e-05,
-      "loss": 0.3085,
+      "learning_rate": 5.649253731343284e-06,
+      "loss": 0.269,
       "step": 5690
     },
     {
       "epoch": 79.17,
-      "learning_rate": 1.0584507042253521e-05,
-      "loss": 0.2926,
+      "learning_rate": 5.611940298507462e-06,
+      "loss": 0.2585,
       "step": 5700
     },
     {
       "epoch": 79.3,
-      "learning_rate": 1.0514084507042255e-05,
-      "loss": 0.3023,
+      "learning_rate": 5.574626865671642e-06,
+      "loss": 0.2609,
       "step": 5710
     },
     {
       "epoch": 79.44,
-      "learning_rate": 1.0443661971830987e-05,
-      "loss": 0.2905,
+      "learning_rate": 5.537313432835821e-06,
+      "loss": 0.256,
       "step": 5720
     },
     {
       "epoch": 79.58,
-      "learning_rate": 1.037323943661972e-05,
-      "loss": 0.3016,
+      "learning_rate": 5.500000000000001e-06,
+      "loss": 0.2615,
       "step": 5730
     },
     {
       "epoch": 79.72,
-      "learning_rate": 1.030281690140845e-05,
-      "loss": 0.301,
+      "learning_rate": 5.46268656716418e-06,
+      "loss": 0.2626,
       "step": 5740
     },
     {
       "epoch": 79.86,
-      "learning_rate": 1.0232394366197182e-05,
-      "loss": 0.2903,
+      "learning_rate": 5.425373134328358e-06,
+      "loss": 0.2545,
       "step": 5750
     },
     {
       "epoch": 80.0,
-      "learning_rate": 1.0161971830985916e-05,
-      "loss": 0.3006,
+      "learning_rate": 5.388059701492538e-06,
+      "loss": 0.2634,
       "step": 5760
     },
     {
       "epoch": 80.0,
-      "eval_loss": 0.48496711254119873,
-      "eval_runtime": 125.2817,
-      "eval_samples_per_second": 4.246,
-      "eval_steps_per_second": 0.535,
-      "eval_wer": 0.20318827537442952,
+      "eval_loss": 0.4841861128807068,
+      "eval_runtime": 131.8698,
+      "eval_samples_per_second": 4.034,
+      "eval_steps_per_second": 0.508,
+      "eval_wer": 0.19277495661117183,
       "step": 5760
     },
     {
       "epoch": 80.14,
-      "learning_rate": 1.0091549295774648e-05,
-      "loss": 0.2919,
+      "learning_rate": 5.3507462686567165e-06,
+      "loss": 0.2575,
       "step": 5770
     },
     {
       "epoch": 80.28,
-      "learning_rate": 1.0028169014084506e-05,
-      "loss": 0.2934,
+      "learning_rate": 5.313432835820896e-06,
+      "loss": 0.2567,
       "step": 5780
     },
     {
       "epoch": 80.42,
-      "learning_rate": 9.95774647887324e-06,
-      "loss": 0.2944,
+      "learning_rate": 5.276119402985075e-06,
+      "loss": 0.2599,
       "step": 5790
     },
     {
       "epoch": 80.55,
-      "learning_rate": 9.887323943661972e-06,
-      "loss": 0.2834,
+      "learning_rate": 5.238805970149254e-06,
+      "loss": 0.246,
       "step": 5800
     },
     {
       "epoch": 80.69,
-      "learning_rate": 9.816901408450704e-06,
-      "loss": 0.2948,
+      "learning_rate": 5.201492537313433e-06,
+      "loss": 0.2546,
       "step": 5810
     },
     {
       "epoch": 80.83,
-      "learning_rate": 9.746478873239437e-06,
-      "loss": 0.3089,
+      "learning_rate": 5.164179104477612e-06,
+      "loss": 0.2675,
       "step": 5820
     },
     {
       "epoch": 80.97,
-      "learning_rate": 9.676056338028169e-06,
-      "loss": 0.2982,
+      "learning_rate": 5.1268656716417914e-06,
+      "loss": 0.2598,
       "step": 5830
     },
     {
       "epoch": 81.0,
-      "eval_loss": 0.49331343173980713,
-      "eval_runtime": 125.9952,
-      "eval_samples_per_second": 4.222,
-      "eval_steps_per_second": 0.532,
-      "eval_wer": 0.2046667095198303,
+      "eval_loss": 0.5017235279083252,
+      "eval_runtime": 132.1658,
+      "eval_samples_per_second": 4.025,
+      "eval_steps_per_second": 0.507,
+      "eval_wer": 0.19463906922928584,
       "step": 5832
     },
     {
       "epoch": 81.11,
-      "learning_rate": 9.605633802816901e-06,
-      "loss": 0.3012,
+      "learning_rate": 5.089552238805971e-06,
+      "loss": 0.2551,
       "step": 5840
     },
     {
       "epoch": 81.25,
-      "learning_rate": 9.535211267605635e-06,
-      "loss": 0.3046,
+      "learning_rate": 5.05223880597015e-06,
+      "loss": 0.2648,
       "step": 5850
     },
     {
       "epoch": 81.39,
-      "learning_rate": 9.464788732394366e-06,
-      "loss": 0.2831,
+      "learning_rate": 5.014925373134328e-06,
+      "loss": 0.2447,
       "step": 5860
     },
     {
       "epoch": 81.53,
-      "learning_rate": 9.3943661971831e-06,
-      "loss": 0.2947,
+      "learning_rate": 4.977611940298508e-06,
+      "loss": 0.2561,
       "step": 5870
     },
     {
       "epoch": 81.66,
-      "learning_rate": 9.323943661971832e-06,
-      "loss": 0.3004,
+      "learning_rate": 4.940298507462686e-06,
+      "loss": 0.261,
       "step": 5880
     },
     {
       "epoch": 81.8,
-      "learning_rate": 9.253521126760564e-06,
-      "loss": 0.3017,
+      "learning_rate": 4.902985074626866e-06,
+      "loss": 0.2653,
       "step": 5890
     },
     {
       "epoch": 81.94,
-      "learning_rate": 9.183098591549297e-06,
-      "loss": 0.2938,
+      "learning_rate": 4.865671641791045e-06,
+      "loss": 0.2571,
       "step": 5900
     },
     {
       "epoch": 82.0,
-      "eval_loss": 0.4996909201145172,
-      "eval_runtime": 126.3183,
-      "eval_samples_per_second": 4.212,
-      "eval_steps_per_second": 0.53,
-      "eval_wer": 0.20524522722890018,
+      "eval_loss": 0.4997570514678955,
+      "eval_runtime": 131.7552,
+      "eval_samples_per_second": 4.038,
+      "eval_steps_per_second": 0.509,
+      "eval_wer": 0.1983672944655139,
       "step": 5904
     },
     {
       "epoch": 82.08,
-      "learning_rate": 9.11267605633803e-06,
-      "loss": 0.2938,
+      "learning_rate": 4.828358208955224e-06,
+      "loss": 0.2628,
       "step": 5910
     },
     {
       "epoch": 82.22,
-      "learning_rate": 9.042253521126761e-06,
-      "loss": 0.2973,
+      "learning_rate": 4.791044776119404e-06,
+      "loss": 0.2625,
       "step": 5920
     },
     {
       "epoch": 82.36,
-      "learning_rate": 8.971830985915493e-06,
-      "loss": 0.2763,
+      "learning_rate": 4.753731343283582e-06,
+      "loss": 0.2396,
       "step": 5930
     },
     {
       "epoch": 82.5,
-      "learning_rate": 8.901408450704225e-06,
-      "loss": 0.307,
+      "learning_rate": 4.716417910447761e-06,
+      "loss": 0.2698,
       "step": 5940
     },
     {
       "epoch": 82.64,
-      "learning_rate": 8.830985915492958e-06,
-      "loss": 0.2967,
+      "learning_rate": 4.6791044776119405e-06,
+      "loss": 0.2571,
       "step": 5950
     },
     {
       "epoch": 82.78,
-      "learning_rate": 8.76056338028169e-06,
-      "loss": 0.2951,
+      "learning_rate": 4.64179104477612e-06,
+      "loss": 0.2592,
       "step": 5960
     },
     {
       "epoch": 82.91,
-      "learning_rate": 8.690140845070422e-06,
-      "loss": 0.2903,
+      "learning_rate": 4.604477611940299e-06,
+      "loss": 0.2564,
       "step": 5970
     },
     {
       "epoch": 83.0,
-      "eval_loss": 0.49193274974823,
-      "eval_runtime": 125.3571,
-      "eval_samples_per_second": 4.244,
-      "eval_steps_per_second": 0.534,
-      "eval_wer": 0.20344539435623835,
+      "eval_loss": 0.4961981475353241,
+      "eval_runtime": 132.3751,
+      "eval_samples_per_second": 4.019,
+      "eval_steps_per_second": 0.506,
+      "eval_wer": 0.19393199202931155,
       "step": 5976
     },
     {
       "epoch": 83.06,
-      "learning_rate": 8.619718309859156e-06,
-      "loss": 0.3021,
+      "learning_rate": 4.567164179104478e-06,
+      "loss": 0.2661,
       "step": 5980
     },
     {
       "epoch": 83.19,
-      "learning_rate": 8.549295774647888e-06,
-      "loss": 0.294,
+      "learning_rate": 4.529850746268656e-06,
+      "loss": 0.2602,
       "step": 5990
     },
     {
       "epoch": 83.33,
-      "learning_rate": 8.47887323943662e-06,
-      "loss": 0.2897,
+      "learning_rate": 4.492537313432836e-06,
+      "loss": 0.2515,
       "step": 6000
     },
     {
       "epoch": 83.47,
-      "learning_rate": 8.408450704225353e-06,
-      "loss": 0.2934,
+      "learning_rate": 4.4552238805970154e-06,
+      "loss": 0.2607,
       "step": 6010
     },
     {
       "epoch": 83.61,
-      "learning_rate": 8.338028169014085e-06,
-      "loss": 0.2776,
+      "learning_rate": 4.417910447761195e-06,
+      "loss": 0.2457,
       "step": 6020
     },
     {
       "epoch": 83.75,
-      "learning_rate": 8.267605633802817e-06,
-      "loss": 0.2944,
+      "learning_rate": 4.380597014925374e-06,
+      "loss": 0.2613,
       "step": 6030
     },
     {
       "epoch": 83.89,
-      "learning_rate": 8.19718309859155e-06,
-      "loss": 0.297,
+      "learning_rate": 4.343283582089552e-06,
+      "loss": 0.2606,
       "step": 6040
     },
     {
       "epoch": 84.0,
-      "eval_loss": 0.4807397127151489,
-      "eval_runtime": 125.5219,
-      "eval_samples_per_second": 4.238,
-      "eval_steps_per_second": 0.534,
-      "eval_wer": 0.20260975766535966,
+      "eval_loss": 0.49531611800193787,
+      "eval_runtime": 131.4182,
+      "eval_samples_per_second": 4.048,
+      "eval_steps_per_second": 0.51,
+      "eval_wer": 0.19155364144757986,
       "step": 6048
     },
     {
       "epoch": 84.03,
-      "learning_rate": 8.126760563380282e-06,
-      "loss": 0.3071,
+      "learning_rate": 4.305970149253732e-06,
+      "loss": 0.2655,
       "step": 6050
     },
     {
       "epoch": 84.17,
-      "learning_rate": 8.056338028169016e-06,
-      "loss": 0.2797,
+      "learning_rate": 4.26865671641791e-06,
+      "loss": 0.2488,
       "step": 6060
     },
     {
       "epoch": 84.3,
-      "learning_rate": 7.985915492957746e-06,
-      "loss": 0.2976,
+      "learning_rate": 4.2313432835820895e-06,
+      "loss": 0.258,
       "step": 6070
     },
     {
       "epoch": 84.44,
-      "learning_rate": 7.915492957746478e-06,
-      "loss": 0.3014,
+      "learning_rate": 4.194029850746269e-06,
+      "loss": 0.2644,
       "step": 6080
     },
     {
       "epoch": 84.58,
-      "learning_rate": 7.845070422535211e-06,
-      "loss": 0.2899,
+      "learning_rate": 4.156716417910448e-06,
+      "loss": 0.2573,
       "step": 6090
     },
     {
       "epoch": 84.72,
-      "learning_rate": 7.774647887323943e-06,
-      "loss": 0.2781,
+      "learning_rate": 4.119402985074627e-06,
+      "loss": 0.243,
       "step": 6100
     },
     {
       "epoch": 84.86,
-      "learning_rate": 7.704225352112677e-06,
-      "loss": 0.2898,
+      "learning_rate": 4.082089552238806e-06,
+      "loss": 0.2581,
       "step": 6110
     },
     {
       "epoch": 85.0,
-      "learning_rate": 7.633802816901409e-06,
-      "loss": 0.2955,
+      "learning_rate": 4.044776119402985e-06,
+      "loss": 0.2587,
       "step": 6120
     },
     {
       "epoch": 85.0,
-      "eval_loss": 0.4953136146068573,
-      "eval_runtime": 125.5494,
-      "eval_samples_per_second": 4.237,
-      "eval_steps_per_second": 0.534,
-      "eval_wer": 0.20479526901073472,
+      "eval_loss": 0.4982677400112152,
+      "eval_runtime": 132.0294,
+      "eval_samples_per_second": 4.029,
+      "eval_steps_per_second": 0.507,
+      "eval_wer": 0.19463906922928584,
       "step": 6120
     },
     {
       "epoch": 85.14,
-      "learning_rate": 7.5633802816901415e-06,
-      "loss": 0.2777,
+      "learning_rate": 4.0074626865671645e-06,
+      "loss": 0.2455,
       "step": 6130
     },
     {
       "epoch": 85.28,
-      "learning_rate": 7.492957746478873e-06,
-      "loss": 0.2856,
+      "learning_rate": 3.970149253731344e-06,
+      "loss": 0.2532,
       "step": 6140
     },
     {
       "epoch": 85.42,
-      "learning_rate": 7.422535211267606e-06,
-      "loss": 0.2983,
+      "learning_rate": 3.932835820895522e-06,
+      "loss": 0.2608,
       "step": 6150
     },
     {
       "epoch": 85.55,
-      "learning_rate": 7.352112676056339e-06,
-      "loss": 0.2938,
+      "learning_rate": 3.895522388059702e-06,
+      "loss": 0.2604,
       "step": 6160
     },
     {
       "epoch": 85.69,
-      "learning_rate": 7.281690140845071e-06,
-      "loss": 0.2858,
+      "learning_rate": 3.85820895522388e-06,
+      "loss": 0.2489,
       "step": 6170
     },
     {
       "epoch": 85.83,
-      "learning_rate": 7.211267605633803e-06,
-      "loss": 0.2948,
+      "learning_rate": 3.82089552238806e-06,
+      "loss": 0.2594,
       "step": 6180
     },
     {
       "epoch": 85.97,
-      "learning_rate": 7.140845070422536e-06,
-      "loss": 0.2847,
+      "learning_rate": 3.7835820895522394e-06,
+      "loss": 0.2499,
       "step": 6190
     },
     {
       "epoch": 86.0,
-      "eval_loss": 0.4950926899909973,
-      "eval_runtime": 126.5682,
-      "eval_samples_per_second": 4.203,
-      "eval_steps_per_second": 0.529,
-      "eval_wer": 0.2021597994471942,
+      "eval_loss": 0.5027219653129578,
+      "eval_runtime": 131.6359,
+      "eval_samples_per_second": 4.041,
+      "eval_steps_per_second": 0.509,
+      "eval_wer": 0.19534614642926015,
       "step": 6192
     },
     {
       "epoch": 86.11,
-      "learning_rate": 7.070422535211269e-06,
-      "loss": 0.287,
+      "learning_rate": 3.746268656716418e-06,
+      "loss": 0.2499,
       "step": 6200
     },
     {
       "epoch": 86.25,
-      "learning_rate": 7.000000000000001e-06,
-      "loss": 0.3001,
+      "learning_rate": 3.7089552238805973e-06,
+      "loss": 0.2627,
       "step": 6210
     },
     {
       "epoch": 86.39,
-      "learning_rate": 6.929577464788732e-06,
-      "loss": 0.2815,
+      "learning_rate": 3.671641791044776e-06,
+      "loss": 0.2471,
       "step": 6220
     },
     {
       "epoch": 86.53,
-      "learning_rate": 6.8591549295774645e-06,
-      "loss": 0.2909,
+      "learning_rate": 3.6343283582089557e-06,
+      "loss": 0.2538,
       "step": 6230
     },
     {
       "epoch": 86.66,
-      "learning_rate": 6.788732394366197e-06,
-      "loss": 0.2766,
+      "learning_rate": 3.5970149253731344e-06,
+      "loss": 0.2432,
       "step": 6240
     },
     {
       "epoch": 86.8,
-      "learning_rate": 6.71830985915493e-06,
-      "loss": 0.2977,
+      "learning_rate": 3.5597014925373135e-06,
+      "loss": 0.2625,
       "step": 6250
     },
     {
       "epoch": 86.94,
-      "learning_rate": 6.647887323943662e-06,
-      "loss": 0.2891,
+      "learning_rate": 3.5223880597014923e-06,
+      "loss": 0.253,
       "step": 6260
     },
     {
       "epoch": 87.0,
-      "eval_loss": 0.494158536195755,
-      "eval_runtime": 124.8943,
-      "eval_samples_per_second": 4.26,
-      "eval_steps_per_second": 0.536,
-      "eval_wer": 0.20357395384714277,
+      "eval_loss": 0.5000377297401428,
+      "eval_runtime": 132.268,
+      "eval_samples_per_second": 4.022,
+      "eval_steps_per_second": 0.507,
+      "eval_wer": 0.19515330719290352,
       "step": 6264
     },
     {
       "epoch": 87.08,
-      "learning_rate": 6.5774647887323945e-06,
-      "loss": 0.3011,
+      "learning_rate": 3.485074626865672e-06,
+      "loss": 0.2605,
       "step": 6270
     },
     {
       "epoch": 87.22,
-      "learning_rate": 6.507042253521127e-06,
-      "loss": 0.2975,
+      "learning_rate": 3.4477611940298506e-06,
+      "loss": 0.2625,
       "step": 6280
     },
     {
       "epoch": 87.36,
-      "learning_rate": 6.436619718309859e-06,
-      "loss": 0.2863,
+      "learning_rate": 3.41044776119403e-06,
+      "loss": 0.2449,
       "step": 6290
     },
     {
       "epoch": 87.5,
-      "learning_rate": 6.366197183098592e-06,
-      "loss": 0.2871,
+      "learning_rate": 3.3731343283582093e-06,
+      "loss": 0.2529,
       "step": 6300
     },
     {
       "epoch": 87.64,
-      "learning_rate": 6.295774647887325e-06,
-      "loss": 0.274,
+      "learning_rate": 3.335820895522388e-06,
+      "loss": 0.2388,
       "step": 6310
     },
     {
       "epoch": 87.78,
-      "learning_rate": 6.225352112676057e-06,
-      "loss": 0.2818,
+      "learning_rate": 3.2985074626865677e-06,
+      "loss": 0.2504,
       "step": 6320
     },
     {
       "epoch": 87.91,
-      "learning_rate": 6.154929577464788e-06,
-      "loss": 0.2895,
+      "learning_rate": 3.2611940298507464e-06,
+      "loss": 0.2614,
       "step": 6330
     },
     {
       "epoch": 88.0,
-      "eval_loss": 0.48882603645324707,
-      "eval_runtime": 126.4329,
-      "eval_samples_per_second": 4.208,
-      "eval_steps_per_second": 0.53,
-      "eval_wer": 0.20203123995628977,
+      "eval_loss": 0.4974157512187958,
+      "eval_runtime": 132.0096,
+      "eval_samples_per_second": 4.03,
+      "eval_steps_per_second": 0.508,
+      "eval_wer": 0.19354631355659832,
       "step": 6336
     },
     {
       "epoch": 88.06,
-      "learning_rate": 6.084507042253521e-06,
-      "loss": 0.3026,
+      "learning_rate": 3.2238805970149255e-06,
+      "loss": 0.2682,
       "step": 6340
     },
     {
       "epoch": 88.19,
-      "learning_rate": 6.014084507042254e-06,
-      "loss": 0.2863,
+      "learning_rate": 3.1865671641791043e-06,
+      "loss": 0.2494,
       "step": 6350
     },
     {
       "epoch": 88.33,
-      "learning_rate": 5.9436619718309865e-06,
-      "loss": 0.2946,
+      "learning_rate": 3.149253731343284e-06,
+      "loss": 0.2554,
       "step": 6360
     },
     {
       "epoch": 88.47,
-      "learning_rate": 5.873239436619718e-06,
-      "loss": 0.2898,
+      "learning_rate": 3.111940298507463e-06,
+      "loss": 0.2546,
       "step": 6370
     },
     {
       "epoch": 88.61,
-      "learning_rate": 5.802816901408451e-06,
-      "loss": 0.2784,
+      "learning_rate": 3.0746268656716418e-06,
+      "loss": 0.2447,
       "step": 6380
     },
     {
       "epoch": 88.75,
-      "learning_rate": 5.732394366197184e-06,
-      "loss": 0.2927,
+      "learning_rate": 3.037313432835821e-06,
+      "loss": 0.257,
       "step": 6390
     },
     {
       "epoch": 88.89,
-      "learning_rate": 5.661971830985916e-06,
-      "loss": 0.2823,
+      "learning_rate": 3e-06,
+      "loss": 0.2533,
       "step": 6400
     },
     {
       "epoch": 89.0,
-      "eval_loss": 0.4967969059944153,
-      "eval_runtime": 125.4743,
-      "eval_samples_per_second": 4.24,
-      "eval_steps_per_second": 0.534,
-      "eval_wer": 0.20325255511988172,
+      "eval_loss": 0.4969758987426758,
+      "eval_runtime": 132.51,
+      "eval_samples_per_second": 4.015,
+      "eval_steps_per_second": 0.506,
+      "eval_wer": 0.1948961882110947,
       "step": 6408
     },
     {
       "epoch": 89.03,
-      "learning_rate": 5.5915492957746476e-06,
-      "loss": 0.2929,
+      "learning_rate": 2.9626865671641792e-06,
+      "loss": 0.2591,
       "step": 6410
     },
     {
       "epoch": 89.17,
-      "learning_rate": 5.52112676056338e-06,
-      "loss": 0.2925,
+      "learning_rate": 2.9253731343283584e-06,
+      "loss": 0.2604,
       "step": 6420
     },
     {
       "epoch": 89.3,
-      "learning_rate": 5.450704225352113e-06,
-      "loss": 0.2773,
+      "learning_rate": 2.8880597014925376e-06,
+      "loss": 0.2445,
       "step": 6430
     },
     {
       "epoch": 89.44,
-      "learning_rate": 5.380281690140846e-06,
-      "loss": 0.2801,
+      "learning_rate": 2.8507462686567167e-06,
+      "loss": 0.2494,
       "step": 6440
     },
     {
       "epoch": 89.58,
-      "learning_rate": 5.309859154929578e-06,
-      "loss": 0.2885,
+      "learning_rate": 2.813432835820896e-06,
+      "loss": 0.255,
       "step": 6450
     },
     {
       "epoch": 89.72,
-      "learning_rate": 5.23943661971831e-06,
-      "loss": 0.2862,
+      "learning_rate": 2.7761194029850746e-06,
+      "loss": 0.2512,
       "step": 6460
     },
     {
       "epoch": 89.86,
-      "learning_rate": 5.169014084507043e-06,
-      "loss": 0.2866,
+      "learning_rate": 2.7388059701492538e-06,
+      "loss": 0.253,
       "step": 6470
     },
     {
       "epoch": 90.0,
-      "learning_rate": 5.098591549295775e-06,
-      "loss": 0.2885,
+      "learning_rate": 2.701492537313433e-06,
+      "loss": 0.2553,
       "step": 6480
     },
     {
       "epoch": 90.0,
-      "eval_loss": 0.5042439103126526,
-      "eval_runtime": 125.2751,
-      "eval_samples_per_second": 4.247,
-      "eval_steps_per_second": 0.535,
-      "eval_wer": 0.203959632319856,
+      "eval_loss": 0.5005418658256531,
+      "eval_runtime": 131.5872,
+      "eval_samples_per_second": 4.043,
+      "eval_steps_per_second": 0.509,
+      "eval_wer": 0.19296779584752843,
       "step": 6480
     },
     {
       "epoch": 90.14,
-      "learning_rate": 5.028169014084507e-06,
-      "loss": 0.2913,
+      "learning_rate": 2.664179104477612e-06,
+      "loss": 0.2521,
       "step": 6490
     },
     {
       "epoch": 90.28,
-      "learning_rate": 4.9577464788732395e-06,
-      "loss": 0.2982,
+      "learning_rate": 2.626865671641791e-06,
+      "loss": 0.2599,
       "step": 6500
     },
     {
       "epoch": 90.42,
-      "learning_rate": 4.887323943661972e-06,
-      "loss": 0.2869,
+      "learning_rate": 2.5895522388059704e-06,
+      "loss": 0.256,
       "step": 6510
     },
     {
       "epoch": 90.55,
-      "learning_rate": 4.816901408450705e-06,
-      "loss": 0.2898,
+      "learning_rate": 2.5522388059701496e-06,
+      "loss": 0.2559,
       "step": 6520
     },
     {
       "epoch": 90.69,
-      "learning_rate": 4.746478873239437e-06,
-      "loss": 0.2867,
+      "learning_rate": 2.5149253731343287e-06,
+      "loss": 0.2507,
       "step": 6530
     },
     {
       "epoch": 90.83,
-      "learning_rate": 4.67605633802817e-06,
-      "loss": 0.287,
+      "learning_rate": 2.4776119402985074e-06,
+      "loss": 0.2525,
       "step": 6540
     },
     {
       "epoch": 90.97,
-      "learning_rate": 4.6056338028169015e-06,
-      "loss": 0.2876,
+      "learning_rate": 2.4402985074626866e-06,
+      "loss": 0.2518,
       "step": 6550
     },
     {
       "epoch": 91.0,
-      "eval_loss": 0.4943309724330902,
-      "eval_runtime": 125.8105,
-      "eval_samples_per_second": 4.229,
-      "eval_steps_per_second": 0.533,
-      "eval_wer": 0.20254547791990743,
+      "eval_loss": 0.5009539127349854,
+      "eval_runtime": 131.4903,
+      "eval_samples_per_second": 4.046,
+      "eval_steps_per_second": 0.51,
+      "eval_wer": 0.1948961882110947,
       "step": 6552
     },
     {
       "epoch": 91.11,
-      "learning_rate": 4.535211267605634e-06,
-      "loss": 0.2954,
+      "learning_rate": 2.4029850746268658e-06,
+      "loss": 0.2618,
       "step": 6560
     },
     {
       "epoch": 91.25,
-      "learning_rate": 4.464788732394366e-06,
-      "loss": 0.2929,
+      "learning_rate": 2.365671641791045e-06,
+      "loss": 0.2592,
       "step": 6570
     },
     {
       "epoch": 91.39,
-      "learning_rate": 4.394366197183099e-06,
-      "loss": 0.2797,
+      "learning_rate": 2.328358208955224e-06,
+      "loss": 0.2505,
       "step": 6580
     },
     {
       "epoch": 91.53,
-      "learning_rate": 4.3239436619718315e-06,
-      "loss": 0.2857,
+      "learning_rate": 2.291044776119403e-06,
+      "loss": 0.2539,
       "step": 6590
     },
     {
       "epoch": 91.66,
-      "learning_rate": 4.253521126760563e-06,
-      "loss": 0.2783,
+      "learning_rate": 2.2537313432835824e-06,
+      "loss": 0.249,
       "step": 6600
     },
     {
       "epoch": 91.8,
-      "learning_rate": 4.183098591549296e-06,
-      "loss": 0.2761,
+      "learning_rate": 2.2164179104477616e-06,
+      "loss": 0.2428,
       "step": 6610
     },
     {
       "epoch": 91.94,
-      "learning_rate": 4.112676056338028e-06,
-      "loss": 0.2765,
+      "learning_rate": 2.1791044776119403e-06,
+      "loss": 0.2481,
       "step": 6620
     },
     {
       "epoch": 92.0,
-      "eval_loss": 0.4961002767086029,
-      "eval_runtime": 124.9062,
-      "eval_samples_per_second": 4.259,
-      "eval_steps_per_second": 0.536,
-      "eval_wer": 0.20035996657453237,
+      "eval_loss": 0.4958493113517761,
+      "eval_runtime": 131.9046,
+      "eval_samples_per_second": 4.033,
+      "eval_steps_per_second": 0.508,
+      "eval_wer": 0.1918107604293887,
       "step": 6624
     },
     {
       "epoch": 92.08,
-      "learning_rate": 4.042253521126761e-06,
-      "loss": 0.3009,
+      "learning_rate": 2.1417910447761194e-06,
+      "loss": 0.2649,
       "step": 6630
     },
     {
       "epoch": 92.22,
-      "learning_rate": 3.9718309859154926e-06,
-      "loss": 0.2869,
+      "learning_rate": 2.1044776119402986e-06,
+      "loss": 0.2549,
       "step": 6640
     },
     {
       "epoch": 92.36,
-      "learning_rate": 3.901408450704225e-06,
-      "loss": 0.2783,
+      "learning_rate": 2.0671641791044778e-06,
+      "loss": 0.2425,
       "step": 6650
     },
     {
       "epoch": 92.5,
-      "learning_rate": 3.830985915492958e-06,
-      "loss": 0.2872,
+      "learning_rate": 2.029850746268657e-06,
+      "loss": 0.2598,
       "step": 6660
     },
     {
       "epoch": 92.64,
-      "learning_rate": 3.7605633802816903e-06,
-      "loss": 0.274,
+      "learning_rate": 1.9925373134328357e-06,
+      "loss": 0.2429,
       "step": 6670
     },
     {
       "epoch": 92.78,
-      "learning_rate": 3.690140845070423e-06,
-      "loss": 0.2937,
+      "learning_rate": 1.955223880597015e-06,
+      "loss": 0.2596,
       "step": 6680
     },
     {
       "epoch": 92.91,
-      "learning_rate": 3.6197183098591553e-06,
-      "loss": 0.282,
+      "learning_rate": 1.9179104477611944e-06,
+      "loss": 0.2491,
       "step": 6690
     },
     {
       "epoch": 93.0,
-      "eval_loss": 0.4846987724304199,
-      "eval_runtime": 125.4194,
-      "eval_samples_per_second": 4.242,
-      "eval_steps_per_second": 0.534,
-      "eval_wer": 0.20170984122902874,
+      "eval_loss": 0.4932185709476471,
+      "eval_runtime": 131.4467,
+      "eval_samples_per_second": 4.047,
+      "eval_steps_per_second": 0.51,
+      "eval_wer": 0.1909108439930578,
       "step": 6696
     },
     {
       "epoch": 93.06,
-      "learning_rate": 3.549295774647887e-06,
-      "loss": 0.2995,
+      "learning_rate": 1.8805970149253733e-06,
+      "loss": 0.2682,
       "step": 6700
     },
     {
       "epoch": 93.19,
-      "learning_rate": 3.4788732394366195e-06,
-      "loss": 0.2756,
+      "learning_rate": 1.8432835820895525e-06,
+      "loss": 0.2455,
       "step": 6710
     },
     {
       "epoch": 93.33,
-      "learning_rate": 3.4084507042253522e-06,
-      "loss": 0.2797,
+      "learning_rate": 1.8059701492537314e-06,
+      "loss": 0.2456,
       "step": 6720
     },
     {
       "epoch": 93.47,
-      "learning_rate": 3.3380281690140845e-06,
-      "loss": 0.2842,
+      "learning_rate": 1.7686567164179106e-06,
+      "loss": 0.2493,
       "step": 6730
     },
     {
       "epoch": 93.61,
-      "learning_rate": 3.2676056338028173e-06,
-      "loss": 0.2866,
+      "learning_rate": 1.7313432835820896e-06,
+      "loss": 0.2526,
       "step": 6740
     },
     {
       "epoch": 93.75,
-      "learning_rate": 3.1971830985915496e-06,
-      "loss": 0.293,
+      "learning_rate": 1.6940298507462687e-06,
+      "loss": 0.2599,
       "step": 6750
     },
     {
       "epoch": 93.89,
-      "learning_rate": 3.1267605633802823e-06,
-      "loss": 0.277,
+      "learning_rate": 1.6567164179104477e-06,
+      "loss": 0.2455,
       "step": 6760
     },
     {
       "epoch": 94.0,
-      "eval_loss": 0.4849061965942383,
-      "eval_runtime": 124.6049,
-      "eval_samples_per_second": 4.269,
-      "eval_steps_per_second": 0.538,
-      "eval_wer": 0.19939577039274925,
+      "eval_loss": 0.493074506521225,
+      "eval_runtime": 131.6855,
+      "eval_samples_per_second": 4.04,
+      "eval_steps_per_second": 0.509,
+      "eval_wer": 0.19123224272031883,
       "step": 6768
     },
     {
       "epoch": 94.03,
-      "learning_rate": 3.056338028169014e-06,
-      "loss": 0.288,
+      "learning_rate": 1.6194029850746268e-06,
+      "loss": 0.2531,
       "step": 6770
     },
     {
       "epoch": 94.17,
-      "learning_rate": 2.985915492957747e-06,
-      "loss": 0.2819,
+      "learning_rate": 1.5820895522388062e-06,
+      "loss": 0.2513,
       "step": 6780
     },
     {
       "epoch": 94.3,
-      "learning_rate": 2.9154929577464788e-06,
-      "loss": 0.2869,
+      "learning_rate": 1.5447761194029851e-06,
+      "loss": 0.2534,
       "step": 6790
     },
     {
       "epoch": 94.44,
-      "learning_rate": 2.8450704225352115e-06,
-      "loss": 0.2735,
+      "learning_rate": 1.5111940298507464e-06,
+      "loss": 0.2418,
       "step": 6800
     },
     {
       "epoch": 94.58,
-      "learning_rate": 2.7746478873239438e-06,
-      "loss": 0.2725,
+      "learning_rate": 1.4738805970149254e-06,
+      "loss": 0.2407,
       "step": 6810
     },
     {
       "epoch": 94.72,
-      "learning_rate": 2.7042253521126765e-06,
-      "loss": 0.2807,
+      "learning_rate": 1.4365671641791045e-06,
+      "loss": 0.251,
       "step": 6820
     },
     {
       "epoch": 94.86,
-      "learning_rate": 2.6338028169014084e-06,
-      "loss": 0.2826,
+      "learning_rate": 1.3992537313432837e-06,
+      "loss": 0.2494,
       "step": 6830
     },
     {
       "epoch": 95.0,
-      "learning_rate": 2.563380281690141e-06,
-      "loss": 0.2883,
+      "learning_rate": 1.3619402985074628e-06,
+      "loss": 0.2581,
       "step": 6840
     },
     {
       "epoch": 95.0,
-      "eval_loss": 0.48485061526298523,
-      "eval_runtime": 125.1761,
-      "eval_samples_per_second": 4.25,
-      "eval_steps_per_second": 0.535,
-      "eval_wer": 0.20029568682908017,
+      "eval_loss": 0.49741461873054504,
+      "eval_runtime": 132.2805,
+      "eval_samples_per_second": 4.022,
+      "eval_steps_per_second": 0.506,
+      "eval_wer": 0.19148936170212766,
       "step": 6840
     },
     {
       "epoch": 95.14,
-      "learning_rate": 2.4929577464788734e-06,
-      "loss": 0.2703,
+      "learning_rate": 1.3246268656716418e-06,
+      "loss": 0.2415,
       "step": 6850
     },
     {
       "epoch": 95.28,
-      "learning_rate": 2.422535211267606e-06,
-      "loss": 0.2825,
+      "learning_rate": 1.287313432835821e-06,
+      "loss": 0.2474,
       "step": 6860
     },
     {
       "epoch": 95.42,
-      "learning_rate": 2.352112676056338e-06,
-      "loss": 0.2778,
+      "learning_rate": 1.25e-06,
+      "loss": 0.2488,
       "step": 6870
     },
     {
       "epoch": 95.55,
-      "learning_rate": 2.2816901408450707e-06,
-      "loss": 0.2909,
+      "learning_rate": 1.2126865671641793e-06,
+      "loss": 0.2566,
       "step": 6880
     },
     {
       "epoch": 95.69,
-      "learning_rate": 2.211267605633803e-06,
-      "loss": 0.2784,
+      "learning_rate": 1.1753731343283582e-06,
+      "loss": 0.2457,
       "step": 6890
     },
     {
       "epoch": 95.83,
-      "learning_rate": 2.1408450704225353e-06,
-      "loss": 0.2789,
+      "learning_rate": 1.1380597014925374e-06,
+      "loss": 0.2489,
       "step": 6900
     },
     {
       "epoch": 95.97,
-      "learning_rate": 2.0704225352112676e-06,
-      "loss": 0.2906,
+      "learning_rate": 1.1007462686567163e-06,
+      "loss": 0.2578,
       "step": 6910
     },
     {
       "epoch": 96.0,
-      "eval_loss": 0.4912075102329254,
-      "eval_runtime": 125.8417,
-      "eval_samples_per_second": 4.228,
-      "eval_steps_per_second": 0.532,
-      "eval_wer": 0.20016712733817574,
+      "eval_loss": 0.4990382492542267,
+      "eval_runtime": 131.6402,
+      "eval_samples_per_second": 4.041,
+      "eval_steps_per_second": 0.509,
+      "eval_wer": 0.1909108439930578,
       "step": 6912
     },
     {
       "epoch": 96.11,
-      "learning_rate": 2.0000000000000003e-06,
-      "loss": 0.2822,
+      "learning_rate": 1.0634328358208957e-06,
+      "loss": 0.2528,
       "step": 6920
     },
     {
       "epoch": 96.25,
-      "learning_rate": 1.9295774647887326e-06,
-      "loss": 0.2836,
+      "learning_rate": 1.0261194029850746e-06,
+      "loss": 0.2517,
       "step": 6930
     },
     {
       "epoch": 96.39,
-      "learning_rate": 1.8591549295774647e-06,
-      "loss": 0.2685,
+      "learning_rate": 9.888059701492538e-07,
+      "loss": 0.2358,
       "step": 6940
     },
     {
       "epoch": 96.53,
-      "learning_rate": 1.7887323943661972e-06,
-      "loss": 0.2838,
+      "learning_rate": 9.514925373134328e-07,
+      "loss": 0.254,
       "step": 6950
     },
     {
       "epoch": 96.66,
-      "learning_rate": 1.7183098591549297e-06,
-      "loss": 0.2775,
+      "learning_rate": 9.141791044776121e-07,
+      "loss": 0.2491,
       "step": 6960
     },
     {
       "epoch": 96.8,
-      "learning_rate": 1.6478873239436623e-06,
-      "loss": 0.2867,
+      "learning_rate": 8.768656716417912e-07,
+      "loss": 0.2494,
       "step": 6970
     },
     {
       "epoch": 96.94,
-      "learning_rate": 1.5774647887323943e-06,
-      "loss": 0.2781,
+      "learning_rate": 8.395522388059702e-07,
+      "loss": 0.2503,
       "step": 6980
     },
     {
       "epoch": 97.0,
-      "eval_loss": 0.4874226152896881,
-      "eval_runtime": 125.3753,
-      "eval_samples_per_second": 4.243,
-      "eval_steps_per_second": 0.534,
-      "eval_wer": 0.2015812817381243,
+      "eval_loss": 0.4989832937717438,
+      "eval_runtime": 131.556,
+      "eval_samples_per_second": 4.044,
+      "eval_steps_per_second": 0.509,
+      "eval_wer": 0.19136080221122323,
       "step": 6984
     },
     {
       "epoch": 97.08,
-      "learning_rate": 1.5070422535211269e-06,
-      "loss": 0.2772,
+      "learning_rate": 8.022388059701493e-07,
+      "loss": 0.2468,
       "step": 6990
     },
     {
       "epoch": 97.22,
-      "learning_rate": 1.4366197183098591e-06,
-      "loss": 0.284,
+      "learning_rate": 7.649253731343284e-07,
+      "loss": 0.2519,
       "step": 7000
     },
     {
       "epoch": 97.36,
-      "learning_rate": 1.3661971830985917e-06,
-      "loss": 0.2733,
+      "learning_rate": 7.276119402985075e-07,
+      "loss": 0.2464,
       "step": 7010
     },
     {
       "epoch": 97.5,
-      "learning_rate": 1.295774647887324e-06,
-      "loss": 0.2783,
+      "learning_rate": 6.902985074626866e-07,
+      "loss": 0.2423,
       "step": 7020
     },
     {
       "epoch": 97.64,
-      "learning_rate": 1.2253521126760565e-06,
-      "loss": 0.2632,
+      "learning_rate": 6.529850746268657e-07,
+      "loss": 0.2327,
       "step": 7030
     },
     {
       "epoch": 97.78,
-      "learning_rate": 1.1549295774647888e-06,
-      "loss": 0.2841,
+      "learning_rate": 6.156716417910448e-07,
+      "loss": 0.2523,
       "step": 7040
     },
     {
       "epoch": 97.91,
-      "learning_rate": 1.0845070422535213e-06,
-      "loss": 0.2836,
+      "learning_rate": 5.783582089552239e-07,
+      "loss": 0.2544,
       "step": 7050
     },
     {
       "epoch": 98.0,
-      "eval_loss": 0.4825427532196045,
-      "eval_runtime": 126.7624,
-      "eval_samples_per_second": 4.197,
-      "eval_steps_per_second": 0.529,
-      "eval_wer": 0.19958860962910588,
+      "eval_loss": 0.4931636154651642,
+      "eval_runtime": 132.145,
+      "eval_samples_per_second": 4.026,
+      "eval_steps_per_second": 0.507,
+      "eval_wer": 0.1903323262839879,
       "step": 7056
     },
     {
       "epoch": 98.06,
-      "learning_rate": 1.0140845070422536e-06,
-      "loss": 0.2835,
+      "learning_rate": 5.41044776119403e-07,
+      "loss": 0.2491,
       "step": 7060
     },
     {
       "epoch": 98.19,
-      "learning_rate": 9.43661971830986e-07,
-      "loss": 0.2791,
+      "learning_rate": 5.037313432835821e-07,
+      "loss": 0.2483,
       "step": 7070
     },
     {
       "epoch": 98.33,
-      "learning_rate": 8.732394366197183e-07,
-      "loss": 0.2818,
+      "learning_rate": 4.6641791044776116e-07,
+      "loss": 0.2478,
       "step": 7080
     },
     {
       "epoch": 98.47,
-      "learning_rate": 8.028169014084508e-07,
-      "loss": 0.2848,
+      "learning_rate": 4.291044776119403e-07,
+      "loss": 0.255,
       "step": 7090
     },
     {
       "epoch": 98.61,
-      "learning_rate": 7.323943661971831e-07,
-      "loss": 0.2723,
+      "learning_rate": 3.9179104477611937e-07,
+      "loss": 0.2429,
       "step": 7100
     },
     {
       "epoch": 98.75,
-      "learning_rate": 6.619718309859155e-07,
-      "loss": 0.2823,
+      "learning_rate": 3.544776119402985e-07,
+      "loss": 0.2524,
       "step": 7110
     },
     {
       "epoch": 98.89,
-      "learning_rate": 5.915492957746479e-07,
-      "loss": 0.2676,
+      "learning_rate": 3.1716417910447763e-07,
+      "loss": 0.2391,
       "step": 7120
     },
     {
       "epoch": 99.0,
-      "eval_loss": 0.4805923402309418,
-      "eval_runtime": 125.4559,
-      "eval_samples_per_second": 4.241,
-      "eval_steps_per_second": 0.534,
-      "eval_wer": 0.19946005013820145,
+      "eval_loss": 0.49354514479637146,
+      "eval_runtime": 131.8773,
+      "eval_samples_per_second": 4.034,
+      "eval_steps_per_second": 0.508,
+      "eval_wer": 0.1903966060294401,
       "step": 7128
     },
     {
       "epoch": 99.03,
-      "learning_rate": 5.211267605633803e-07,
-      "loss": 0.3002,
+      "learning_rate": 2.7985074626865674e-07,
+      "loss": 0.265,
       "step": 7130
     },
     {
       "epoch": 99.17,
-      "learning_rate": 4.507042253521127e-07,
-      "loss": 0.2796,
+      "learning_rate": 2.4253731343283584e-07,
+      "loss": 0.2477,
       "step": 7140
     },
     {
       "epoch": 99.3,
-      "learning_rate": 3.8028169014084507e-07,
-      "loss": 0.2774,
+      "learning_rate": 2.0522388059701492e-07,
+      "loss": 0.2497,
       "step": 7150
     },
     {
       "epoch": 99.44,
-      "learning_rate": 3.0985915492957747e-07,
-      "loss": 0.2858,
+      "learning_rate": 1.6791044776119403e-07,
+      "loss": 0.2535,
       "step": 7160
     },
     {
       "epoch": 99.58,
-      "learning_rate": 2.394366197183099e-07,
-      "loss": 0.2805,
+      "learning_rate": 1.3059701492537316e-07,
+      "loss": 0.2511,
       "step": 7170
     },
     {
       "epoch": 99.72,
-      "learning_rate": 1.6901408450704228e-07,
-      "loss": 0.2736,
+      "learning_rate": 9.328358208955224e-08,
+      "loss": 0.2441,
       "step": 7180
     },
     {
       "epoch": 99.86,
-      "learning_rate": 9.859154929577465e-08,
-      "loss": 0.2705,
+      "learning_rate": 5.597014925373135e-08,
+      "loss": 0.2413,
       "step": 7190
     },
     {
       "epoch": 100.0,
-      "learning_rate": 2.8169014084507045e-08,
-      "loss": 0.28,
+      "learning_rate": 1.8656716417910447e-08,
+      "loss": 0.2498,
       "step": 7200
     },
     {
       "epoch": 100.0,
-      "eval_loss": 0.4813876152038574,
-      "eval_runtime": 126.071,
-      "eval_samples_per_second": 4.22,
-      "eval_steps_per_second": 0.531,
-      "eval_wer": 0.19939577039274925,
+      "eval_loss": 0.4945172667503357,
+      "eval_runtime": 132.2872,
+      "eval_samples_per_second": 4.022,
+      "eval_steps_per_second": 0.506,
+      "eval_wer": 0.1909108439930578,
       "step": 7200
     },
     {
       "epoch": 100.0,
       "step": 7200,
       "total_flos": 2.7202925310621965e+20,
-      "train_loss": 0.3457774919602606,
-      "train_runtime": 189878.6422,
-      "train_samples_per_second": 2.432,
-      "train_steps_per_second": 0.038
+      "train_loss": 0.2786852648191982,
+      "train_runtime": 194751.2153,
+      "train_samples_per_second": 2.371,
+      "train_steps_per_second": 0.037
     }
   ],
   "max_steps": 7200,