diff --git "a/trainer_state.json" "b/trainer_state.json"
--- "a/trainer_state.json"
+++ "b/trainer_state.json"
@@ -1,6 +1,6 @@
 {
-  "best_metric": 2.5726349353790283,
-  "best_model_checkpoint": "ai-light-dance_singing3_ft_pretrain2_wav2vec2-large-xlsr-53/checkpoint-6624",
+  "best_metric": 2.4279065132141113,
+  "best_model_checkpoint": "ai-light-dance_singing3_ft_pretrain2_wav2vec2-large-xlsr-53/checkpoint-3096",
   "epoch": 99.99653979238754,
   "global_step": 7200,
   "is_hyper_param_search": false,
@@ -9,5231 +9,5231 @@
   "log_history": [
     {
       "epoch": 0.14,
-      "learning_rate": 3.5000000000000004e-06,
-      "loss": 37.8998,
+      "learning_rate": 5e-06,
+      "loss": 1.2057,
       "step": 10
     },
     {
       "epoch": 0.28,
-      "learning_rate": 8.500000000000002e-06,
-      "loss": 20.6294,
+      "learning_rate": 1e-05,
+      "loss": 1.2085,
       "step": 20
     },
     {
       "epoch": 0.42,
-      "learning_rate": 1.3500000000000001e-05,
-      "loss": 4.2282,
+      "learning_rate": 1.5e-05,
+      "loss": 1.1873,
       "step": 30
     },
     {
       "epoch": 0.55,
-      "learning_rate": 1.85e-05,
-      "loss": 3.1117,
+      "learning_rate": 2e-05,
+      "loss": 1.2064,
       "step": 40
     },
     {
       "epoch": 0.69,
-      "learning_rate": 2.35e-05,
-      "loss": 2.7247,
+      "learning_rate": 2.5e-05,
+      "loss": 1.2001,
       "step": 50
     },
     {
       "epoch": 0.83,
-      "learning_rate": 2.8499999999999998e-05,
-      "loss": 2.6035,
+      "learning_rate": 3e-05,
+      "loss": 1.2256,
       "step": 60
     },
     {
       "epoch": 0.97,
-      "learning_rate": 3.35e-05,
-      "loss": 2.1979,
+      "learning_rate": 3.5e-05,
+      "loss": 1.209,
       "step": 70
     },
     {
       "epoch": 1.0,
-      "eval_loss": 3.4874308109283447,
-      "eval_runtime": 130.8827,
-      "eval_samples_per_second": 4.065,
-      "eval_steps_per_second": 0.512,
-      "eval_wer": 0.9922221508002829,
+      "eval_loss": 2.5599188804626465,
+      "eval_runtime": 128.8737,
+      "eval_samples_per_second": 4.128,
+      "eval_steps_per_second": 0.52,
+      "eval_wer": 0.988879604036768,
       "step": 72
     },
     {
       "epoch": 1.11,
-      "learning_rate": 3.85e-05,
-      "loss": 2.2698,
+      "learning_rate": 4e-05,
+      "loss": 1.2654,
       "step": 80
     },
     {
       "epoch": 1.25,
-      "learning_rate": 4.35e-05,
-      "loss": 2.0742,
+      "learning_rate": 4.5e-05,
+      "loss": 1.2185,
       "step": 90
     },
     {
       "epoch": 1.39,
-      "learning_rate": 4.85e-05,
-      "loss": 2.3481,
+      "learning_rate": 5e-05,
+      "loss": 1.294,
       "step": 100
     },
     {
       "epoch": 1.53,
-      "learning_rate": 4.995070422535211e-05,
-      "loss": 1.9461,
+      "learning_rate": 4.992957746478874e-05,
+      "loss": 1.2704,
       "step": 110
     },
     {
       "epoch": 1.66,
-      "learning_rate": 4.988028169014085e-05,
-      "loss": 2.0287,
+      "learning_rate": 4.9859154929577466e-05,
+      "loss": 1.2611,
       "step": 120
     },
     {
       "epoch": 1.8,
-      "learning_rate": 4.980985915492958e-05,
-      "loss": 2.0229,
+      "learning_rate": 4.97887323943662e-05,
+      "loss": 1.3165,
       "step": 130
     },
     {
       "epoch": 1.94,
-      "learning_rate": 4.973943661971831e-05,
-      "loss": 1.9612,
+      "learning_rate": 4.971830985915493e-05,
+      "loss": 1.3395,
       "step": 140
     },
     {
       "epoch": 2.0,
-      "eval_loss": 4.510500431060791,
-      "eval_runtime": 131.7347,
-      "eval_samples_per_second": 4.038,
-      "eval_steps_per_second": 0.509,
-      "eval_wer": 0.988879604036768,
+      "eval_loss": 2.7188220024108887,
+      "eval_runtime": 127.016,
+      "eval_samples_per_second": 4.188,
+      "eval_steps_per_second": 0.527,
+      "eval_wer": 0.9877225686186283,
       "step": 144
     },
     {
       "epoch": 2.08,
-      "learning_rate": 4.966901408450704e-05,
-      "loss": 1.9514,
+      "learning_rate": 4.9647887323943665e-05,
+      "loss": 1.351,
       "step": 150
     },
     {
       "epoch": 2.22,
-      "learning_rate": 4.9598591549295776e-05,
-      "loss": 2.1498,
+      "learning_rate": 4.95774647887324e-05,
+      "loss": 1.2436,
       "step": 160
     },
     {
       "epoch": 2.36,
-      "learning_rate": 4.952816901408451e-05,
-      "loss": 2.1645,
+      "learning_rate": 4.950704225352113e-05,
+      "loss": 1.2872,
       "step": 170
     },
     {
       "epoch": 2.5,
-      "learning_rate": 4.945774647887324e-05,
-      "loss": 1.8589,
+      "learning_rate": 4.9436619718309864e-05,
+      "loss": 1.2178,
       "step": 180
     },
     {
       "epoch": 2.64,
-      "learning_rate": 4.9387323943661975e-05,
-      "loss": 1.8882,
+      "learning_rate": 4.936619718309859e-05,
+      "loss": 1.3138,
       "step": 190
     },
     {
       "epoch": 2.78,
-      "learning_rate": 4.93169014084507e-05,
-      "loss": 1.856,
+      "learning_rate": 4.929577464788733e-05,
+      "loss": 1.248,
       "step": 200
     },
     {
       "epoch": 2.91,
-      "learning_rate": 4.924647887323944e-05,
-      "loss": 1.8052,
+      "learning_rate": 4.9225352112676056e-05,
+      "loss": 1.2695,
       "step": 210
     },
     {
       "epoch": 3.0,
-      "eval_loss": 4.081934452056885,
-      "eval_runtime": 130.8104,
-      "eval_samples_per_second": 4.067,
-      "eval_steps_per_second": 0.512,
-      "eval_wer": 0.9786591245098669,
+      "eval_loss": 2.9988839626312256,
+      "eval_runtime": 126.8102,
+      "eval_samples_per_second": 4.195,
+      "eval_steps_per_second": 0.528,
+      "eval_wer": 0.970945555055602,
       "step": 216
     },
     {
       "epoch": 3.06,
-      "learning_rate": 4.917605633802817e-05,
-      "loss": 1.8272,
+      "learning_rate": 4.915492957746479e-05,
+      "loss": 1.2978,
       "step": 220
     },
     {
       "epoch": 3.19,
-      "learning_rate": 4.91056338028169e-05,
-      "loss": 1.7397,
+      "learning_rate": 4.908450704225353e-05,
+      "loss": 1.2646,
       "step": 230
     },
     {
       "epoch": 3.33,
-      "learning_rate": 4.903521126760564e-05,
-      "loss": 1.92,
+      "learning_rate": 4.9014084507042255e-05,
+      "loss": 1.3252,
       "step": 240
     },
     {
       "epoch": 3.47,
-      "learning_rate": 4.8964788732394366e-05,
-      "loss": 1.8222,
+      "learning_rate": 4.894366197183099e-05,
+      "loss": 1.2454,
       "step": 250
     },
     {
       "epoch": 3.61,
-      "learning_rate": 4.88943661971831e-05,
-      "loss": 1.8094,
+      "learning_rate": 4.887323943661972e-05,
+      "loss": 1.2817,
       "step": 260
     },
     {
       "epoch": 3.75,
-      "learning_rate": 4.882394366197183e-05,
-      "loss": 1.6907,
+      "learning_rate": 4.8802816901408454e-05,
+      "loss": 1.2217,
       "step": 270
     },
     {
       "epoch": 3.89,
-      "learning_rate": 4.8753521126760565e-05,
-      "loss": 1.827,
+      "learning_rate": 4.873239436619719e-05,
+      "loss": 1.2818,
       "step": 280
     },
     {
       "epoch": 4.0,
-      "eval_loss": 4.405837059020996,
-      "eval_runtime": 130.0974,
-      "eval_samples_per_second": 4.089,
-      "eval_steps_per_second": 0.515,
-      "eval_wer": 0.9844443016005656,
+      "eval_loss": 3.2352070808410645,
+      "eval_runtime": 126.3395,
+      "eval_samples_per_second": 4.211,
+      "eval_steps_per_second": 0.53,
+      "eval_wer": 0.9757022562190654,
       "step": 288
     },
     {
       "epoch": 4.03,
-      "learning_rate": 4.86830985915493e-05,
-      "loss": 1.7429,
+      "learning_rate": 4.866197183098592e-05,
+      "loss": 1.2945,
       "step": 290
     },
     {
       "epoch": 4.17,
-      "learning_rate": 4.861267605633803e-05,
-      "loss": 1.778,
+      "learning_rate": 4.8591549295774653e-05,
+      "loss": 1.2647,
       "step": 300
     },
     {
       "epoch": 4.3,
-      "learning_rate": 4.8542253521126764e-05,
-      "loss": 1.7108,
+      "learning_rate": 4.852112676056338e-05,
+      "loss": 1.2678,
       "step": 310
     },
     {
       "epoch": 4.44,
-      "learning_rate": 4.847183098591549e-05,
-      "loss": 1.7007,
+      "learning_rate": 4.845070422535212e-05,
+      "loss": 1.2601,
       "step": 320
     },
     {
       "epoch": 4.58,
-      "learning_rate": 4.840140845070423e-05,
-      "loss": 1.8591,
+      "learning_rate": 4.838028169014085e-05,
+      "loss": 1.2823,
       "step": 330
     },
     {
       "epoch": 4.72,
-      "learning_rate": 4.833098591549296e-05,
-      "loss": 1.6263,
+      "learning_rate": 4.830985915492958e-05,
+      "loss": 1.2328,
       "step": 340
     },
     {
       "epoch": 4.86,
-      "learning_rate": 4.826056338028169e-05,
-      "loss": 1.8142,
+      "learning_rate": 4.8239436619718316e-05,
+      "loss": 1.2988,
       "step": 350
     },
     {
       "epoch": 5.0,
-      "learning_rate": 4.819014084507043e-05,
-      "loss": 1.6249,
+      "learning_rate": 4.8169014084507045e-05,
+      "loss": 1.2389,
       "step": 360
     },
     {
       "epoch": 5.0,
-      "eval_loss": 4.889533519744873,
-      "eval_runtime": 130.9002,
-      "eval_samples_per_second": 4.064,
-      "eval_steps_per_second": 0.512,
-      "eval_wer": 0.9872083306550106,
+      "eval_loss": 3.686650514602661,
+      "eval_runtime": 125.8503,
+      "eval_samples_per_second": 4.227,
+      "eval_steps_per_second": 0.532,
+      "eval_wer": 0.9783377257826059,
       "step": 360
     },
     {
       "epoch": 5.14,
-      "learning_rate": 4.8119718309859155e-05,
-      "loss": 1.7359,
+      "learning_rate": 4.809859154929578e-05,
+      "loss": 1.3495,
       "step": 370
     },
     {
       "epoch": 5.28,
-      "learning_rate": 4.804929577464789e-05,
-      "loss": 1.8771,
+      "learning_rate": 4.8028169014084515e-05,
+      "loss": 1.2582,
       "step": 380
     },
     {
       "epoch": 5.42,
-      "learning_rate": 4.7978873239436626e-05,
-      "loss": 1.6211,
+      "learning_rate": 4.7957746478873244e-05,
+      "loss": 1.2744,
       "step": 390
     },
     {
       "epoch": 5.55,
-      "learning_rate": 4.7908450704225354e-05,
-      "loss": 1.6833,
+      "learning_rate": 4.788732394366197e-05,
+      "loss": 1.2988,
       "step": 400
     },
     {
       "epoch": 5.69,
-      "learning_rate": 4.783802816901409e-05,
-      "loss": 1.6535,
+      "learning_rate": 4.78169014084507e-05,
+      "loss": 1.2398,
       "step": 410
     },
     {
       "epoch": 5.83,
-      "learning_rate": 4.776760563380282e-05,
-      "loss": 1.727,
+      "learning_rate": 4.7746478873239436e-05,
+      "loss": 1.2788,
       "step": 420
     },
     {
       "epoch": 5.97,
-      "learning_rate": 4.769718309859155e-05,
-      "loss": 1.6285,
+      "learning_rate": 4.767605633802817e-05,
+      "loss": 1.2368,
       "step": 430
     },
     {
       "epoch": 6.0,
-      "eval_loss": 4.753390312194824,
-      "eval_runtime": 130.2251,
-      "eval_samples_per_second": 4.085,
-      "eval_steps_per_second": 0.514,
-      "eval_wer": 0.9837372244005914,
+      "eval_loss": 3.3188819885253906,
+      "eval_runtime": 126.8606,
+      "eval_samples_per_second": 4.194,
+      "eval_steps_per_second": 0.528,
+      "eval_wer": 0.9811017548370509,
       "step": 432
     },
     {
       "epoch": 6.11,
-      "learning_rate": 4.762676056338028e-05,
-      "loss": 1.7857,
+      "learning_rate": 4.76056338028169e-05,
+      "loss": 1.3373,
       "step": 440
     },
     {
       "epoch": 6.25,
-      "learning_rate": 4.755633802816902e-05,
-      "loss": 1.5781,
+      "learning_rate": 4.7535211267605635e-05,
+      "loss": 1.2149,
       "step": 450
     },
     {
       "epoch": 6.39,
-      "learning_rate": 4.748591549295775e-05,
-      "loss": 1.7005,
+      "learning_rate": 4.7464788732394363e-05,
+      "loss": 1.2758,
       "step": 460
     },
     {
       "epoch": 6.53,
-      "learning_rate": 4.741549295774648e-05,
-      "loss": 1.6211,
+      "learning_rate": 4.73943661971831e-05,
+      "loss": 1.2639,
       "step": 470
     },
     {
       "epoch": 6.66,
-      "learning_rate": 4.7345070422535216e-05,
-      "loss": 1.769,
+      "learning_rate": 4.7323943661971834e-05,
+      "loss": 1.2651,
       "step": 480
     },
     {
       "epoch": 6.8,
-      "learning_rate": 4.7274647887323945e-05,
-      "loss": 1.7096,
+      "learning_rate": 4.725352112676056e-05,
+      "loss": 1.2955,
       "step": 490
     },
     {
       "epoch": 6.94,
-      "learning_rate": 4.720422535211268e-05,
-      "loss": 1.6713,
+      "learning_rate": 4.71830985915493e-05,
+      "loss": 1.2307,
       "step": 500
     },
     {
       "epoch": 7.0,
-      "eval_loss": 4.558750629425049,
-      "eval_runtime": 130.6897,
-      "eval_samples_per_second": 4.071,
-      "eval_steps_per_second": 0.513,
-      "eval_wer": 0.9818731117824774,
+      "eval_loss": 3.0785598754882812,
+      "eval_runtime": 126.8144,
+      "eval_samples_per_second": 4.195,
+      "eval_steps_per_second": 0.528,
+      "eval_wer": 0.9657388956739731,
       "step": 504
     },
     {
       "epoch": 7.08,
-      "learning_rate": 4.7133802816901415e-05,
-      "loss": 1.7235,
+      "learning_rate": 4.7112676056338026e-05,
+      "loss": 1.3135,
       "step": 510
     },
     {
       "epoch": 7.22,
-      "learning_rate": 4.7063380281690144e-05,
-      "loss": 1.6237,
+      "learning_rate": 4.704225352112676e-05,
+      "loss": 1.2488,
       "step": 520
     },
     {
       "epoch": 7.36,
-      "learning_rate": 4.699295774647888e-05,
-      "loss": 1.6671,
+      "learning_rate": 4.69718309859155e-05,
+      "loss": 1.2778,
       "step": 530
     },
     {
       "epoch": 7.5,
-      "learning_rate": 4.692253521126761e-05,
-      "loss": 1.5521,
+      "learning_rate": 4.6901408450704225e-05,
+      "loss": 1.2424,
       "step": 540
     },
     {
       "epoch": 7.64,
-      "learning_rate": 4.685211267605634e-05,
-      "loss": 1.6938,
+      "learning_rate": 4.683098591549296e-05,
+      "loss": 1.2831,
       "step": 550
     },
     {
       "epoch": 7.78,
-      "learning_rate": 4.678169014084508e-05,
-      "loss": 1.7133,
+      "learning_rate": 4.676056338028169e-05,
+      "loss": 1.2642,
       "step": 560
     },
     {
       "epoch": 7.91,
-      "learning_rate": 4.6711267605633806e-05,
-      "loss": 1.777,
+      "learning_rate": 4.6690140845070424e-05,
+      "loss": 1.2607,
       "step": 570
     },
     {
       "epoch": 8.0,
-      "eval_loss": 4.741362571716309,
-      "eval_runtime": 131.3054,
-      "eval_samples_per_second": 4.052,
-      "eval_steps_per_second": 0.51,
-      "eval_wer": 0.9859870154914187,
+      "eval_loss": 2.971968412399292,
+      "eval_runtime": 126.4051,
+      "eval_samples_per_second": 4.209,
+      "eval_steps_per_second": 0.53,
+      "eval_wer": 0.9677315677829916,
       "step": 576
     },
     {
       "epoch": 8.06,
-      "learning_rate": 4.664084507042254e-05,
-      "loss": 1.7278,
+      "learning_rate": 4.661971830985915e-05,
+      "loss": 1.2819,
       "step": 580
     },
     {
       "epoch": 8.19,
-      "learning_rate": 4.657042253521127e-05,
-      "loss": 1.6539,
+      "learning_rate": 4.654929577464789e-05,
+      "loss": 1.2396,
       "step": 590
     },
     {
       "epoch": 8.33,
-      "learning_rate": 4.6500000000000005e-05,
-      "loss": 1.6884,
+      "learning_rate": 4.647887323943662e-05,
+      "loss": 1.2609,
       "step": 600
     },
     {
       "epoch": 8.47,
-      "learning_rate": 4.6429577464788734e-05,
-      "loss": 1.6151,
+      "learning_rate": 4.640845070422535e-05,
+      "loss": 1.2216,
       "step": 610
     },
     {
       "epoch": 8.61,
-      "learning_rate": 4.635915492957747e-05,
-      "loss": 1.7166,
+      "learning_rate": 4.633802816901409e-05,
+      "loss": 1.2899,
       "step": 620
     },
     {
       "epoch": 8.75,
-      "learning_rate": 4.6288732394366204e-05,
-      "loss": 1.4914,
+      "learning_rate": 4.6267605633802816e-05,
+      "loss": 1.2237,
       "step": 630
     },
     {
       "epoch": 8.89,
-      "learning_rate": 4.6218309859154926e-05,
-      "loss": 1.6759,
+      "learning_rate": 4.619718309859155e-05,
+      "loss": 1.2584,
       "step": 640
     },
     {
       "epoch": 9.0,
-      "eval_loss": 4.522234916687012,
-      "eval_runtime": 130.413,
-      "eval_samples_per_second": 4.079,
-      "eval_steps_per_second": 0.514,
-      "eval_wer": 0.9874011698913673,
+      "eval_loss": 2.561340570449829,
+      "eval_runtime": 126.4151,
+      "eval_samples_per_second": 4.208,
+      "eval_steps_per_second": 0.53,
+      "eval_wer": 0.9701741981101755,
       "step": 648
     },
     {
       "epoch": 9.03,
-      "learning_rate": 4.614788732394366e-05,
-      "loss": 1.6397,
+      "learning_rate": 4.6126760563380286e-05,
+      "loss": 1.2901,
       "step": 650
     },
     {
       "epoch": 9.17,
-      "learning_rate": 4.60774647887324e-05,
-      "loss": 1.5902,
+      "learning_rate": 4.6056338028169015e-05,
+      "loss": 1.2446,
       "step": 660
     },
     {
       "epoch": 9.3,
-      "learning_rate": 4.6007042253521125e-05,
-      "loss": 1.752,
+      "learning_rate": 4.598591549295775e-05,
+      "loss": 1.2673,
       "step": 670
     },
     {
       "epoch": 9.44,
-      "learning_rate": 4.593661971830986e-05,
-      "loss": 1.6492,
+      "learning_rate": 4.591549295774648e-05,
+      "loss": 1.2929,
       "step": 680
     },
     {
       "epoch": 9.58,
-      "learning_rate": 4.586619718309859e-05,
-      "loss": 1.6478,
+      "learning_rate": 4.5845070422535214e-05,
+      "loss": 1.3131,
       "step": 690
     },
     {
       "epoch": 9.72,
-      "learning_rate": 4.5795774647887324e-05,
-      "loss": 1.593,
+      "learning_rate": 4.577464788732395e-05,
+      "loss": 1.2339,
       "step": 700
     },
     {
       "epoch": 9.86,
-      "learning_rate": 4.572535211267606e-05,
-      "loss": 1.6711,
+      "learning_rate": 4.570422535211268e-05,
+      "loss": 1.2997,
       "step": 710
     },
     {
       "epoch": 10.0,
-      "learning_rate": 4.565492957746479e-05,
-      "loss": 1.5533,
+      "learning_rate": 4.563380281690141e-05,
+      "loss": 1.2266,
       "step": 720
     },
     {
       "epoch": 10.0,
-      "eval_loss": 3.6153128147125244,
-      "eval_runtime": 129.7947,
-      "eval_samples_per_second": 4.099,
-      "eval_steps_per_second": 0.516,
-      "eval_wer": 0.982387349746095,
+      "eval_loss": 2.6937201023101807,
+      "eval_runtime": 126.5235,
+      "eval_samples_per_second": 4.205,
+      "eval_steps_per_second": 0.53,
+      "eval_wer": 0.9609821945105097,
       "step": 720
     },
     {
       "epoch": 10.14,
-      "learning_rate": 4.558450704225352e-05,
-      "loss": 1.6175,
+      "learning_rate": 4.556338028169014e-05,
+      "loss": 1.2622,
       "step": 730
     },
     {
       "epoch": 10.28,
-      "learning_rate": 4.551408450704225e-05,
-      "loss": 1.6756,
+      "learning_rate": 4.5492957746478876e-05,
+      "loss": 1.2754,
       "step": 740
     },
     {
       "epoch": 10.42,
-      "learning_rate": 4.544366197183099e-05,
-      "loss": 1.6021,
+      "learning_rate": 4.542253521126761e-05,
+      "loss": 1.2391,
       "step": 750
     },
     {
       "epoch": 10.55,
-      "learning_rate": 4.537323943661972e-05,
-      "loss": 1.5629,
+      "learning_rate": 4.535211267605634e-05,
+      "loss": 1.2542,
       "step": 760
     },
     {
       "epoch": 10.69,
-      "learning_rate": 4.530281690140845e-05,
-      "loss": 1.5889,
+      "learning_rate": 4.5281690140845075e-05,
+      "loss": 1.2628,
       "step": 770
     },
     {
       "epoch": 10.83,
-      "learning_rate": 4.5232394366197186e-05,
-      "loss": 1.7345,
+      "learning_rate": 4.5211267605633804e-05,
+      "loss": 1.3325,
       "step": 780
     },
     {
       "epoch": 10.97,
-      "learning_rate": 4.5161971830985914e-05,
-      "loss": 1.5053,
+      "learning_rate": 4.514084507042254e-05,
+      "loss": 1.262,
       "step": 790
     },
     {
       "epoch": 11.0,
-      "eval_loss": 5.648496150970459,
-      "eval_runtime": 129.8957,
-      "eval_samples_per_second": 4.096,
-      "eval_steps_per_second": 0.516,
-      "eval_wer": 0.9906794369094298,
+      "eval_loss": 3.9060168266296387,
+      "eval_runtime": 125.7452,
+      "eval_samples_per_second": 4.231,
+      "eval_steps_per_second": 0.533,
+      "eval_wer": 0.9744809410554734,
       "step": 792
     },
     {
       "epoch": 11.11,
-      "learning_rate": 4.509154929577465e-05,
-      "loss": 1.6899,
+      "learning_rate": 4.507042253521127e-05,
+      "loss": 1.3256,
       "step": 800
     },
     {
       "epoch": 11.25,
-      "learning_rate": 4.502112676056338e-05,
-      "loss": 1.5438,
+      "learning_rate": 4.5e-05,
+      "loss": 1.2094,
       "step": 810
     },
     {
       "epoch": 11.39,
-      "learning_rate": 4.4950704225352113e-05,
-      "loss": 1.6434,
+      "learning_rate": 4.492957746478874e-05,
+      "loss": 1.2787,
       "step": 820
     },
     {
       "epoch": 11.53,
-      "learning_rate": 4.488028169014085e-05,
-      "loss": 1.5148,
+      "learning_rate": 4.4859154929577467e-05,
+      "loss": 1.248,
       "step": 830
     },
     {
       "epoch": 11.66,
-      "learning_rate": 4.480985915492958e-05,
-      "loss": 1.5729,
+      "learning_rate": 4.47887323943662e-05,
+      "loss": 1.2499,
       "step": 840
     },
     {
       "epoch": 11.8,
-      "learning_rate": 4.473943661971831e-05,
-      "loss": 1.6381,
+      "learning_rate": 4.471830985915493e-05,
+      "loss": 1.2635,
       "step": 850
     },
     {
       "epoch": 11.94,
-      "learning_rate": 4.466901408450704e-05,
-      "loss": 1.6541,
+      "learning_rate": 4.4647887323943666e-05,
+      "loss": 1.2361,
       "step": 860
     },
     {
       "epoch": 12.0,
-      "eval_loss": 5.4656782150268555,
-      "eval_runtime": 130.1688,
-      "eval_samples_per_second": 4.087,
-      "eval_steps_per_second": 0.515,
-      "eval_wer": 0.9843157421096612,
+      "eval_loss": 3.6138112545013428,
+      "eval_runtime": 126.2162,
+      "eval_samples_per_second": 4.215,
+      "eval_steps_per_second": 0.531,
+      "eval_wer": 0.9718454714919329,
       "step": 864
     },
     {
       "epoch": 12.08,
-      "learning_rate": 4.4598591549295776e-05,
-      "loss": 1.6775,
+      "learning_rate": 4.45774647887324e-05,
+      "loss": 1.2703,
       "step": 870
     },
     {
       "epoch": 12.22,
-      "learning_rate": 4.452816901408451e-05,
-      "loss": 1.546,
+      "learning_rate": 4.450704225352113e-05,
+      "loss": 1.2139,
       "step": 880
     },
     {
       "epoch": 12.36,
-      "learning_rate": 4.445774647887324e-05,
-      "loss": 1.7313,
+      "learning_rate": 4.4436619718309865e-05,
+      "loss": 1.281,
       "step": 890
     },
     {
       "epoch": 12.5,
-      "learning_rate": 4.4387323943661975e-05,
-      "loss": 1.5004,
+      "learning_rate": 4.436619718309859e-05,
+      "loss": 1.2076,
       "step": 900
     },
     {
       "epoch": 12.64,
-      "learning_rate": 4.4316901408450704e-05,
-      "loss": 1.6062,
+      "learning_rate": 4.429577464788733e-05,
+      "loss": 1.2545,
       "step": 910
     },
     {
       "epoch": 12.78,
-      "learning_rate": 4.424647887323944e-05,
-      "loss": 1.5941,
+      "learning_rate": 4.4225352112676064e-05,
+      "loss": 1.2332,
       "step": 920
     },
     {
       "epoch": 12.91,
-      "learning_rate": 4.4176056338028174e-05,
-      "loss": 1.5891,
+      "learning_rate": 4.415492957746479e-05,
+      "loss": 1.2348,
       "step": 930
     },
     {
       "epoch": 13.0,
-      "eval_loss": 4.453381061553955,
-      "eval_runtime": 130.587,
-      "eval_samples_per_second": 4.074,
-      "eval_steps_per_second": 0.513,
-      "eval_wer": 0.9877225686186283,
+      "eval_loss": 3.4837868213653564,
+      "eval_runtime": 126.6207,
+      "eval_samples_per_second": 4.202,
+      "eval_steps_per_second": 0.529,
+      "eval_wer": 0.9745452208009256,
       "step": 936
     },
     {
       "epoch": 13.06,
-      "learning_rate": 4.41056338028169e-05,
-      "loss": 1.6226,
+      "learning_rate": 4.408450704225353e-05,
+      "loss": 1.3286,
       "step": 940
     },
     {
       "epoch": 13.19,
-      "learning_rate": 4.403521126760564e-05,
-      "loss": 1.5122,
+      "learning_rate": 4.4014084507042256e-05,
+      "loss": 1.211,
       "step": 950
     },
     {
       "epoch": 13.33,
-      "learning_rate": 4.3964788732394366e-05,
-      "loss": 1.5418,
+      "learning_rate": 4.394366197183099e-05,
+      "loss": 1.2424,
       "step": 960
     },
     {
       "epoch": 13.47,
-      "learning_rate": 4.38943661971831e-05,
-      "loss": 1.5366,
+      "learning_rate": 4.3873239436619726e-05,
+      "loss": 1.2201,
       "step": 970
     },
     {
       "epoch": 13.61,
-      "learning_rate": 4.382394366197183e-05,
-      "loss": 1.6477,
+      "learning_rate": 4.3802816901408455e-05,
+      "loss": 1.2663,
       "step": 980
     },
     {
       "epoch": 13.75,
-      "learning_rate": 4.3753521126760566e-05,
-      "loss": 1.5524,
+      "learning_rate": 4.373239436619718e-05,
+      "loss": 1.2221,
       "step": 990
     },
     {
       "epoch": 13.89,
-      "learning_rate": 4.36830985915493e-05,
-      "loss": 1.6809,
+      "learning_rate": 4.366197183098591e-05,
+      "loss": 1.2715,
       "step": 1000
     },
     {
       "epoch": 14.0,
-      "eval_loss": 4.840369701385498,
-      "eval_runtime": 129.3528,
-      "eval_samples_per_second": 4.113,
-      "eval_steps_per_second": 0.518,
-      "eval_wer": 0.9890724432731246,
+      "eval_loss": 3.3127832412719727,
+      "eval_runtime": 126.5745,
+      "eval_samples_per_second": 4.203,
+      "eval_steps_per_second": 0.529,
+      "eval_wer": 0.9750594587645433,
       "step": 1008
     },
     {
       "epoch": 14.03,
-      "learning_rate": 4.361267605633803e-05,
-      "loss": 1.5893,
+      "learning_rate": 4.359154929577465e-05,
+      "loss": 1.2714,
       "step": 1010
     },
     {
       "epoch": 14.17,
-      "learning_rate": 4.3542253521126765e-05,
-      "loss": 1.5516,
+      "learning_rate": 4.352112676056338e-05,
+      "loss": 1.2512,
       "step": 1020
     },
     {
       "epoch": 14.3,
-      "learning_rate": 4.347183098591549e-05,
-      "loss": 1.6468,
+      "learning_rate": 4.345070422535211e-05,
+      "loss": 1.247,
       "step": 1030
     },
     {
       "epoch": 14.44,
-      "learning_rate": 4.340140845070423e-05,
-      "loss": 1.5277,
+      "learning_rate": 4.3380281690140846e-05,
+      "loss": 1.2068,
       "step": 1040
     },
     {
       "epoch": 14.58,
-      "learning_rate": 4.3330985915492964e-05,
-      "loss": 1.5323,
+      "learning_rate": 4.3309859154929575e-05,
+      "loss": 1.2731,
       "step": 1050
     },
     {
       "epoch": 14.72,
-      "learning_rate": 4.326056338028169e-05,
-      "loss": 1.4823,
+      "learning_rate": 4.323943661971831e-05,
+      "loss": 1.2367,
       "step": 1060
     },
     {
       "epoch": 14.86,
-      "learning_rate": 4.319014084507043e-05,
-      "loss": 1.6693,
+      "learning_rate": 4.3169014084507045e-05,
+      "loss": 1.2975,
       "step": 1070
     },
     {
       "epoch": 15.0,
-      "learning_rate": 4.3119718309859156e-05,
-      "loss": 1.5506,
+      "learning_rate": 4.3098591549295774e-05,
+      "loss": 1.2505,
       "step": 1080
     },
     {
       "epoch": 15.0,
-      "eval_loss": 3.7248387336730957,
-      "eval_runtime": 129.7977,
-      "eval_samples_per_second": 4.099,
-      "eval_steps_per_second": 0.516,
-      "eval_wer": 0.9872726104004628,
+      "eval_loss": 3.201465129852295,
+      "eval_runtime": 127.8848,
+      "eval_samples_per_second": 4.16,
+      "eval_steps_per_second": 0.524,
+      "eval_wer": 0.9710098348010542,
       "step": 1080
     },
     {
       "epoch": 15.14,
-      "learning_rate": 4.304929577464789e-05,
-      "loss": 1.6658,
+      "learning_rate": 4.302816901408451e-05,
+      "loss": 1.2885,
       "step": 1090
     },
     {
       "epoch": 15.28,
-      "learning_rate": 4.2978873239436626e-05,
-      "loss": 1.5241,
+      "learning_rate": 4.295774647887324e-05,
+      "loss": 1.2462,
       "step": 1100
     },
     {
       "epoch": 15.42,
-      "learning_rate": 4.2908450704225355e-05,
-      "loss": 1.6534,
+      "learning_rate": 4.288732394366197e-05,
+      "loss": 1.242,
       "step": 1110
     },
     {
       "epoch": 15.55,
-      "learning_rate": 4.283802816901409e-05,
-      "loss": 1.5788,
+      "learning_rate": 4.281690140845071e-05,
+      "loss": 1.2736,
       "step": 1120
     },
     {
       "epoch": 15.69,
-      "learning_rate": 4.276760563380282e-05,
-      "loss": 1.5388,
+      "learning_rate": 4.2746478873239436e-05,
+      "loss": 1.2562,
       "step": 1130
     },
     {
       "epoch": 15.83,
-      "learning_rate": 4.2697183098591554e-05,
-      "loss": 1.6622,
+      "learning_rate": 4.267605633802817e-05,
+      "loss": 1.2479,
       "step": 1140
     },
     {
       "epoch": 15.97,
-      "learning_rate": 4.262676056338029e-05,
-      "loss": 1.4754,
+      "learning_rate": 4.26056338028169e-05,
+      "loss": 1.211,
       "step": 1150
     },
     {
       "epoch": 16.0,
-      "eval_loss": 4.396983623504639,
-      "eval_runtime": 130.0288,
-      "eval_samples_per_second": 4.091,
-      "eval_steps_per_second": 0.515,
-      "eval_wer": 0.9864369737095842,
+      "eval_loss": 3.4709153175354004,
+      "eval_runtime": 126.1647,
+      "eval_samples_per_second": 4.217,
+      "eval_steps_per_second": 0.531,
+      "eval_wer": 0.9708812753101498,
       "step": 1152
     },
     {
       "epoch": 16.11,
-      "learning_rate": 4.255633802816902e-05,
-      "loss": 1.6914,
+      "learning_rate": 4.2535211267605635e-05,
+      "loss": 1.2747,
       "step": 1160
     },
     {
       "epoch": 16.25,
-      "learning_rate": 4.248591549295775e-05,
-      "loss": 1.514,
+      "learning_rate": 4.2464788732394364e-05,
+      "loss": 1.1978,
       "step": 1170
     },
     {
       "epoch": 16.39,
-      "learning_rate": 4.241549295774648e-05,
-      "loss": 1.5457,
+      "learning_rate": 4.23943661971831e-05,
+      "loss": 1.2175,
       "step": 1180
     },
     {
       "epoch": 16.53,
-      "learning_rate": 4.2345070422535217e-05,
-      "loss": 1.49,
+      "learning_rate": 4.2323943661971834e-05,
+      "loss": 1.2125,
       "step": 1190
     },
     {
       "epoch": 16.66,
-      "learning_rate": 4.2274647887323945e-05,
-      "loss": 1.672,
+      "learning_rate": 4.225352112676056e-05,
+      "loss": 1.2241,
       "step": 1200
     },
     {
       "epoch": 16.8,
-      "learning_rate": 4.220422535211268e-05,
-      "loss": 1.5788,
+      "learning_rate": 4.21830985915493e-05,
+      "loss": 1.2471,
       "step": 1210
     },
     {
       "epoch": 16.94,
-      "learning_rate": 4.2133802816901416e-05,
-      "loss": 1.4576,
+      "learning_rate": 4.211267605633803e-05,
+      "loss": 1.2067,
       "step": 1220
     },
     {
       "epoch": 17.0,
-      "eval_loss": 4.477088928222656,
-      "eval_runtime": 130.6058,
-      "eval_samples_per_second": 4.073,
-      "eval_steps_per_second": 0.513,
-      "eval_wer": 0.9854727775278009,
+      "eval_loss": 3.0566349029541016,
+      "eval_runtime": 126.7481,
+      "eval_samples_per_second": 4.197,
+      "eval_steps_per_second": 0.529,
+      "eval_wer": 0.9672816095648261,
       "step": 1224
     },
     {
       "epoch": 17.08,
-      "learning_rate": 4.206338028169014e-05,
-      "loss": 1.7007,
+      "learning_rate": 4.204225352112676e-05,
+      "loss": 1.3137,
       "step": 1230
     },
     {
       "epoch": 17.22,
-      "learning_rate": 4.199295774647887e-05,
-      "loss": 1.6807,
+      "learning_rate": 4.19718309859155e-05,
+      "loss": 1.1953,
       "step": 1240
     },
     {
       "epoch": 17.36,
-      "learning_rate": 4.192253521126761e-05,
-      "loss": 1.5547,
+      "learning_rate": 4.1901408450704226e-05,
+      "loss": 1.2506,
       "step": 1250
     },
     {
       "epoch": 17.5,
-      "learning_rate": 4.1852112676056336e-05,
-      "loss": 1.4415,
+      "learning_rate": 4.183098591549296e-05,
+      "loss": 1.2202,
       "step": 1260
     },
     {
       "epoch": 17.64,
-      "learning_rate": 4.178169014084507e-05,
-      "loss": 1.5736,
+      "learning_rate": 4.176056338028169e-05,
+      "loss": 1.2442,
       "step": 1270
     },
     {
       "epoch": 17.78,
-      "learning_rate": 4.17112676056338e-05,
-      "loss": 1.5291,
+      "learning_rate": 4.1690140845070425e-05,
+      "loss": 1.2342,
       "step": 1280
     },
     {
       "epoch": 17.91,
-      "learning_rate": 4.1640845070422535e-05,
-      "loss": 1.5614,
+      "learning_rate": 4.161971830985916e-05,
+      "loss": 1.2536,
       "step": 1290
     },
     {
       "epoch": 18.0,
-      "eval_loss": 3.5102336406707764,
-      "eval_runtime": 129.8253,
-      "eval_samples_per_second": 4.098,
-      "eval_steps_per_second": 0.516,
-      "eval_wer": 0.986115574982323,
+      "eval_loss": 2.5478615760803223,
+      "eval_runtime": 128.2066,
+      "eval_samples_per_second": 4.15,
+      "eval_steps_per_second": 0.523,
+      "eval_wer": 0.9788519637462235,
       "step": 1296
     },
     {
       "epoch": 18.06,
-      "learning_rate": 4.157042253521127e-05,
-      "loss": 1.5626,
+      "learning_rate": 4.154929577464789e-05,
+      "loss": 1.2678,
       "step": 1300
     },
     {
       "epoch": 18.19,
-      "learning_rate": 4.15e-05,
-      "loss": 1.4813,
+      "learning_rate": 4.1478873239436624e-05,
+      "loss": 1.1947,
       "step": 1310
     },
     {
       "epoch": 18.33,
-      "learning_rate": 4.1429577464788734e-05,
-      "loss": 1.5684,
+      "learning_rate": 4.140845070422535e-05,
+      "loss": 1.2312,
       "step": 1320
     },
     {
       "epoch": 18.47,
-      "learning_rate": 4.135915492957746e-05,
-      "loss": 1.4684,
+      "learning_rate": 4.133802816901409e-05,
+      "loss": 1.2167,
       "step": 1330
     },
     {
       "epoch": 18.61,
-      "learning_rate": 4.12887323943662e-05,
-      "loss": 1.5375,
+      "learning_rate": 4.126760563380282e-05,
+      "loss": 1.2303,
       "step": 1340
     },
     {
       "epoch": 18.75,
-      "learning_rate": 4.1218309859154927e-05,
-      "loss": 1.4983,
+      "learning_rate": 4.119718309859155e-05,
+      "loss": 1.2039,
       "step": 1350
     },
     {
       "epoch": 18.89,
-      "learning_rate": 4.114788732394366e-05,
-      "loss": 1.655,
+      "learning_rate": 4.1126760563380286e-05,
+      "loss": 1.2297,
       "step": 1360
     },
     {
       "epoch": 19.0,
-      "eval_loss": 4.806183338165283,
-      "eval_runtime": 129.9512,
-      "eval_samples_per_second": 4.094,
-      "eval_steps_per_second": 0.516,
-      "eval_wer": 0.9863084142186797,
+      "eval_loss": 2.830681562423706,
+      "eval_runtime": 127.0885,
+      "eval_samples_per_second": 4.186,
+      "eval_steps_per_second": 0.527,
+      "eval_wer": 0.9710098348010542,
       "step": 1368
     },
     {
       "epoch": 19.03,
-      "learning_rate": 4.10774647887324e-05,
-      "loss": 1.6348,
+      "learning_rate": 4.1056338028169015e-05,
+      "loss": 1.2467,
       "step": 1370
     },
     {
       "epoch": 19.17,
-      "learning_rate": 4.1007042253521126e-05,
-      "loss": 1.558,
+      "learning_rate": 4.098591549295775e-05,
+      "loss": 1.2161,
       "step": 1380
     },
     {
       "epoch": 19.3,
-      "learning_rate": 4.093661971830986e-05,
-      "loss": 1.5928,
+      "learning_rate": 4.091549295774648e-05,
+      "loss": 1.218,
       "step": 1390
     },
     {
       "epoch": 19.44,
-      "learning_rate": 4.086619718309859e-05,
-      "loss": 1.5327,
+      "learning_rate": 4.0845070422535214e-05,
+      "loss": 1.1997,
       "step": 1400
     },
     {
       "epoch": 19.58,
-      "learning_rate": 4.0795774647887325e-05,
-      "loss": 1.5945,
+      "learning_rate": 4.077464788732395e-05,
+      "loss": 1.2471,
       "step": 1410
     },
     {
       "epoch": 19.72,
-      "learning_rate": 4.072535211267606e-05,
-      "loss": 1.4834,
+      "learning_rate": 4.070422535211268e-05,
+      "loss": 1.2013,
       "step": 1420
     },
     {
       "epoch": 19.86,
-      "learning_rate": 4.065492957746479e-05,
-      "loss": 1.5911,
+      "learning_rate": 4.063380281690141e-05,
+      "loss": 1.2477,
       "step": 1430
     },
     {
       "epoch": 20.0,
-      "learning_rate": 4.0584507042253524e-05,
-      "loss": 1.455,
+      "learning_rate": 4.056338028169014e-05,
+      "loss": 1.1949,
       "step": 1440
     },
     {
       "epoch": 20.0,
-      "eval_loss": 5.0790300369262695,
-      "eval_runtime": 129.7071,
-      "eval_samples_per_second": 4.102,
-      "eval_steps_per_second": 0.517,
-      "eval_wer": 0.9879796876004371,
+      "eval_loss": 3.4112143516540527,
+      "eval_runtime": 126.2652,
+      "eval_samples_per_second": 4.213,
+      "eval_steps_per_second": 0.531,
+      "eval_wer": 0.9776949283280838,
       "step": 1440
     },
     {
       "epoch": 20.14,
-      "learning_rate": 4.051408450704225e-05,
-      "loss": 1.6204,
+      "learning_rate": 4.049295774647888e-05,
+      "loss": 1.2656,
       "step": 1450
     },
     {
       "epoch": 20.28,
-      "learning_rate": 4.044366197183099e-05,
-      "loss": 1.5141,
+      "learning_rate": 4.042253521126761e-05,
+      "loss": 1.1936,
       "step": 1460
     },
     {
       "epoch": 20.42,
-      "learning_rate": 4.037323943661972e-05,
-      "loss": 1.5035,
+      "learning_rate": 4.035211267605634e-05,
+      "loss": 1.1896,
       "step": 1470
     },
     {
       "epoch": 20.55,
-      "learning_rate": 4.030281690140845e-05,
-      "loss": 1.6486,
+      "learning_rate": 4.0281690140845076e-05,
+      "loss": 1.2437,
       "step": 1480
     },
     {
       "epoch": 20.69,
-      "learning_rate": 4.0232394366197186e-05,
-      "loss": 1.4833,
+      "learning_rate": 4.0211267605633804e-05,
+      "loss": 1.2124,
       "step": 1490
     },
     {
       "epoch": 20.83,
-      "learning_rate": 4.0161971830985915e-05,
-      "loss": 1.616,
+      "learning_rate": 4.014084507042254e-05,
+      "loss": 1.2475,
       "step": 1500
     },
     {
       "epoch": 20.97,
-      "learning_rate": 4.009154929577465e-05,
-      "loss": 1.4701,
+      "learning_rate": 4.0070422535211275e-05,
+      "loss": 1.2181,
       "step": 1510
     },
     {
       "epoch": 21.0,
-      "eval_loss": 4.231964111328125,
-      "eval_runtime": 130.4172,
-      "eval_samples_per_second": 4.079,
-      "eval_steps_per_second": 0.514,
-      "eval_wer": 0.9850228193096355,
+      "eval_loss": 2.6784112453460693,
+      "eval_runtime": 127.2435,
+      "eval_samples_per_second": 4.181,
+      "eval_steps_per_second": 0.527,
+      "eval_wer": 0.9682458057466092,
       "step": 1512
     },
     {
       "epoch": 21.11,
-      "learning_rate": 4.0021126760563385e-05,
-      "loss": 1.6554,
+      "learning_rate": 4e-05,
+      "loss": 1.2798,
       "step": 1520
     },
     {
       "epoch": 21.25,
-      "learning_rate": 3.9950704225352114e-05,
-      "loss": 1.5313,
+      "learning_rate": 3.992957746478874e-05,
+      "loss": 1.2006,
       "step": 1530
     },
     {
       "epoch": 21.39,
-      "learning_rate": 3.988028169014085e-05,
-      "loss": 1.5128,
+      "learning_rate": 3.985915492957747e-05,
+      "loss": 1.2242,
       "step": 1540
     },
     {
       "epoch": 21.53,
-      "learning_rate": 3.980985915492958e-05,
-      "loss": 1.5099,
+      "learning_rate": 3.97887323943662e-05,
+      "loss": 1.2146,
       "step": 1550
     },
     {
       "epoch": 21.66,
-      "learning_rate": 3.973943661971831e-05,
-      "loss": 1.5616,
+      "learning_rate": 3.971830985915493e-05,
+      "loss": 1.2338,
       "step": 1560
     },
     {
       "epoch": 21.8,
-      "learning_rate": 3.966901408450704e-05,
-      "loss": 1.5503,
+      "learning_rate": 3.9647887323943666e-05,
+      "loss": 1.2157,
       "step": 1570
     },
     {
       "epoch": 21.94,
-      "learning_rate": 3.959859154929578e-05,
-      "loss": 1.495,
+      "learning_rate": 3.9577464788732395e-05,
+      "loss": 1.195,
       "step": 1580
     },
     {
       "epoch": 22.0,
-      "eval_loss": 5.213191986083984,
-      "eval_runtime": 130.3188,
-      "eval_samples_per_second": 4.082,
-      "eval_steps_per_second": 0.514,
-      "eval_wer": 0.988879604036768,
+      "eval_loss": 3.039508104324341,
+      "eval_runtime": 126.6407,
+      "eval_samples_per_second": 4.201,
+      "eval_steps_per_second": 0.529,
+      "eval_wer": 0.963874783055859,
       "step": 1584
     },
     {
       "epoch": 22.08,
-      "learning_rate": 3.952816901408451e-05,
-      "loss": 1.5799,
+      "learning_rate": 3.950704225352112e-05,
+      "loss": 1.2549,
       "step": 1590
     },
     {
       "epoch": 22.22,
-      "learning_rate": 3.945774647887324e-05,
-      "loss": 1.4768,
+      "learning_rate": 3.943661971830986e-05,
+      "loss": 1.2158,
       "step": 1600
     },
     {
       "epoch": 22.36,
-      "learning_rate": 3.9387323943661976e-05,
-      "loss": 1.5627,
+      "learning_rate": 3.9366197183098594e-05,
+      "loss": 1.2259,
       "step": 1610
     },
     {
       "epoch": 22.5,
-      "learning_rate": 3.9316901408450704e-05,
-      "loss": 1.502,
+      "learning_rate": 3.929577464788732e-05,
+      "loss": 1.1934,
       "step": 1620
     },
     {
       "epoch": 22.64,
-      "learning_rate": 3.924647887323944e-05,
-      "loss": 1.5224,
+      "learning_rate": 3.922535211267606e-05,
+      "loss": 1.235,
       "step": 1630
     },
     {
       "epoch": 22.78,
-      "learning_rate": 3.9176056338028175e-05,
-      "loss": 1.4793,
+      "learning_rate": 3.9154929577464786e-05,
+      "loss": 1.2188,
       "step": 1640
     },
     {
       "epoch": 22.91,
-      "learning_rate": 3.91056338028169e-05,
-      "loss": 1.5676,
+      "learning_rate": 3.908450704225352e-05,
+      "loss": 1.2047,
       "step": 1650
     },
     {
       "epoch": 23.0,
-      "eval_loss": 4.159689426422119,
-      "eval_runtime": 130.5889,
-      "eval_samples_per_second": 4.074,
-      "eval_steps_per_second": 0.513,
-      "eval_wer": 0.9805875168734332,
+      "eval_loss": 3.193483352661133,
+      "eval_runtime": 126.5244,
+      "eval_samples_per_second": 4.205,
+      "eval_steps_per_second": 0.53,
+      "eval_wer": 0.9725525486919072,
       "step": 1656
     },
     {
       "epoch": 23.06,
-      "learning_rate": 3.903521126760564e-05,
-      "loss": 1.5326,
+      "learning_rate": 3.9014084507042256e-05,
+      "loss": 1.2639,
       "step": 1660
     },
     {
       "epoch": 23.19,
-      "learning_rate": 3.896478873239437e-05,
-      "loss": 1.4812,
+      "learning_rate": 3.8943661971830985e-05,
+      "loss": 1.2001,
       "step": 1670
     },
     {
       "epoch": 23.33,
-      "learning_rate": 3.88943661971831e-05,
-      "loss": 1.5815,
+      "learning_rate": 3.887323943661972e-05,
+      "loss": 1.1907,
       "step": 1680
     },
     {
       "epoch": 23.47,
-      "learning_rate": 3.882394366197184e-05,
-      "loss": 1.4233,
+      "learning_rate": 3.880281690140845e-05,
+      "loss": 1.1901,
       "step": 1690
     },
     {
       "epoch": 23.61,
-      "learning_rate": 3.8753521126760566e-05,
-      "loss": 1.528,
+      "learning_rate": 3.8732394366197184e-05,
+      "loss": 1.2286,
       "step": 1700
     },
     {
       "epoch": 23.75,
-      "learning_rate": 3.86830985915493e-05,
-      "loss": 1.4451,
+      "learning_rate": 3.866197183098592e-05,
+      "loss": 1.1921,
       "step": 1710
     },
     {
       "epoch": 23.89,
-      "learning_rate": 3.861267605633803e-05,
-      "loss": 1.5837,
+      "learning_rate": 3.859154929577465e-05,
+      "loss": 1.2306,
       "step": 1720
     },
     {
       "epoch": 24.0,
-      "eval_loss": 5.098143577575684,
-      "eval_runtime": 130.0277,
-      "eval_samples_per_second": 4.091,
-      "eval_steps_per_second": 0.515,
-      "eval_wer": 0.986694092691393,
+      "eval_loss": 3.2648956775665283,
+      "eval_runtime": 126.6373,
+      "eval_samples_per_second": 4.201,
+      "eval_steps_per_second": 0.529,
+      "eval_wer": 0.9722954297100983,
       "step": 1728
     },
     {
       "epoch": 24.03,
-      "learning_rate": 3.8542253521126765e-05,
-      "loss": 1.5209,
+      "learning_rate": 3.852112676056338e-05,
+      "loss": 1.2327,
       "step": 1730
     },
     {
       "epoch": 24.17,
-      "learning_rate": 3.84718309859155e-05,
-      "loss": 1.5073,
+      "learning_rate": 3.845070422535211e-05,
+      "loss": 1.2164,
       "step": 1740
     },
     {
       "epoch": 24.3,
-      "learning_rate": 3.840140845070423e-05,
-      "loss": 1.5739,
+      "learning_rate": 3.8380281690140847e-05,
+      "loss": 1.231,
       "step": 1750
     },
     {
       "epoch": 24.44,
-      "learning_rate": 3.8330985915492964e-05,
-      "loss": 1.4875,
+      "learning_rate": 3.8309859154929575e-05,
+      "loss": 1.1602,
       "step": 1760
     },
     {
       "epoch": 24.58,
-      "learning_rate": 3.826056338028169e-05,
-      "loss": 1.5407,
+      "learning_rate": 3.823943661971831e-05,
+      "loss": 1.2379,
       "step": 1770
     },
     {
       "epoch": 24.72,
-      "learning_rate": 3.819014084507043e-05,
-      "loss": 1.4561,
+      "learning_rate": 3.8169014084507046e-05,
+      "loss": 1.2011,
       "step": 1780
     },
     {
       "epoch": 24.86,
-      "learning_rate": 3.8119718309859156e-05,
-      "loss": 1.5288,
+      "learning_rate": 3.8098591549295774e-05,
+      "loss": 1.2345,
       "step": 1790
     },
     {
       "epoch": 25.0,
-      "learning_rate": 3.804929577464789e-05,
-      "loss": 1.4437,
+      "learning_rate": 3.802816901408451e-05,
+      "loss": 1.199,
       "step": 1800
     },
     {
       "epoch": 25.0,
-      "eval_loss": 3.6790771484375,
-      "eval_runtime": 130.0598,
-      "eval_samples_per_second": 4.09,
-      "eval_steps_per_second": 0.515,
-      "eval_wer": 0.9812945940734075,
+      "eval_loss": 3.137786865234375,
+      "eval_runtime": 126.7378,
+      "eval_samples_per_second": 4.198,
+      "eval_steps_per_second": 0.529,
+      "eval_wer": 0.9645175805103812,
       "step": 1800
     },
     {
       "epoch": 25.14,
-      "learning_rate": 3.797887323943663e-05,
-      "loss": 1.5601,
+      "learning_rate": 3.795774647887324e-05,
+      "loss": 1.2533,
       "step": 1810
     },
     {
       "epoch": 25.28,
-      "learning_rate": 3.790845070422535e-05,
-      "loss": 1.6149,
+      "learning_rate": 3.788732394366197e-05,
+      "loss": 1.1875,
       "step": 1820
     },
     {
       "epoch": 25.42,
-      "learning_rate": 3.7838028169014084e-05,
-      "loss": 1.4645,
+      "learning_rate": 3.781690140845071e-05,
+      "loss": 1.2041,
       "step": 1830
     },
     {
       "epoch": 25.55,
-      "learning_rate": 3.776760563380282e-05,
-      "loss": 1.492,
+      "learning_rate": 3.774647887323944e-05,
+      "loss": 1.1978,
       "step": 1840
     },
     {
       "epoch": 25.69,
-      "learning_rate": 3.769718309859155e-05,
-      "loss": 1.4875,
+      "learning_rate": 3.767605633802817e-05,
+      "loss": 1.1949,
       "step": 1850
     },
     {
       "epoch": 25.83,
-      "learning_rate": 3.762676056338028e-05,
-      "loss": 1.5415,
+      "learning_rate": 3.76056338028169e-05,
+      "loss": 1.2128,
       "step": 1860
     },
     {
       "epoch": 25.97,
-      "learning_rate": 3.755633802816901e-05,
-      "loss": 1.4684,
+      "learning_rate": 3.7535211267605636e-05,
+      "loss": 1.1945,
       "step": 1870
     },
     {
       "epoch": 26.0,
-      "eval_loss": 4.066702842712402,
-      "eval_runtime": 129.879,
-      "eval_samples_per_second": 4.096,
-      "eval_steps_per_second": 0.516,
-      "eval_wer": 0.9868869319277496,
+      "eval_loss": 2.8143184185028076,
+      "eval_runtime": 126.9454,
+      "eval_samples_per_second": 4.191,
+      "eval_steps_per_second": 0.528,
+      "eval_wer": 0.9595680401105612,
       "step": 1872
     },
     {
       "epoch": 26.11,
-      "learning_rate": 3.7485915492957746e-05,
-      "loss": 1.5748,
+      "learning_rate": 3.746478873239437e-05,
+      "loss": 1.2274,
       "step": 1880
     },
     {
       "epoch": 26.25,
-      "learning_rate": 3.741549295774648e-05,
-      "loss": 1.4697,
+      "learning_rate": 3.73943661971831e-05,
+      "loss": 1.1666,
       "step": 1890
     },
     {
       "epoch": 26.39,
-      "learning_rate": 3.734507042253521e-05,
-      "loss": 1.5225,
+      "learning_rate": 3.7323943661971835e-05,
+      "loss": 1.2108,
       "step": 1900
     },
     {
       "epoch": 26.53,
-      "learning_rate": 3.7274647887323945e-05,
-      "loss": 1.4683,
+      "learning_rate": 3.725352112676056e-05,
+      "loss": 1.2026,
       "step": 1910
     },
     {
       "epoch": 26.66,
-      "learning_rate": 3.7204225352112674e-05,
-      "loss": 1.446,
+      "learning_rate": 3.71830985915493e-05,
+      "loss": 1.185,
       "step": 1920
     },
     {
       "epoch": 26.8,
-      "learning_rate": 3.713380281690141e-05,
-      "loss": 1.5124,
+      "learning_rate": 3.711267605633803e-05,
+      "loss": 1.2213,
       "step": 1930
     },
     {
       "epoch": 26.94,
-      "learning_rate": 3.706338028169014e-05,
-      "loss": 1.4617,
+      "learning_rate": 3.704225352112676e-05,
+      "loss": 1.19,
       "step": 1940
     },
     {
       "epoch": 27.0,
-      "eval_loss": 5.2262139320373535,
-      "eval_runtime": 130.3549,
-      "eval_samples_per_second": 4.081,
-      "eval_steps_per_second": 0.514,
-      "eval_wer": 0.9902294786912644,
+      "eval_loss": 3.5174083709716797,
+      "eval_runtime": 126.9336,
+      "eval_samples_per_second": 4.191,
+      "eval_steps_per_second": 0.528,
+      "eval_wer": 0.9786591245098669,
       "step": 1944
     },
     {
       "epoch": 27.08,
-      "learning_rate": 3.699295774647887e-05,
-      "loss": 1.5184,
+      "learning_rate": 3.69718309859155e-05,
+      "loss": 1.2137,
       "step": 1950
     },
     {
       "epoch": 27.22,
-      "learning_rate": 3.692253521126761e-05,
-      "loss": 1.5419,
+      "learning_rate": 3.6901408450704226e-05,
+      "loss": 1.1797,
       "step": 1960
     },
     {
       "epoch": 27.36,
-      "learning_rate": 3.685211267605634e-05,
-      "loss": 1.5046,
+      "learning_rate": 3.683098591549296e-05,
+      "loss": 1.201,
       "step": 1970
     },
     {
       "epoch": 27.5,
-      "learning_rate": 3.678169014084507e-05,
-      "loss": 1.4151,
+      "learning_rate": 3.676056338028169e-05,
+      "loss": 1.1638,
       "step": 1980
     },
     {
       "epoch": 27.64,
-      "learning_rate": 3.67112676056338e-05,
-      "loss": 1.5072,
+      "learning_rate": 3.6690140845070425e-05,
+      "loss": 1.2258,
       "step": 1990
     },
     {
       "epoch": 27.78,
-      "learning_rate": 3.6640845070422536e-05,
-      "loss": 1.4767,
+      "learning_rate": 3.661971830985916e-05,
+      "loss": 1.1811,
       "step": 2000
     },
     {
       "epoch": 27.91,
-      "learning_rate": 3.657042253521127e-05,
-      "loss": 1.5259,
+      "learning_rate": 3.654929577464789e-05,
+      "loss": 1.1976,
       "step": 2010
     },
     {
       "epoch": 28.0,
-      "eval_loss": 4.5762858390808105,
-      "eval_runtime": 130.307,
-      "eval_samples_per_second": 4.083,
-      "eval_steps_per_second": 0.514,
-      "eval_wer": 0.9875940091277239,
+      "eval_loss": 2.9665908813476562,
+      "eval_runtime": 126.9918,
+      "eval_samples_per_second": 4.189,
+      "eval_steps_per_second": 0.528,
+      "eval_wer": 0.9593752008742046,
       "step": 2016
     },
     {
       "epoch": 28.06,
-      "learning_rate": 3.65e-05,
-      "loss": 1.5078,
+      "learning_rate": 3.648591549295775e-05,
+      "loss": 1.2387,
       "step": 2020
     },
     {
       "epoch": 28.19,
-      "learning_rate": 3.6429577464788735e-05,
-      "loss": 1.4085,
+      "learning_rate": 3.6415492957746485e-05,
+      "loss": 1.1871,
       "step": 2030
     },
     {
       "epoch": 28.33,
-      "learning_rate": 3.635915492957746e-05,
-      "loss": 1.4948,
+      "learning_rate": 3.634507042253521e-05,
+      "loss": 1.187,
       "step": 2040
     },
     {
       "epoch": 28.47,
-      "learning_rate": 3.62887323943662e-05,
-      "loss": 1.4401,
+      "learning_rate": 3.627464788732395e-05,
+      "loss": 1.1664,
       "step": 2050
     },
     {
       "epoch": 28.61,
-      "learning_rate": 3.6218309859154934e-05,
-      "loss": 1.5285,
+      "learning_rate": 3.620422535211268e-05,
+      "loss": 1.1957,
       "step": 2060
     },
     {
       "epoch": 28.75,
-      "learning_rate": 3.614788732394366e-05,
-      "loss": 1.4384,
+      "learning_rate": 3.6133802816901405e-05,
+      "loss": 1.1636,
       "step": 2070
     },
     {
       "epoch": 28.89,
-      "learning_rate": 3.60774647887324e-05,
-      "loss": 1.5127,
+      "learning_rate": 3.606338028169014e-05,
+      "loss": 1.2229,
       "step": 2080
     },
     {
       "epoch": 29.0,
-      "eval_loss": 3.5296952724456787,
-      "eval_runtime": 130.1354,
-      "eval_samples_per_second": 4.088,
-      "eval_steps_per_second": 0.515,
-      "eval_wer": 0.9830944269460693,
+      "eval_loss": 2.867180585861206,
+      "eval_runtime": 126.6524,
+      "eval_samples_per_second": 4.2,
+      "eval_steps_per_second": 0.529,
+      "eval_wer": 0.9588609629105869,
       "step": 2088
     },
     {
       "epoch": 29.03,
-      "learning_rate": 3.6007042253521126e-05,
-      "loss": 1.5654,
+      "learning_rate": 3.5992957746478876e-05,
+      "loss": 1.2067,
       "step": 2090
     },
     {
       "epoch": 29.17,
-      "learning_rate": 3.593661971830986e-05,
-      "loss": 1.4627,
+      "learning_rate": 3.5922535211267604e-05,
+      "loss": 1.1802,
       "step": 2100
     },
     {
       "epoch": 29.3,
-      "learning_rate": 3.5866197183098597e-05,
-      "loss": 1.4326,
+      "learning_rate": 3.585211267605634e-05,
+      "loss": 1.1715,
       "step": 2110
     },
     {
       "epoch": 29.44,
-      "learning_rate": 3.5795774647887325e-05,
-      "loss": 1.4128,
+      "learning_rate": 3.578169014084507e-05,
+      "loss": 1.1805,
       "step": 2120
     },
     {
       "epoch": 29.58,
-      "learning_rate": 3.572535211267606e-05,
-      "loss": 1.5034,
+      "learning_rate": 3.57112676056338e-05,
+      "loss": 1.1944,
       "step": 2130
     },
     {
       "epoch": 29.72,
-      "learning_rate": 3.565492957746479e-05,
-      "loss": 1.4244,
+      "learning_rate": 3.564084507042253e-05,
+      "loss": 1.1686,
       "step": 2140
     },
     {
       "epoch": 29.86,
-      "learning_rate": 3.5584507042253524e-05,
-      "loss": 1.4928,
+      "learning_rate": 3.557042253521127e-05,
+      "loss": 1.1822,
       "step": 2150
     },
     {
       "epoch": 30.0,
-      "learning_rate": 3.551408450704225e-05,
-      "loss": 1.4726,
+      "learning_rate": 3.55e-05,
+      "loss": 1.1548,
       "step": 2160
     },
     {
       "epoch": 30.0,
-      "eval_loss": 3.237917900085449,
-      "eval_runtime": 130.2259,
-      "eval_samples_per_second": 4.085,
-      "eval_steps_per_second": 0.514,
-      "eval_wer": 0.9776306485826316,
+      "eval_loss": 2.656834840774536,
+      "eval_runtime": 127.1543,
+      "eval_samples_per_second": 4.184,
+      "eval_steps_per_second": 0.527,
+      "eval_wer": 0.9627177476377193,
       "step": 2160
     },
     {
       "epoch": 30.14,
-      "learning_rate": 3.544366197183099e-05,
-      "loss": 1.5197,
+      "learning_rate": 3.542957746478873e-05,
+      "loss": 1.2254,
       "step": 2170
     },
     {
       "epoch": 30.28,
-      "learning_rate": 3.537323943661972e-05,
-      "loss": 1.4335,
+      "learning_rate": 3.5359154929577466e-05,
+      "loss": 1.176,
       "step": 2180
     },
     {
       "epoch": 30.42,
-      "learning_rate": 3.530281690140845e-05,
-      "loss": 1.4897,
+      "learning_rate": 3.5288732394366195e-05,
+      "loss": 1.1744,
       "step": 2190
     },
     {
       "epoch": 30.55,
-      "learning_rate": 3.523239436619719e-05,
-      "loss": 1.4403,
+      "learning_rate": 3.521830985915493e-05,
+      "loss": 1.1846,
       "step": 2200
     },
     {
       "epoch": 30.69,
-      "learning_rate": 3.5161971830985915e-05,
-      "loss": 1.4808,
+      "learning_rate": 3.5147887323943665e-05,
+      "loss": 1.1895,
       "step": 2210
     },
     {
       "epoch": 30.83,
-      "learning_rate": 3.509154929577465e-05,
-      "loss": 1.5281,
+      "learning_rate": 3.5077464788732394e-05,
+      "loss": 1.2154,
       "step": 2220
     },
     {
       "epoch": 30.97,
-      "learning_rate": 3.5021126760563386e-05,
-      "loss": 1.446,
+      "learning_rate": 3.500704225352113e-05,
+      "loss": 1.169,
       "step": 2230
     },
     {
       "epoch": 31.0,
-      "eval_loss": 3.9459383487701416,
-      "eval_runtime": 129.8148,
-      "eval_samples_per_second": 4.098,
-      "eval_steps_per_second": 0.516,
-      "eval_wer": 0.9834801054187825,
+      "eval_loss": 2.879929542541504,
+      "eval_runtime": 127.1506,
+      "eval_samples_per_second": 4.184,
+      "eval_steps_per_second": 0.527,
+      "eval_wer": 0.9653532172012599,
       "step": 2232
     },
     {
       "epoch": 31.11,
-      "learning_rate": 3.4950704225352114e-05,
-      "loss": 1.595,
+      "learning_rate": 3.493661971830986e-05,
+      "loss": 1.2204,
       "step": 2240
     },
     {
       "epoch": 31.25,
-      "learning_rate": 3.488028169014085e-05,
-      "loss": 1.3812,
+      "learning_rate": 3.486619718309859e-05,
+      "loss": 1.151,
       "step": 2250
     },
     {
       "epoch": 31.39,
-      "learning_rate": 3.480985915492958e-05,
-      "loss": 1.5127,
+      "learning_rate": 3.479577464788733e-05,
+      "loss": 1.1853,
       "step": 2260
     },
     {
       "epoch": 31.53,
-      "learning_rate": 3.473943661971831e-05,
-      "loss": 1.4908,
+      "learning_rate": 3.4725352112676056e-05,
+      "loss": 1.174,
       "step": 2270
     },
     {
       "epoch": 31.66,
-      "learning_rate": 3.466901408450705e-05,
-      "loss": 1.4987,
+      "learning_rate": 3.465492957746479e-05,
+      "loss": 1.1718,
       "step": 2280
     },
     {
       "epoch": 31.8,
-      "learning_rate": 3.459859154929578e-05,
-      "loss": 1.4607,
+      "learning_rate": 3.458450704225352e-05,
+      "loss": 1.1823,
       "step": 2290
     },
     {
       "epoch": 31.94,
-      "learning_rate": 3.452816901408451e-05,
-      "loss": 1.4116,
+      "learning_rate": 3.4514084507042255e-05,
+      "loss": 1.1857,
       "step": 2300
     },
     {
       "epoch": 32.0,
-      "eval_loss": 3.535161256790161,
-      "eval_runtime": 130.6285,
-      "eval_samples_per_second": 4.073,
-      "eval_steps_per_second": 0.513,
-      "eval_wer": 0.9814874333097641,
+      "eval_loss": 2.8691256046295166,
+      "eval_runtime": 127.1415,
+      "eval_samples_per_second": 4.184,
+      "eval_steps_per_second": 0.527,
+      "eval_wer": 0.9625249084013627,
       "step": 2304
     },
     {
       "epoch": 32.08,
-      "learning_rate": 3.445774647887324e-05,
-      "loss": 1.5251,
+      "learning_rate": 3.4443661971830984e-05,
+      "loss": 1.2006,
       "step": 2310
     },
     {
       "epoch": 32.22,
-      "learning_rate": 3.4387323943661976e-05,
-      "loss": 1.3876,
+      "learning_rate": 3.437323943661972e-05,
+      "loss": 1.1676,
       "step": 2320
     },
     {
       "epoch": 32.36,
-      "learning_rate": 3.4316901408450705e-05,
-      "loss": 1.5109,
+      "learning_rate": 3.4302816901408454e-05,
+      "loss": 1.1918,
       "step": 2330
     },
     {
       "epoch": 32.5,
-      "learning_rate": 3.424647887323944e-05,
-      "loss": 1.4014,
+      "learning_rate": 3.423239436619718e-05,
+      "loss": 1.1461,
       "step": 2340
     },
     {
       "epoch": 32.64,
-      "learning_rate": 3.4176056338028175e-05,
-      "loss": 1.5301,
+      "learning_rate": 3.416197183098592e-05,
+      "loss": 1.1791,
       "step": 2350
     },
     {
       "epoch": 32.78,
-      "learning_rate": 3.4105633802816904e-05,
-      "loss": 1.4588,
+      "learning_rate": 3.4091549295774647e-05,
+      "loss": 1.1853,
       "step": 2360
     },
     {
       "epoch": 32.91,
-      "learning_rate": 3.403521126760564e-05,
-      "loss": 1.462,
+      "learning_rate": 3.402112676056338e-05,
+      "loss": 1.1862,
       "step": 2370
     },
     {
       "epoch": 33.0,
-      "eval_loss": 3.3658456802368164,
-      "eval_runtime": 130.4437,
-      "eval_samples_per_second": 4.078,
-      "eval_steps_per_second": 0.514,
-      "eval_wer": 0.9760879346917786,
+      "eval_loss": 2.825140953063965,
+      "eval_runtime": 127.5529,
+      "eval_samples_per_second": 4.171,
+      "eval_steps_per_second": 0.525,
+      "eval_wer": 0.955518416147072,
       "step": 2376
     },
     {
       "epoch": 33.06,
-      "learning_rate": 3.396478873239437e-05,
-      "loss": 1.4869,
+      "learning_rate": 3.395070422535212e-05,
+      "loss": 1.2053,
       "step": 2380
     },
     {
       "epoch": 33.19,
-      "learning_rate": 3.38943661971831e-05,
-      "loss": 1.4332,
+      "learning_rate": 3.3880281690140846e-05,
+      "loss": 1.1704,
       "step": 2390
     },
     {
       "epoch": 33.33,
-      "learning_rate": 3.382394366197184e-05,
-      "loss": 1.4464,
+      "learning_rate": 3.380985915492958e-05,
+      "loss": 1.1749,
       "step": 2400
     },
     {
       "epoch": 33.47,
-      "learning_rate": 3.3753521126760566e-05,
-      "loss": 1.3993,
+      "learning_rate": 3.373943661971831e-05,
+      "loss": 1.1554,
       "step": 2410
     },
     {
       "epoch": 33.61,
-      "learning_rate": 3.3683098591549295e-05,
-      "loss": 1.4726,
+      "learning_rate": 3.3669014084507045e-05,
+      "loss": 1.1958,
       "step": 2420
     },
     {
       "epoch": 33.75,
-      "learning_rate": 3.361267605633803e-05,
-      "loss": 1.4268,
+      "learning_rate": 3.359859154929578e-05,
+      "loss": 1.156,
       "step": 2430
     },
     {
       "epoch": 33.89,
-      "learning_rate": 3.354225352112676e-05,
-      "loss": 1.498,
+      "learning_rate": 3.352816901408451e-05,
+      "loss": 1.1721,
       "step": 2440
     },
     {
       "epoch": 34.0,
-      "eval_loss": 3.862046718597412,
-      "eval_runtime": 131.0934,
-      "eval_samples_per_second": 4.058,
-      "eval_steps_per_second": 0.511,
-      "eval_wer": 0.9794947612007456,
+      "eval_loss": 3.596752405166626,
+      "eval_runtime": 127.0297,
+      "eval_samples_per_second": 4.188,
+      "eval_steps_per_second": 0.527,
+      "eval_wer": 0.9725525486919072,
       "step": 2448
     },
     {
       "epoch": 34.03,
-      "learning_rate": 3.3471830985915494e-05,
-      "loss": 1.4754,
+      "learning_rate": 3.3457746478873244e-05,
+      "loss": 1.2103,
       "step": 2450
     },
     {
       "epoch": 34.17,
-      "learning_rate": 3.340140845070422e-05,
-      "loss": 1.3934,
+      "learning_rate": 3.338732394366197e-05,
+      "loss": 1.1833,
       "step": 2460
     },
     {
       "epoch": 34.3,
-      "learning_rate": 3.333098591549296e-05,
-      "loss": 1.4224,
+      "learning_rate": 3.331690140845071e-05,
+      "loss": 1.1502,
       "step": 2470
     },
     {
       "epoch": 34.44,
-      "learning_rate": 3.326056338028169e-05,
-      "loss": 1.4338,
+      "learning_rate": 3.324647887323944e-05,
+      "loss": 1.1613,
       "step": 2480
     },
     {
       "epoch": 34.58,
-      "learning_rate": 3.319014084507042e-05,
-      "loss": 1.5069,
+      "learning_rate": 3.317605633802817e-05,
+      "loss": 1.1949,
       "step": 2490
     },
     {
       "epoch": 34.72,
-      "learning_rate": 3.311971830985916e-05,
-      "loss": 1.4498,
+      "learning_rate": 3.3105633802816906e-05,
+      "loss": 1.1708,
       "step": 2500
     },
     {
       "epoch": 34.86,
-      "learning_rate": 3.3049295774647885e-05,
-      "loss": 1.5188,
+      "learning_rate": 3.3035211267605635e-05,
+      "loss": 1.2031,
       "step": 2510
     },
     {
       "epoch": 35.0,
-      "learning_rate": 3.297887323943662e-05,
-      "loss": 1.382,
+      "learning_rate": 3.296478873239437e-05,
+      "loss": 1.1293,
       "step": 2520
     },
     {
       "epoch": 35.0,
-      "eval_loss": 3.473414182662964,
-      "eval_runtime": 130.3567,
-      "eval_samples_per_second": 4.081,
-      "eval_steps_per_second": 0.514,
-      "eval_wer": 0.9797518801825544,
+      "eval_loss": 3.4130284786224365,
+      "eval_runtime": 126.7659,
+      "eval_samples_per_second": 4.197,
+      "eval_steps_per_second": 0.529,
+      "eval_wer": 0.9650960982194511,
       "step": 2520
     },
     {
       "epoch": 35.14,
-      "learning_rate": 3.290845070422535e-05,
-      "loss": 1.5214,
+      "learning_rate": 3.28943661971831e-05,
+      "loss": 1.1864,
       "step": 2530
     },
     {
       "epoch": 35.28,
-      "learning_rate": 3.2838028169014084e-05,
-      "loss": 1.4588,
+      "learning_rate": 3.2823943661971834e-05,
+      "loss": 1.1632,
       "step": 2540
     },
     {
       "epoch": 35.42,
-      "learning_rate": 3.276760563380282e-05,
-      "loss": 1.4542,
+      "learning_rate": 3.275352112676057e-05,
+      "loss": 1.1728,
       "step": 2550
     },
     {
       "epoch": 35.55,
-      "learning_rate": 3.269718309859155e-05,
-      "loss": 1.449,
+      "learning_rate": 3.26830985915493e-05,
+      "loss": 1.1831,
       "step": 2560
     },
     {
       "epoch": 35.69,
-      "learning_rate": 3.262676056338028e-05,
-      "loss": 1.4157,
+      "learning_rate": 3.261267605633803e-05,
+      "loss": 1.1664,
       "step": 2570
     },
     {
       "epoch": 35.83,
-      "learning_rate": 3.255633802816901e-05,
-      "loss": 1.4518,
+      "learning_rate": 3.254225352112676e-05,
+      "loss": 1.1644,
       "step": 2580
     },
     {
       "epoch": 35.97,
-      "learning_rate": 3.248591549295775e-05,
-      "loss": 1.3969,
+      "learning_rate": 3.24718309859155e-05,
+      "loss": 1.1513,
       "step": 2590
     },
     {
       "epoch": 36.0,
-      "eval_loss": 3.7284700870513916,
-      "eval_runtime": 130.4462,
-      "eval_samples_per_second": 4.078,
-      "eval_steps_per_second": 0.514,
-      "eval_wer": 0.9780163270553449,
+      "eval_loss": 2.8804469108581543,
+      "eval_runtime": 126.7553,
+      "eval_samples_per_second": 4.197,
+      "eval_steps_per_second": 0.529,
+      "eval_wer": 0.9629748666195281,
       "step": 2592
     },
     {
       "epoch": 36.11,
-      "learning_rate": 3.241549295774648e-05,
-      "loss": 1.4896,
+      "learning_rate": 3.240140845070423e-05,
+      "loss": 1.1767,
       "step": 2600
     },
     {
       "epoch": 36.25,
-      "learning_rate": 3.234507042253521e-05,
-      "loss": 1.3764,
+      "learning_rate": 3.233098591549296e-05,
+      "loss": 1.1267,
       "step": 2610
     },
     {
       "epoch": 36.39,
-      "learning_rate": 3.2274647887323946e-05,
-      "loss": 1.4335,
+      "learning_rate": 3.2260563380281696e-05,
+      "loss": 1.1499,
       "step": 2620
     },
     {
       "epoch": 36.53,
-      "learning_rate": 3.2204225352112674e-05,
-      "loss": 1.4429,
+      "learning_rate": 3.2190140845070424e-05,
+      "loss": 1.1703,
       "step": 2630
     },
     {
       "epoch": 36.66,
-      "learning_rate": 3.213380281690141e-05,
-      "loss": 1.4849,
+      "learning_rate": 3.211971830985916e-05,
+      "loss": 1.1565,
       "step": 2640
     },
     {
       "epoch": 36.8,
-      "learning_rate": 3.2063380281690145e-05,
-      "loss": 1.4526,
+      "learning_rate": 3.204929577464789e-05,
+      "loss": 1.1573,
       "step": 2650
     },
     {
       "epoch": 36.94,
-      "learning_rate": 3.1992957746478873e-05,
-      "loss": 1.3945,
+      "learning_rate": 3.1978873239436616e-05,
+      "loss": 1.1537,
       "step": 2660
     },
     {
       "epoch": 37.0,
-      "eval_loss": 3.725541353225708,
-      "eval_runtime": 130.2241,
-      "eval_samples_per_second": 4.085,
-      "eval_steps_per_second": 0.514,
-      "eval_wer": 0.9820016712733818,
+      "eval_loss": 2.582429885864258,
+      "eval_runtime": 127.0846,
+      "eval_samples_per_second": 4.186,
+      "eval_steps_per_second": 0.527,
+      "eval_wer": 0.9575110882560905,
       "step": 2664
     },
     {
       "epoch": 37.08,
-      "learning_rate": 3.192253521126761e-05,
-      "loss": 1.4576,
+      "learning_rate": 3.190845070422535e-05,
+      "loss": 1.1845,
       "step": 2670
     },
     {
       "epoch": 37.22,
-      "learning_rate": 3.185211267605634e-05,
-      "loss": 1.3657,
+      "learning_rate": 3.183802816901409e-05,
+      "loss": 1.1543,
       "step": 2680
     },
     {
       "epoch": 37.36,
-      "learning_rate": 3.178169014084507e-05,
-      "loss": 1.4399,
+      "learning_rate": 3.1767605633802815e-05,
+      "loss": 1.1641,
       "step": 2690
     },
     {
       "epoch": 37.5,
-      "learning_rate": 3.17112676056338e-05,
-      "loss": 1.3822,
+      "learning_rate": 3.169718309859155e-05,
+      "loss": 1.1547,
       "step": 2700
     },
     {
       "epoch": 37.64,
-      "learning_rate": 3.1640845070422536e-05,
-      "loss": 1.4901,
+      "learning_rate": 3.162676056338028e-05,
+      "loss": 1.1611,
       "step": 2710
     },
     {
       "epoch": 37.78,
-      "learning_rate": 3.157042253521127e-05,
-      "loss": 1.4733,
+      "learning_rate": 3.1556338028169014e-05,
+      "loss": 1.1515,
       "step": 2720
     },
     {
       "epoch": 37.91,
-      "learning_rate": 3.15e-05,
-      "loss": 1.4616,
+      "learning_rate": 3.148591549295774e-05,
+      "loss": 1.1818,
       "step": 2730
     },
     {
       "epoch": 38.0,
-      "eval_loss": 3.878848075866699,
-      "eval_runtime": 130.4377,
-      "eval_samples_per_second": 4.079,
-      "eval_steps_per_second": 0.514,
-      "eval_wer": 0.9812303143279553,
+      "eval_loss": 2.844338893890381,
+      "eval_runtime": 127.2865,
+      "eval_samples_per_second": 4.18,
+      "eval_steps_per_second": 0.526,
+      "eval_wer": 0.9613035932377708,
       "step": 2736
     },
     {
       "epoch": 38.06,
-      "learning_rate": 3.1429577464788735e-05,
-      "loss": 1.4583,
+      "learning_rate": 3.141549295774648e-05,
+      "loss": 1.1751,
       "step": 2740
     },
     {
       "epoch": 38.19,
-      "learning_rate": 3.1359154929577464e-05,
-      "loss": 1.3932,
+      "learning_rate": 3.1345070422535213e-05,
+      "loss": 1.1363,
       "step": 2750
     },
     {
       "epoch": 38.33,
-      "learning_rate": 3.12887323943662e-05,
-      "loss": 1.4282,
+      "learning_rate": 3.127464788732394e-05,
+      "loss": 1.1511,
       "step": 2760
     },
     {
       "epoch": 38.47,
-      "learning_rate": 3.1218309859154934e-05,
-      "loss": 1.3665,
+      "learning_rate": 3.120422535211268e-05,
+      "loss": 1.1406,
       "step": 2770
     },
     {
       "epoch": 38.61,
-      "learning_rate": 3.114788732394366e-05,
-      "loss": 1.5116,
+      "learning_rate": 3.1133802816901406e-05,
+      "loss": 1.1566,
       "step": 2780
     },
     {
       "epoch": 38.75,
-      "learning_rate": 3.10774647887324e-05,
-      "loss": 1.4327,
+      "learning_rate": 3.106338028169014e-05,
+      "loss": 1.159,
       "step": 2790
     },
     {
       "epoch": 38.89,
-      "learning_rate": 3.1007042253521126e-05,
-      "loss": 1.4791,
+      "learning_rate": 3.0992957746478876e-05,
+      "loss": 1.1835,
       "step": 2800
     },
     {
       "epoch": 39.0,
-      "eval_loss": 3.189089298248291,
-      "eval_runtime": 130.1914,
-      "eval_samples_per_second": 4.086,
-      "eval_steps_per_second": 0.515,
-      "eval_wer": 0.9780806068007971,
+      "eval_loss": 2.6431169509887695,
+      "eval_runtime": 127.4608,
+      "eval_samples_per_second": 4.174,
+      "eval_steps_per_second": 0.526,
+      "eval_wer": 0.9619463906922928,
       "step": 2808
     },
     {
       "epoch": 39.03,
-      "learning_rate": 3.093661971830986e-05,
-      "loss": 1.4395,
+      "learning_rate": 3.0922535211267605e-05,
+      "loss": 1.19,
       "step": 2810
     },
     {
       "epoch": 39.17,
-      "learning_rate": 3.08661971830986e-05,
-      "loss": 1.3824,
+      "learning_rate": 3.085211267605634e-05,
+      "loss": 1.1424,
       "step": 2820
     },
     {
       "epoch": 39.3,
-      "learning_rate": 3.0795774647887325e-05,
-      "loss": 1.4604,
+      "learning_rate": 3.078169014084507e-05,
+      "loss": 1.1494,
       "step": 2830
     },
     {
       "epoch": 39.44,
-      "learning_rate": 3.072535211267606e-05,
-      "loss": 1.4136,
+      "learning_rate": 3.0711267605633804e-05,
+      "loss": 1.153,
       "step": 2840
     },
     {
       "epoch": 39.58,
-      "learning_rate": 3.065492957746479e-05,
-      "loss": 1.42,
+      "learning_rate": 3.064084507042254e-05,
+      "loss": 1.1729,
       "step": 2850
     },
     {
       "epoch": 39.72,
-      "learning_rate": 3.0584507042253524e-05,
-      "loss": 1.3876,
+      "learning_rate": 3.057042253521127e-05,
+      "loss": 1.1471,
       "step": 2860
     },
     {
       "epoch": 39.86,
-      "learning_rate": 3.0514084507042256e-05,
-      "loss": 1.4762,
+      "learning_rate": 3.05e-05,
+      "loss": 1.1594,
       "step": 2870
     },
     {
       "epoch": 40.0,
-      "learning_rate": 3.0443661971830988e-05,
-      "loss": 1.3875,
+      "learning_rate": 3.0429577464788735e-05,
+      "loss": 1.1457,
       "step": 2880
     },
     {
       "epoch": 40.0,
-      "eval_loss": 3.6854491233825684,
-      "eval_runtime": 130.6968,
-      "eval_samples_per_second": 4.07,
-      "eval_steps_per_second": 0.513,
-      "eval_wer": 0.9800732789098155,
+      "eval_loss": 2.925424098968506,
+      "eval_runtime": 127.1921,
+      "eval_samples_per_second": 4.183,
+      "eval_steps_per_second": 0.527,
+      "eval_wer": 0.963874783055859,
       "step": 2880
     },
     {
       "epoch": 40.14,
-      "learning_rate": 3.037323943661972e-05,
-      "loss": 1.4414,
+      "learning_rate": 3.0359154929577467e-05,
+      "loss": 1.1912,
       "step": 2890
     },
     {
       "epoch": 40.28,
-      "learning_rate": 3.0302816901408455e-05,
-      "loss": 1.3793,
+      "learning_rate": 3.02887323943662e-05,
+      "loss": 1.13,
       "step": 2900
     },
     {
       "epoch": 40.42,
-      "learning_rate": 3.0232394366197187e-05,
-      "loss": 1.4228,
+      "learning_rate": 3.021830985915493e-05,
+      "loss": 1.1434,
       "step": 2910
     },
     {
       "epoch": 40.55,
-      "learning_rate": 3.016197183098592e-05,
-      "loss": 1.403,
+      "learning_rate": 3.0147887323943662e-05,
+      "loss": 1.151,
       "step": 2920
     },
     {
       "epoch": 40.69,
-      "learning_rate": 3.009154929577465e-05,
-      "loss": 1.3963,
+      "learning_rate": 3.0077464788732397e-05,
+      "loss": 1.1432,
       "step": 2930
     },
     {
       "epoch": 40.83,
-      "learning_rate": 3.0021126760563383e-05,
-      "loss": 1.4233,
+      "learning_rate": 3.000704225352113e-05,
+      "loss": 1.1518,
       "step": 2940
     },
     {
       "epoch": 40.97,
-      "learning_rate": 2.9950704225352115e-05,
-      "loss": 1.3914,
+      "learning_rate": 2.993661971830986e-05,
+      "loss": 1.1591,
       "step": 2950
     },
     {
       "epoch": 41.0,
-      "eval_loss": 3.4011056423187256,
-      "eval_runtime": 130.7999,
-      "eval_samples_per_second": 4.067,
-      "eval_steps_per_second": 0.512,
-      "eval_wer": 0.9742238220736645,
+      "eval_loss": 2.8194425106048584,
+      "eval_runtime": 127.6605,
+      "eval_samples_per_second": 4.167,
+      "eval_steps_per_second": 0.525,
+      "eval_wer": 0.9560969338561419,
       "step": 2952
     },
     {
       "epoch": 41.11,
-      "learning_rate": 2.988028169014085e-05,
-      "loss": 1.4879,
+      "learning_rate": 2.9866197183098593e-05,
+      "loss": 1.193,
       "step": 2960
     },
     {
       "epoch": 41.25,
-      "learning_rate": 2.9809859154929582e-05,
-      "loss": 1.3831,
+      "learning_rate": 2.9795774647887325e-05,
+      "loss": 1.1611,
       "step": 2970
     },
     {
       "epoch": 41.39,
-      "learning_rate": 2.9739436619718314e-05,
-      "loss": 1.4669,
+      "learning_rate": 2.9725352112676057e-05,
+      "loss": 1.1543,
       "step": 2980
     },
     {
       "epoch": 41.53,
-      "learning_rate": 2.9669014084507046e-05,
-      "loss": 1.3642,
+      "learning_rate": 2.9654929577464792e-05,
+      "loss": 1.1137,
       "step": 2990
     },
     {
       "epoch": 41.66,
-      "learning_rate": 2.9598591549295778e-05,
-      "loss": 1.4006,
+      "learning_rate": 2.9584507042253524e-05,
+      "loss": 1.1464,
       "step": 3000
     },
     {
       "epoch": 41.8,
-      "learning_rate": 2.9528169014084506e-05,
-      "loss": 1.4594,
+      "learning_rate": 2.9514084507042256e-05,
+      "loss": 1.1452,
       "step": 3010
     },
     {
       "epoch": 41.94,
-      "learning_rate": 2.9457746478873238e-05,
-      "loss": 1.3744,
+      "learning_rate": 2.9443661971830988e-05,
+      "loss": 1.1284,
       "step": 3020
     },
     {
       "epoch": 42.0,
-      "eval_loss": 3.5792057514190674,
-      "eval_runtime": 130.4782,
-      "eval_samples_per_second": 4.077,
-      "eval_steps_per_second": 0.513,
-      "eval_wer": 0.9808446358552421,
+      "eval_loss": 2.643159866333008,
+      "eval_runtime": 127.5772,
+      "eval_samples_per_second": 4.17,
+      "eval_steps_per_second": 0.525,
+      "eval_wer": 0.9805875168734332,
       "step": 3024
     },
     {
       "epoch": 42.08,
-      "learning_rate": 2.938732394366197e-05,
-      "loss": 1.4386,
+      "learning_rate": 2.937323943661972e-05,
+      "loss": 1.1777,
       "step": 3030
     },
     {
       "epoch": 42.22,
-      "learning_rate": 2.93169014084507e-05,
-      "loss": 1.3701,
+      "learning_rate": 2.930281690140845e-05,
+      "loss": 1.1343,
       "step": 3040
     },
     {
       "epoch": 42.36,
-      "learning_rate": 2.9246478873239437e-05,
-      "loss": 1.4619,
+      "learning_rate": 2.9232394366197187e-05,
+      "loss": 1.1599,
       "step": 3050
     },
     {
       "epoch": 42.5,
-      "learning_rate": 2.917605633802817e-05,
-      "loss": 1.3623,
+      "learning_rate": 2.916197183098592e-05,
+      "loss": 1.1131,
       "step": 3060
     },
     {
       "epoch": 42.64,
-      "learning_rate": 2.91056338028169e-05,
-      "loss": 1.3855,
+      "learning_rate": 2.909154929577465e-05,
+      "loss": 1.1535,
       "step": 3070
     },
     {
       "epoch": 42.78,
-      "learning_rate": 2.9035211267605633e-05,
-      "loss": 1.3731,
+      "learning_rate": 2.9021126760563382e-05,
+      "loss": 1.1454,
       "step": 3080
     },
     {
       "epoch": 42.91,
-      "learning_rate": 2.8964788732394364e-05,
-      "loss": 1.4085,
+      "learning_rate": 2.8950704225352114e-05,
+      "loss": 1.1602,
       "step": 3090
     },
     {
       "epoch": 43.0,
-      "eval_loss": 2.9806365966796875,
-      "eval_runtime": 130.7636,
-      "eval_samples_per_second": 4.068,
-      "eval_steps_per_second": 0.512,
-      "eval_wer": 0.9742881018191168,
+      "eval_loss": 2.4279065132141113,
+      "eval_runtime": 127.4359,
+      "eval_samples_per_second": 4.175,
+      "eval_steps_per_second": 0.526,
+      "eval_wer": 1.0087420453815004,
       "step": 3096
     },
     {
       "epoch": 43.06,
-      "learning_rate": 2.88943661971831e-05,
-      "loss": 1.3944,
+      "learning_rate": 2.888028169014085e-05,
+      "loss": 1.1669,
       "step": 3100
     },
     {
       "epoch": 43.19,
-      "learning_rate": 2.882394366197183e-05,
-      "loss": 1.3556,
+      "learning_rate": 2.880985915492958e-05,
+      "loss": 1.1278,
       "step": 3110
     },
     {
       "epoch": 43.33,
-      "learning_rate": 2.8753521126760563e-05,
-      "loss": 1.415,
+      "learning_rate": 2.8739436619718313e-05,
+      "loss": 1.1531,
       "step": 3120
     },
     {
       "epoch": 43.47,
-      "learning_rate": 2.8683098591549295e-05,
-      "loss": 1.3848,
+      "learning_rate": 2.8669014084507045e-05,
+      "loss": 1.1365,
       "step": 3130
     },
     {
       "epoch": 43.61,
-      "learning_rate": 2.8612676056338027e-05,
-      "loss": 1.468,
+      "learning_rate": 2.8598591549295777e-05,
+      "loss": 1.1681,
       "step": 3140
     },
     {
       "epoch": 43.75,
-      "learning_rate": 2.854225352112676e-05,
-      "loss": 1.3518,
+      "learning_rate": 2.852816901408451e-05,
+      "loss": 1.1311,
       "step": 3150
     },
     {
       "epoch": 43.89,
-      "learning_rate": 2.8471830985915494e-05,
-      "loss": 1.407,
+      "learning_rate": 2.8457746478873244e-05,
+      "loss": 1.1556,
       "step": 3160
     },
     {
       "epoch": 44.0,
-      "eval_loss": 3.3188397884368896,
-      "eval_runtime": 130.8549,
-      "eval_samples_per_second": 4.066,
-      "eval_steps_per_second": 0.512,
-      "eval_wer": 0.9821302307642862,
+      "eval_loss": 2.503970146179199,
+      "eval_runtime": 127.6942,
+      "eval_samples_per_second": 4.166,
+      "eval_steps_per_second": 0.525,
+      "eval_wer": 1.0030211480362539,
       "step": 3168
     },
     {
       "epoch": 44.03,
-      "learning_rate": 2.8401408450704226e-05,
-      "loss": 1.4181,
+      "learning_rate": 2.8387323943661976e-05,
+      "loss": 1.1653,
       "step": 3170
     },
     {
       "epoch": 44.17,
-      "learning_rate": 2.8330985915492958e-05,
-      "loss": 1.3522,
+      "learning_rate": 2.8316901408450708e-05,
+      "loss": 1.1135,
       "step": 3180
     },
     {
       "epoch": 44.3,
-      "learning_rate": 2.826056338028169e-05,
-      "loss": 1.3983,
+      "learning_rate": 2.824647887323944e-05,
+      "loss": 1.1454,
       "step": 3190
     },
     {
       "epoch": 44.44,
-      "learning_rate": 2.8190140845070422e-05,
-      "loss": 1.4091,
+      "learning_rate": 2.817605633802817e-05,
+      "loss": 1.1401,
       "step": 3200
     },
     {
       "epoch": 44.58,
-      "learning_rate": 2.8119718309859157e-05,
-      "loss": 1.4292,
+      "learning_rate": 2.8105633802816907e-05,
+      "loss": 1.1465,
       "step": 3210
     },
     {
       "epoch": 44.72,
-      "learning_rate": 2.804929577464789e-05,
-      "loss": 1.348,
+      "learning_rate": 2.803521126760564e-05,
+      "loss": 1.1378,
       "step": 3220
     },
     {
       "epoch": 44.86,
-      "learning_rate": 2.797887323943662e-05,
-      "loss": 1.3977,
+      "learning_rate": 2.796478873239437e-05,
+      "loss": 1.1362,
       "step": 3230
     },
     {
       "epoch": 45.0,
-      "learning_rate": 2.7908450704225353e-05,
-      "loss": 1.3678,
+      "learning_rate": 2.7894366197183096e-05,
+      "loss": 1.1256,
       "step": 3240
     },
     {
       "epoch": 45.0,
-      "eval_loss": 3.6300089359283447,
-      "eval_runtime": 130.6888,
-      "eval_samples_per_second": 4.071,
-      "eval_steps_per_second": 0.513,
-      "eval_wer": 0.9816802725461207,
+      "eval_loss": 3.1641151905059814,
+      "eval_runtime": 128.4723,
+      "eval_samples_per_second": 4.141,
+      "eval_steps_per_second": 0.522,
+      "eval_wer": 0.9607893552741531,
       "step": 3240
     },
     {
       "epoch": 45.14,
-      "learning_rate": 2.7838028169014085e-05,
-      "loss": 1.43,
+      "learning_rate": 2.782394366197183e-05,
+      "loss": 1.166,
       "step": 3250
     },
     {
       "epoch": 45.28,
-      "learning_rate": 2.7767605633802816e-05,
-      "loss": 1.3639,
+      "learning_rate": 2.7753521126760563e-05,
+      "loss": 1.1158,
       "step": 3260
     },
     {
       "epoch": 45.42,
-      "learning_rate": 2.7697183098591552e-05,
-      "loss": 1.3611,
+      "learning_rate": 2.7683098591549295e-05,
+      "loss": 1.111,
       "step": 3270
     },
     {
       "epoch": 45.55,
-      "learning_rate": 2.7626760563380284e-05,
-      "loss": 1.4093,
+      "learning_rate": 2.7612676056338027e-05,
+      "loss": 1.127,
       "step": 3280
     },
     {
       "epoch": 45.69,
-      "learning_rate": 2.7556338028169015e-05,
-      "loss": 1.3651,
+      "learning_rate": 2.754225352112676e-05,
+      "loss": 1.1143,
       "step": 3290
     },
     {
       "epoch": 45.83,
-      "learning_rate": 2.7485915492957747e-05,
-      "loss": 1.4214,
+      "learning_rate": 2.7471830985915494e-05,
+      "loss": 1.1177,
       "step": 3300
     },
     {
       "epoch": 45.97,
-      "learning_rate": 2.741549295774648e-05,
-      "loss": 1.3762,
+      "learning_rate": 2.7401408450704226e-05,
+      "loss": 1.1256,
       "step": 3310
     },
     {
       "epoch": 46.0,
-      "eval_loss": 3.098646640777588,
-      "eval_runtime": 130.2199,
-      "eval_samples_per_second": 4.085,
-      "eval_steps_per_second": 0.515,
-      "eval_wer": 0.977437809346275,
+      "eval_loss": 2.9522297382354736,
+      "eval_runtime": 127.8911,
+      "eval_samples_per_second": 4.16,
+      "eval_steps_per_second": 0.524,
+      "eval_wer": 0.9676672880375393,
       "step": 3312
     },
     {
       "epoch": 46.11,
-      "learning_rate": 2.734507042253521e-05,
-      "loss": 1.421,
+      "learning_rate": 2.7330985915492957e-05,
+      "loss": 1.1654,
       "step": 3320
     },
     {
       "epoch": 46.25,
-      "learning_rate": 2.7274647887323946e-05,
-      "loss": 1.3568,
+      "learning_rate": 2.726056338028169e-05,
+      "loss": 1.1123,
       "step": 3330
     },
     {
       "epoch": 46.39,
-      "learning_rate": 2.7204225352112678e-05,
-      "loss": 1.3897,
+      "learning_rate": 2.719014084507042e-05,
+      "loss": 1.1265,
       "step": 3340
     },
     {
       "epoch": 46.53,
-      "learning_rate": 2.713380281690141e-05,
-      "loss": 1.3806,
+      "learning_rate": 2.7119718309859153e-05,
+      "loss": 1.12,
       "step": 3350
     },
     {
       "epoch": 46.66,
-      "learning_rate": 2.7063380281690142e-05,
-      "loss": 1.3745,
+      "learning_rate": 2.704929577464789e-05,
+      "loss": 1.1198,
       "step": 3360
     },
     {
       "epoch": 46.8,
-      "learning_rate": 2.6992957746478874e-05,
-      "loss": 1.3942,
+      "learning_rate": 2.697887323943662e-05,
+      "loss": 1.1124,
       "step": 3370
     },
     {
       "epoch": 46.94,
-      "learning_rate": 2.692253521126761e-05,
-      "loss": 1.3507,
+      "learning_rate": 2.6908450704225352e-05,
+      "loss": 1.1211,
       "step": 3380
     },
     {
       "epoch": 47.0,
-      "eval_loss": 2.8956308364868164,
-      "eval_runtime": 131.1936,
-      "eval_samples_per_second": 4.055,
-      "eval_steps_per_second": 0.511,
-      "eval_wer": 0.9732596258918814,
+      "eval_loss": 2.6318135261535645,
+      "eval_runtime": 127.2211,
+      "eval_samples_per_second": 4.182,
+      "eval_steps_per_second": 0.527,
+      "eval_wer": 0.957961046474256,
       "step": 3384
     },
     {
       "epoch": 47.08,
-      "learning_rate": 2.685211267605634e-05,
-      "loss": 1.4291,
+      "learning_rate": 2.6838028169014084e-05,
+      "loss": 1.1519,
       "step": 3390
     },
     {
       "epoch": 47.22,
-      "learning_rate": 2.6781690140845073e-05,
-      "loss": 1.3325,
+      "learning_rate": 2.6767605633802816e-05,
+      "loss": 1.1076,
       "step": 3400
     },
     {
       "epoch": 47.36,
-      "learning_rate": 2.6711267605633805e-05,
-      "loss": 1.419,
+      "learning_rate": 2.6697183098591548e-05,
+      "loss": 1.1234,
       "step": 3410
     },
     {
       "epoch": 47.5,
-      "learning_rate": 2.6640845070422537e-05,
-      "loss": 1.362,
+      "learning_rate": 2.6626760563380283e-05,
+      "loss": 1.096,
       "step": 3420
     },
     {
       "epoch": 47.64,
-      "learning_rate": 2.657042253521127e-05,
-      "loss": 1.3878,
+      "learning_rate": 2.6556338028169015e-05,
+      "loss": 1.1399,
       "step": 3430
     },
     {
       "epoch": 47.78,
-      "learning_rate": 2.6500000000000004e-05,
-      "loss": 1.3695,
+      "learning_rate": 2.6485915492957747e-05,
+      "loss": 1.1256,
       "step": 3440
     },
     {
       "epoch": 47.91,
-      "learning_rate": 2.6429577464788736e-05,
-      "loss": 1.3646,
+      "learning_rate": 2.641549295774648e-05,
+      "loss": 1.1142,
       "step": 3450
     },
     {
       "epoch": 48.0,
-      "eval_loss": 2.836045503616333,
-      "eval_runtime": 130.9286,
-      "eval_samples_per_second": 4.063,
-      "eval_steps_per_second": 0.512,
-      "eval_wer": 0.9730667866555248,
+      "eval_loss": 2.72981858253479,
+      "eval_runtime": 126.8959,
+      "eval_samples_per_second": 4.192,
+      "eval_steps_per_second": 0.528,
+      "eval_wer": 0.953332904801697,
       "step": 3456
     },
     {
       "epoch": 48.06,
-      "learning_rate": 2.6359154929577468e-05,
-      "loss": 1.404,
+      "learning_rate": 2.634507042253521e-05,
+      "loss": 1.1586,
       "step": 3460
     },
     {
       "epoch": 48.19,
-      "learning_rate": 2.62887323943662e-05,
-      "loss": 1.3409,
+      "learning_rate": 2.6274647887323946e-05,
+      "loss": 1.122,
       "step": 3470
     },
     {
       "epoch": 48.33,
-      "learning_rate": 2.621830985915493e-05,
-      "loss": 1.3885,
+      "learning_rate": 2.6204225352112678e-05,
+      "loss": 1.1289,
       "step": 3480
     },
     {
       "epoch": 48.47,
-      "learning_rate": 2.6147887323943667e-05,
-      "loss": 1.3468,
+      "learning_rate": 2.613380281690141e-05,
+      "loss": 1.1061,
       "step": 3490
     },
     {
       "epoch": 48.61,
-      "learning_rate": 2.60774647887324e-05,
-      "loss": 1.3811,
+      "learning_rate": 2.606338028169014e-05,
+      "loss": 1.125,
       "step": 3500
     },
     {
       "epoch": 48.75,
-      "learning_rate": 2.600704225352113e-05,
-      "loss": 1.3525,
+      "learning_rate": 2.5992957746478873e-05,
+      "loss": 1.1044,
       "step": 3510
     },
     {
       "epoch": 48.89,
-      "learning_rate": 2.5936619718309862e-05,
-      "loss": 1.4027,
+      "learning_rate": 2.5922535211267605e-05,
+      "loss": 1.1237,
       "step": 3520
     },
     {
       "epoch": 49.0,
-      "eval_loss": 3.518127918243408,
-      "eval_runtime": 130.7441,
-      "eval_samples_per_second": 4.069,
-      "eval_steps_per_second": 0.512,
-      "eval_wer": 0.9744166613100212,
+      "eval_loss": 2.5441715717315674,
+      "eval_runtime": 127.7445,
+      "eval_samples_per_second": 4.165,
+      "eval_steps_per_second": 0.524,
+      "eval_wer": 0.9673458893102783,
       "step": 3528
     },
     {
       "epoch": 49.03,
-      "learning_rate": 2.5866197183098594e-05,
-      "loss": 1.4143,
+      "learning_rate": 2.585211267605634e-05,
+      "loss": 1.1566,
       "step": 3530
     },
     {
       "epoch": 49.17,
-      "learning_rate": 2.5795774647887326e-05,
-      "loss": 1.3482,
+      "learning_rate": 2.5781690140845072e-05,
+      "loss": 1.1107,
       "step": 3540
     },
     {
       "epoch": 49.3,
-      "learning_rate": 2.572535211267606e-05,
-      "loss": 1.379,
+      "learning_rate": 2.5711267605633804e-05,
+      "loss": 1.1394,
       "step": 3550
     },
     {
       "epoch": 49.44,
-      "learning_rate": 2.5654929577464793e-05,
-      "loss": 1.4315,
+      "learning_rate": 2.5640845070422536e-05,
+      "loss": 1.1306,
       "step": 3560
     },
     {
       "epoch": 49.58,
-      "learning_rate": 2.5584507042253525e-05,
-      "loss": 1.389,
+      "learning_rate": 2.5570422535211268e-05,
+      "loss": 1.1309,
       "step": 3570
     },
     {
       "epoch": 49.72,
-      "learning_rate": 2.5514084507042257e-05,
-      "loss": 1.3345,
+      "learning_rate": 2.5500000000000003e-05,
+      "loss": 1.1092,
       "step": 3580
     },
     {
       "epoch": 49.86,
-      "learning_rate": 2.544366197183099e-05,
-      "loss": 1.3777,
+      "learning_rate": 2.5429577464788735e-05,
+      "loss": 1.1353,
       "step": 3590
     },
     {
       "epoch": 50.0,
-      "learning_rate": 2.5373239436619717e-05,
-      "loss": 1.3383,
+      "learning_rate": 2.5359154929577467e-05,
+      "loss": 1.0976,
       "step": 3600
     },
     {
       "epoch": 50.0,
-      "eval_loss": 3.27095627784729,
-      "eval_runtime": 131.6783,
-      "eval_samples_per_second": 4.04,
-      "eval_steps_per_second": 0.509,
-      "eval_wer": 0.9768592916372051,
+      "eval_loss": 2.77671480178833,
+      "eval_runtime": 127.064,
+      "eval_samples_per_second": 4.187,
+      "eval_steps_per_second": 0.527,
+      "eval_wer": 0.9609821945105097,
       "step": 3600
     },
     {
       "epoch": 50.14,
-      "learning_rate": 2.530281690140845e-05,
-      "loss": 1.4179,
+      "learning_rate": 2.52887323943662e-05,
+      "loss": 1.1575,
       "step": 3610
     },
     {
       "epoch": 50.28,
-      "learning_rate": 2.523239436619718e-05,
-      "loss": 1.4183,
+      "learning_rate": 2.521830985915493e-05,
+      "loss": 1.0924,
       "step": 3620
     },
     {
       "epoch": 50.42,
-      "learning_rate": 2.5161971830985913e-05,
-      "loss": 1.3428,
+      "learning_rate": 2.5147887323943663e-05,
+      "loss": 1.1045,
       "step": 3630
     },
     {
       "epoch": 50.55,
-      "learning_rate": 2.5091549295774648e-05,
-      "loss": 1.3861,
+      "learning_rate": 2.5077464788732398e-05,
+      "loss": 1.1188,
       "step": 3640
     },
     {
       "epoch": 50.69,
-      "learning_rate": 2.502112676056338e-05,
-      "loss": 1.3596,
+      "learning_rate": 2.500704225352113e-05,
+      "loss": 1.1055,
       "step": 3650
     },
     {
       "epoch": 50.83,
-      "learning_rate": 2.4950704225352115e-05,
-      "loss": 1.3923,
+      "learning_rate": 2.493661971830986e-05,
+      "loss": 1.1335,
       "step": 3660
     },
     {
       "epoch": 50.97,
-      "learning_rate": 2.4880281690140847e-05,
-      "loss": 1.3517,
+      "learning_rate": 2.4866197183098593e-05,
+      "loss": 1.1154,
       "step": 3670
     },
     {
       "epoch": 51.0,
-      "eval_loss": 2.997406244277954,
-      "eval_runtime": 129.9638,
-      "eval_samples_per_second": 4.093,
-      "eval_steps_per_second": 0.516,
-      "eval_wer": 0.975573696728161,
+      "eval_loss": 2.684926986694336,
+      "eval_runtime": 127.3605,
+      "eval_samples_per_second": 4.177,
+      "eval_steps_per_second": 0.526,
+      "eval_wer": 0.9645818602558334,
       "step": 3672
     },
     {
       "epoch": 51.11,
-      "learning_rate": 2.480985915492958e-05,
-      "loss": 1.4026,
+      "learning_rate": 2.4795774647887325e-05,
+      "loss": 1.1468,
       "step": 3680
     },
     {
       "epoch": 51.25,
-      "learning_rate": 2.473943661971831e-05,
-      "loss": 1.3317,
+      "learning_rate": 2.4725352112676057e-05,
+      "loss": 1.0912,
       "step": 3690
     },
     {
       "epoch": 51.39,
-      "learning_rate": 2.4669014084507046e-05,
-      "loss": 1.3968,
+      "learning_rate": 2.465492957746479e-05,
+      "loss": 1.1212,
       "step": 3700
     },
     {
       "epoch": 51.53,
-      "learning_rate": 2.4598591549295778e-05,
-      "loss": 1.3303,
+      "learning_rate": 2.458450704225352e-05,
+      "loss": 1.1276,
       "step": 3710
     },
     {
       "epoch": 51.66,
-      "learning_rate": 2.452816901408451e-05,
-      "loss": 1.3477,
+      "learning_rate": 2.4514084507042253e-05,
+      "loss": 1.1093,
       "step": 3720
     },
     {
       "epoch": 51.8,
-      "learning_rate": 2.4457746478873242e-05,
-      "loss": 1.3425,
+      "learning_rate": 2.4443661971830985e-05,
+      "loss": 1.1032,
       "step": 3730
     },
     {
       "epoch": 51.94,
-      "learning_rate": 2.4387323943661974e-05,
-      "loss": 1.3551,
+      "learning_rate": 2.437323943661972e-05,
+      "loss": 1.1012,
       "step": 3740
     },
     {
       "epoch": 52.0,
-      "eval_loss": 3.3179967403411865,
-      "eval_runtime": 131.3646,
-      "eval_samples_per_second": 4.05,
-      "eval_steps_per_second": 0.51,
-      "eval_wer": 0.9751237385099955,
+      "eval_loss": 2.538395643234253,
+      "eval_runtime": 128.8802,
+      "eval_samples_per_second": 4.128,
+      "eval_steps_per_second": 0.52,
+      "eval_wer": 0.9620749501831972,
       "step": 3744
     },
     {
       "epoch": 52.08,
-      "learning_rate": 2.4316901408450705e-05,
-      "loss": 1.3972,
+      "learning_rate": 2.4302816901408452e-05,
+      "loss": 1.1407,
       "step": 3750
     },
     {
       "epoch": 52.22,
-      "learning_rate": 2.4246478873239437e-05,
-      "loss": 1.3394,
+      "learning_rate": 2.4232394366197184e-05,
+      "loss": 1.1212,
       "step": 3760
     },
     {
       "epoch": 52.36,
-      "learning_rate": 2.417605633802817e-05,
-      "loss": 1.3707,
+      "learning_rate": 2.4161971830985916e-05,
+      "loss": 1.1136,
       "step": 3770
     },
     {
       "epoch": 52.5,
-      "learning_rate": 2.41056338028169e-05,
-      "loss": 1.3463,
+      "learning_rate": 2.4091549295774647e-05,
+      "loss": 1.1121,
       "step": 3780
     },
     {
       "epoch": 52.64,
-      "learning_rate": 2.4035211267605633e-05,
-      "loss": 1.3658,
+      "learning_rate": 2.4021126760563383e-05,
+      "loss": 1.1137,
       "step": 3790
     },
     {
       "epoch": 52.78,
-      "learning_rate": 2.3964788732394365e-05,
-      "loss": 1.358,
+      "learning_rate": 2.3950704225352115e-05,
+      "loss": 1.0869,
       "step": 3800
     },
     {
       "epoch": 52.91,
-      "learning_rate": 2.38943661971831e-05,
-      "loss": 1.3493,
+      "learning_rate": 2.3880281690140846e-05,
+      "loss": 1.1077,
       "step": 3810
     },
     {
       "epoch": 53.0,
-      "eval_loss": 2.6087403297424316,
-      "eval_runtime": 130.8988,
-      "eval_samples_per_second": 4.064,
-      "eval_steps_per_second": 0.512,
-      "eval_wer": 0.9994214822909301,
+      "eval_loss": 2.450518846511841,
+      "eval_runtime": 127.7053,
+      "eval_samples_per_second": 4.166,
+      "eval_steps_per_second": 0.525,
+      "eval_wer": 1.006749373272482,
       "step": 3816
     },
     {
       "epoch": 53.06,
-      "learning_rate": 2.3823943661971832e-05,
-      "loss": 1.4151,
+      "learning_rate": 2.380985915492958e-05,
+      "loss": 1.1386,
       "step": 3820
     },
     {
       "epoch": 53.19,
-      "learning_rate": 2.3753521126760564e-05,
-      "loss": 1.3187,
+      "learning_rate": 2.373943661971831e-05,
+      "loss": 1.0952,
       "step": 3830
     },
     {
       "epoch": 53.33,
-      "learning_rate": 2.3683098591549296e-05,
-      "loss": 1.3759,
+      "learning_rate": 2.3669014084507042e-05,
+      "loss": 1.1137,
       "step": 3840
     },
     {
       "epoch": 53.47,
-      "learning_rate": 2.3612676056338028e-05,
-      "loss": 1.3332,
+      "learning_rate": 2.3598591549295777e-05,
+      "loss": 1.1057,
       "step": 3850
     },
     {
       "epoch": 53.61,
-      "learning_rate": 2.3542253521126763e-05,
-      "loss": 1.3438,
+      "learning_rate": 2.352816901408451e-05,
+      "loss": 1.1082,
       "step": 3860
     },
     {
       "epoch": 53.75,
-      "learning_rate": 2.3471830985915495e-05,
-      "loss": 1.3296,
+      "learning_rate": 2.345774647887324e-05,
+      "loss": 1.1082,
       "step": 3870
     },
     {
       "epoch": 53.89,
-      "learning_rate": 2.3401408450704227e-05,
-      "loss": 1.3977,
+      "learning_rate": 2.3387323943661973e-05,
+      "loss": 1.0936,
       "step": 3880
     },
     {
       "epoch": 54.0,
-      "eval_loss": 2.6875009536743164,
-      "eval_runtime": 132.1432,
-      "eval_samples_per_second": 4.026,
-      "eval_steps_per_second": 0.507,
-      "eval_wer": 0.9778234878189882,
+      "eval_loss": 2.5847084522247314,
+      "eval_runtime": 127.5732,
+      "eval_samples_per_second": 4.17,
+      "eval_steps_per_second": 0.525,
+      "eval_wer": 0.9686957639647747,
       "step": 3888
     },
     {
       "epoch": 54.03,
-      "learning_rate": 2.333098591549296e-05,
-      "loss": 1.3677,
+      "learning_rate": 2.3316901408450705e-05,
+      "loss": 1.1251,
       "step": 3890
     },
     {
       "epoch": 54.17,
-      "learning_rate": 2.326056338028169e-05,
-      "loss": 1.3172,
+      "learning_rate": 2.3246478873239437e-05,
+      "loss": 1.1062,
       "step": 3900
     },
     {
       "epoch": 54.3,
-      "learning_rate": 2.3190140845070422e-05,
-      "loss": 1.362,
+      "learning_rate": 2.3176056338028172e-05,
+      "loss": 1.1175,
       "step": 3910
     },
     {
       "epoch": 54.44,
-      "learning_rate": 2.3119718309859158e-05,
-      "loss": 1.3375,
+      "learning_rate": 2.3105633802816904e-05,
+      "loss": 1.0885,
       "step": 3920
     },
     {
       "epoch": 54.58,
-      "learning_rate": 2.304929577464789e-05,
-      "loss": 1.3644,
+      "learning_rate": 2.3035211267605636e-05,
+      "loss": 1.1181,
       "step": 3930
     },
     {
       "epoch": 54.72,
-      "learning_rate": 2.297887323943662e-05,
-      "loss": 1.3165,
+      "learning_rate": 2.2964788732394368e-05,
+      "loss": 1.074,
       "step": 3940
     },
     {
       "epoch": 54.86,
-      "learning_rate": 2.2908450704225353e-05,
-      "loss": 1.3723,
+      "learning_rate": 2.28943661971831e-05,
+      "loss": 1.0951,
       "step": 3950
     },
     {
       "epoch": 55.0,
-      "learning_rate": 2.2838028169014085e-05,
-      "loss": 1.3254,
+      "learning_rate": 2.2823943661971835e-05,
+      "loss": 1.0772,
       "step": 3960
     },
     {
       "epoch": 55.0,
-      "eval_loss": 3.3662502765655518,
-      "eval_runtime": 130.7472,
-      "eval_samples_per_second": 4.069,
-      "eval_steps_per_second": 0.512,
-      "eval_wer": 0.9787234042553191,
+      "eval_loss": 2.4575040340423584,
+      "eval_runtime": 127.5148,
+      "eval_samples_per_second": 4.172,
+      "eval_steps_per_second": 0.525,
+      "eval_wer": 0.9760879346917786,
       "step": 3960
     },
     {
       "epoch": 55.14,
-      "learning_rate": 2.276760563380282e-05,
-      "loss": 1.3937,
+      "learning_rate": 2.2753521126760567e-05,
+      "loss": 1.1337,
       "step": 3970
     },
     {
       "epoch": 55.28,
-      "learning_rate": 2.2697183098591552e-05,
-      "loss": 1.3371,
+      "learning_rate": 2.2683098591549295e-05,
+      "loss": 1.1033,
       "step": 3980
     },
     {
       "epoch": 55.42,
-      "learning_rate": 2.2626760563380284e-05,
-      "loss": 1.3096,
+      "learning_rate": 2.2612676056338027e-05,
+      "loss": 1.0893,
       "step": 3990
     },
     {
       "epoch": 55.55,
-      "learning_rate": 2.2556338028169016e-05,
-      "loss": 1.3341,
+      "learning_rate": 2.254225352112676e-05,
+      "loss": 1.0987,
       "step": 4000
     },
     {
       "epoch": 55.69,
-      "learning_rate": 2.2485915492957748e-05,
-      "loss": 1.3173,
+      "learning_rate": 2.2471830985915494e-05,
+      "loss": 1.0999,
       "step": 4010
     },
     {
       "epoch": 55.83,
-      "learning_rate": 2.241549295774648e-05,
-      "loss": 1.3174,
+      "learning_rate": 2.2401408450704226e-05,
+      "loss": 1.0823,
       "step": 4020
     },
     {
       "epoch": 55.97,
-      "learning_rate": 2.2345070422535215e-05,
-      "loss": 1.303,
+      "learning_rate": 2.2330985915492958e-05,
+      "loss": 1.092,
       "step": 4030
     },
     {
       "epoch": 56.0,
-      "eval_loss": 2.6154274940490723,
-      "eval_runtime": 131.383,
-      "eval_samples_per_second": 4.049,
-      "eval_steps_per_second": 0.51,
-      "eval_wer": 0.9692100019283924,
+      "eval_loss": 2.488936185836792,
+      "eval_runtime": 127.2806,
+      "eval_samples_per_second": 4.18,
+      "eval_steps_per_second": 0.526,
+      "eval_wer": 0.98020183840072,
       "step": 4032
     },
     {
       "epoch": 56.11,
-      "learning_rate": 2.2274647887323943e-05,
-      "loss": 1.337,
+      "learning_rate": 2.226056338028169e-05,
+      "loss": 1.1073,
       "step": 4040
     },
     {
       "epoch": 56.25,
-      "learning_rate": 2.2204225352112675e-05,
-      "loss": 1.2723,
+      "learning_rate": 2.219014084507042e-05,
+      "loss": 1.0755,
       "step": 4050
     },
     {
       "epoch": 56.39,
-      "learning_rate": 2.2133802816901407e-05,
-      "loss": 1.3505,
+      "learning_rate": 2.2119718309859157e-05,
+      "loss": 1.1181,
       "step": 4060
     },
     {
       "epoch": 56.53,
-      "learning_rate": 2.2063380281690142e-05,
-      "loss": 1.3265,
+      "learning_rate": 2.204929577464789e-05,
+      "loss": 1.0712,
       "step": 4070
     },
     {
       "epoch": 56.66,
-      "learning_rate": 2.1992957746478874e-05,
-      "loss": 1.2951,
+      "learning_rate": 2.197887323943662e-05,
+      "loss": 1.098,
       "step": 4080
     },
     {
       "epoch": 56.8,
-      "learning_rate": 2.1922535211267606e-05,
-      "loss": 1.3154,
+      "learning_rate": 2.1908450704225353e-05,
+      "loss": 1.1074,
       "step": 4090
     },
     {
       "epoch": 56.94,
-      "learning_rate": 2.1852112676056338e-05,
-      "loss": 1.2612,
+      "learning_rate": 2.1838028169014084e-05,
+      "loss": 1.0868,
       "step": 4100
     },
     {
       "epoch": 57.0,
-      "eval_loss": 3.2773284912109375,
-      "eval_runtime": 131.5337,
-      "eval_samples_per_second": 4.045,
-      "eval_steps_per_second": 0.509,
-      "eval_wer": 0.9760879346917786,
+      "eval_loss": 2.5885348320007324,
+      "eval_runtime": 127.2611,
+      "eval_samples_per_second": 4.18,
+      "eval_steps_per_second": 0.526,
+      "eval_wer": 0.9664459728739474,
       "step": 4104
     },
     {
       "epoch": 57.08,
-      "learning_rate": 2.178169014084507e-05,
-      "loss": 1.3172,
+      "learning_rate": 2.1767605633802816e-05,
+      "loss": 1.0985,
       "step": 4110
     },
     {
       "epoch": 57.22,
-      "learning_rate": 2.1711267605633802e-05,
-      "loss": 1.2591,
+      "learning_rate": 2.169718309859155e-05,
+      "loss": 1.0843,
       "step": 4120
     },
     {
       "epoch": 57.36,
-      "learning_rate": 2.1640845070422537e-05,
-      "loss": 1.3305,
+      "learning_rate": 2.1626760563380283e-05,
+      "loss": 1.0861,
       "step": 4130
     },
     {
       "epoch": 57.5,
-      "learning_rate": 2.157042253521127e-05,
-      "loss": 1.2637,
+      "learning_rate": 2.1563380281690144e-05,
+      "loss": 1.1,
       "step": 4140
     },
     {
       "epoch": 57.64,
-      "learning_rate": 2.15e-05,
-      "loss": 1.3198,
+      "learning_rate": 2.1492957746478876e-05,
+      "loss": 1.0991,
       "step": 4150
     },
     {
       "epoch": 57.78,
-      "learning_rate": 2.1429577464788733e-05,
-      "loss": 1.2798,
+      "learning_rate": 2.1422535211267604e-05,
+      "loss": 1.072,
       "step": 4160
     },
     {
       "epoch": 57.91,
-      "learning_rate": 2.1359154929577465e-05,
-      "loss": 1.3161,
+      "learning_rate": 2.135211267605634e-05,
+      "loss": 1.0979,
       "step": 4170
     },
     {
       "epoch": 58.0,
-      "eval_loss": 2.902003765106201,
-      "eval_runtime": 131.0979,
-      "eval_samples_per_second": 4.058,
-      "eval_steps_per_second": 0.511,
-      "eval_wer": 0.9717169120010285,
+      "eval_loss": 2.636953353881836,
+      "eval_runtime": 127.0998,
+      "eval_samples_per_second": 4.186,
+      "eval_steps_per_second": 0.527,
+      "eval_wer": 0.9606607957832487,
       "step": 4176
     },
     {
       "epoch": 58.06,
-      "learning_rate": 2.12887323943662e-05,
-      "loss": 1.3129,
+      "learning_rate": 2.128169014084507e-05,
+      "loss": 1.1102,
       "step": 4180
     },
     {
       "epoch": 58.19,
-      "learning_rate": 2.1218309859154932e-05,
-      "loss": 1.2534,
+      "learning_rate": 2.1211267605633803e-05,
+      "loss": 1.0738,
       "step": 4190
     },
     {
       "epoch": 58.33,
-      "learning_rate": 2.1147887323943664e-05,
-      "loss": 1.3106,
+      "learning_rate": 2.1140845070422535e-05,
+      "loss": 1.0783,
       "step": 4200
     },
     {
       "epoch": 58.47,
-      "learning_rate": 2.1077464788732395e-05,
-      "loss": 1.2513,
+      "learning_rate": 2.1070422535211267e-05,
+      "loss": 1.0796,
       "step": 4210
     },
     {
       "epoch": 58.61,
-      "learning_rate": 2.1007042253521127e-05,
-      "loss": 1.2999,
+      "learning_rate": 2.1e-05,
+      "loss": 1.0833,
       "step": 4220
     },
     {
       "epoch": 58.75,
-      "learning_rate": 2.093661971830986e-05,
-      "loss": 1.2241,
+      "learning_rate": 2.0929577464788734e-05,
+      "loss": 1.0659,
       "step": 4230
     },
     {
       "epoch": 58.89,
-      "learning_rate": 2.0866197183098594e-05,
-      "loss": 1.3026,
+      "learning_rate": 2.0859154929577466e-05,
+      "loss": 1.094,
       "step": 4240
     },
     {
       "epoch": 59.0,
-      "eval_loss": 3.151747703552246,
-      "eval_runtime": 131.6083,
-      "eval_samples_per_second": 4.042,
-      "eval_steps_per_second": 0.509,
-      "eval_wer": 0.9708812753101498,
+      "eval_loss": 2.619495153427124,
+      "eval_runtime": 127.8141,
+      "eval_samples_per_second": 4.162,
+      "eval_steps_per_second": 0.524,
+      "eval_wer": 0.9605322362923443,
       "step": 4248
     },
     {
       "epoch": 59.03,
-      "learning_rate": 2.0795774647887326e-05,
-      "loss": 1.3299,
+      "learning_rate": 2.0788732394366198e-05,
+      "loss": 1.1189,
       "step": 4250
     },
     {
       "epoch": 59.17,
-      "learning_rate": 2.0725352112676058e-05,
-      "loss": 1.2785,
+      "learning_rate": 2.071830985915493e-05,
+      "loss": 1.0937,
       "step": 4260
     },
     {
       "epoch": 59.3,
-      "learning_rate": 2.065492957746479e-05,
-      "loss": 1.2908,
+      "learning_rate": 2.064788732394366e-05,
+      "loss": 1.0792,
       "step": 4270
     },
     {
       "epoch": 59.44,
-      "learning_rate": 2.0584507042253522e-05,
-      "loss": 1.2814,
+      "learning_rate": 2.0577464788732393e-05,
+      "loss": 1.0852,
       "step": 4280
     },
     {
       "epoch": 59.58,
-      "learning_rate": 2.0514084507042254e-05,
-      "loss": 1.2841,
+      "learning_rate": 2.050704225352113e-05,
+      "loss": 1.0955,
       "step": 4290
     },
     {
       "epoch": 59.72,
-      "learning_rate": 2.044366197183099e-05,
-      "loss": 1.2436,
+      "learning_rate": 2.043661971830986e-05,
+      "loss": 1.0697,
       "step": 4300
     },
     {
       "epoch": 59.86,
-      "learning_rate": 2.037323943661972e-05,
-      "loss": 1.3191,
+      "learning_rate": 2.0366197183098592e-05,
+      "loss": 1.0777,
       "step": 4310
     },
     {
       "epoch": 60.0,
-      "learning_rate": 2.0302816901408453e-05,
-      "loss": 1.262,
+      "learning_rate": 2.0295774647887324e-05,
+      "loss": 1.0745,
       "step": 4320
     },
     {
       "epoch": 60.0,
-      "eval_loss": 3.129873752593994,
-      "eval_runtime": 130.7498,
-      "eval_samples_per_second": 4.069,
-      "eval_steps_per_second": 0.512,
-      "eval_wer": 0.9674744488011827,
+      "eval_loss": 2.534641981124878,
+      "eval_runtime": 128.8728,
+      "eval_samples_per_second": 4.128,
+      "eval_steps_per_second": 0.52,
+      "eval_wer": 0.9834158256733303,
       "step": 4320
     },
     {
       "epoch": 60.14,
-      "learning_rate": 2.0232394366197185e-05,
-      "loss": 1.3176,
+      "learning_rate": 2.0225352112676056e-05,
+      "loss": 1.1234,
       "step": 4330
     },
     {
       "epoch": 60.28,
-      "learning_rate": 2.0161971830985917e-05,
-      "loss": 1.271,
+      "learning_rate": 2.015492957746479e-05,
+      "loss": 1.0765,
       "step": 4340
     },
     {
       "epoch": 60.42,
-      "learning_rate": 2.009154929577465e-05,
-      "loss": 1.2509,
+      "learning_rate": 2.0084507042253523e-05,
+      "loss": 1.0785,
       "step": 4350
     },
     {
       "epoch": 60.55,
-      "learning_rate": 2.002112676056338e-05,
-      "loss": 1.2607,
+      "learning_rate": 2.0014084507042255e-05,
+      "loss": 1.074,
       "step": 4360
     },
     {
       "epoch": 60.69,
-      "learning_rate": 1.9950704225352112e-05,
-      "loss": 1.2664,
+      "learning_rate": 1.9943661971830987e-05,
+      "loss": 1.0766,
       "step": 4370
     },
     {
       "epoch": 60.83,
-      "learning_rate": 1.9880281690140844e-05,
-      "loss": 1.318,
+      "learning_rate": 1.987323943661972e-05,
+      "loss": 1.0948,
       "step": 4380
     },
     {
       "epoch": 60.97,
-      "learning_rate": 1.9809859154929576e-05,
-      "loss": 1.2885,
+      "learning_rate": 1.980281690140845e-05,
+      "loss": 1.1057,
       "step": 4390
     },
     {
       "epoch": 61.0,
-      "eval_loss": 3.1023097038269043,
-      "eval_runtime": 130.9774,
-      "eval_samples_per_second": 4.062,
-      "eval_steps_per_second": 0.512,
-      "eval_wer": 0.9723597094555505,
+      "eval_loss": 2.687922239303589,
+      "eval_runtime": 127.6816,
+      "eval_samples_per_second": 4.167,
+      "eval_steps_per_second": 0.525,
+      "eval_wer": 0.9602751173105355,
       "step": 4392
     },
     {
       "epoch": 61.11,
-      "learning_rate": 1.973943661971831e-05,
-      "loss": 1.3469,
+      "learning_rate": 1.9732394366197186e-05,
+      "loss": 1.1104,
       "step": 4400
     },
     {
       "epoch": 61.25,
-      "learning_rate": 1.9669014084507043e-05,
-      "loss": 1.2218,
+      "learning_rate": 1.9661971830985918e-05,
+      "loss": 1.0503,
       "step": 4410
     },
     {
       "epoch": 61.39,
-      "learning_rate": 1.9598591549295775e-05,
-      "loss": 1.2931,
+      "learning_rate": 1.959154929577465e-05,
+      "loss": 1.0896,
       "step": 4420
     },
     {
       "epoch": 61.53,
-      "learning_rate": 1.9528169014084507e-05,
-      "loss": 1.2769,
+      "learning_rate": 1.9521126760563382e-05,
+      "loss": 1.0813,
       "step": 4430
     },
     {
       "epoch": 61.66,
-      "learning_rate": 1.945774647887324e-05,
-      "loss": 1.2587,
+      "learning_rate": 1.9450704225352114e-05,
+      "loss": 1.0609,
       "step": 4440
     },
     {
       "epoch": 61.8,
-      "learning_rate": 1.9387323943661974e-05,
-      "loss": 1.3005,
+      "learning_rate": 1.938028169014085e-05,
+      "loss": 1.0867,
       "step": 4450
     },
     {
       "epoch": 61.94,
-      "learning_rate": 1.9316901408450706e-05,
-      "loss": 1.2538,
+      "learning_rate": 1.9309859154929577e-05,
+      "loss": 1.0722,
       "step": 4460
     },
     {
       "epoch": 62.0,
-      "eval_loss": 2.7857325077056885,
-      "eval_runtime": 131.3498,
-      "eval_samples_per_second": 4.05,
-      "eval_steps_per_second": 0.51,
-      "eval_wer": 0.9877225686186283,
+      "eval_loss": 2.5426149368286133,
+      "eval_runtime": 127.3462,
+      "eval_samples_per_second": 4.178,
+      "eval_steps_per_second": 0.526,
+      "eval_wer": 0.9735167448736903,
       "step": 4464
     },
     {
       "epoch": 62.08,
-      "learning_rate": 1.9246478873239438e-05,
-      "loss": 1.3198,
+      "learning_rate": 1.923943661971831e-05,
+      "loss": 1.1005,
       "step": 4470
     },
     {
       "epoch": 62.22,
-      "learning_rate": 1.917605633802817e-05,
-      "loss": 1.2637,
+      "learning_rate": 1.916901408450704e-05,
+      "loss": 1.0734,
       "step": 4480
     },
     {
       "epoch": 62.36,
-      "learning_rate": 1.91056338028169e-05,
-      "loss": 1.3048,
+      "learning_rate": 1.9098591549295773e-05,
+      "loss": 1.0887,
       "step": 4490
     },
     {
       "epoch": 62.5,
-      "learning_rate": 1.9035211267605633e-05,
-      "loss": 1.2401,
+      "learning_rate": 1.9028169014084508e-05,
+      "loss": 1.0536,
       "step": 4500
     },
     {
       "epoch": 62.64,
-      "learning_rate": 1.896478873239437e-05,
-      "loss": 1.2781,
+      "learning_rate": 1.895774647887324e-05,
+      "loss": 1.0741,
       "step": 4510
     },
     {
       "epoch": 62.78,
-      "learning_rate": 1.88943661971831e-05,
-      "loss": 1.2786,
+      "learning_rate": 1.8887323943661972e-05,
+      "loss": 1.0537,
       "step": 4520
     },
     {
       "epoch": 62.91,
-      "learning_rate": 1.8823943661971832e-05,
-      "loss": 1.2536,
+      "learning_rate": 1.8816901408450704e-05,
+      "loss": 1.0731,
       "step": 4530
     },
     {
       "epoch": 63.0,
-      "eval_loss": 3.014954090118408,
-      "eval_runtime": 130.7805,
-      "eval_samples_per_second": 4.068,
-      "eval_steps_per_second": 0.512,
-      "eval_wer": 0.9703670373465321,
+      "eval_loss": 2.8259387016296387,
+      "eval_runtime": 127.1101,
+      "eval_samples_per_second": 4.185,
+      "eval_steps_per_second": 0.527,
+      "eval_wer": 0.9535257440380536,
       "step": 4536
     },
     {
       "epoch": 63.06,
-      "learning_rate": 1.8753521126760564e-05,
-      "loss": 1.321,
+      "learning_rate": 1.8746478873239436e-05,
+      "loss": 1.095,
       "step": 4540
     },
     {
       "epoch": 63.19,
-      "learning_rate": 1.8683098591549296e-05,
-      "loss": 1.2476,
+      "learning_rate": 1.867605633802817e-05,
+      "loss": 1.0617,
       "step": 4550
     },
     {
       "epoch": 63.33,
-      "learning_rate": 1.861267605633803e-05,
-      "loss": 1.2828,
+      "learning_rate": 1.8605633802816903e-05,
+      "loss": 1.0617,
       "step": 4560
     },
     {
       "epoch": 63.47,
-      "learning_rate": 1.8542253521126763e-05,
-      "loss": 1.2221,
+      "learning_rate": 1.8535211267605635e-05,
+      "loss": 1.0575,
       "step": 4570
     },
     {
       "epoch": 63.61,
-      "learning_rate": 1.8471830985915495e-05,
-      "loss": 1.3011,
+      "learning_rate": 1.8464788732394367e-05,
+      "loss": 1.069,
       "step": 4580
     },
     {
       "epoch": 63.75,
-      "learning_rate": 1.8401408450704227e-05,
-      "loss": 1.2338,
+      "learning_rate": 1.83943661971831e-05,
+      "loss": 1.0292,
       "step": 4590
     },
     {
       "epoch": 63.89,
-      "learning_rate": 1.833098591549296e-05,
-      "loss": 1.3339,
+      "learning_rate": 1.832394366197183e-05,
+      "loss": 1.0862,
       "step": 4600
     },
     {
       "epoch": 64.0,
-      "eval_loss": 2.930506706237793,
-      "eval_runtime": 130.9798,
-      "eval_samples_per_second": 4.062,
-      "eval_steps_per_second": 0.512,
-      "eval_wer": 0.9722311499646461,
+      "eval_loss": 2.7631983757019043,
+      "eval_runtime": 127.2256,
+      "eval_samples_per_second": 4.182,
+      "eval_steps_per_second": 0.527,
+      "eval_wer": 0.9559040946197853,
       "step": 4608
     },
     {
       "epoch": 64.03,
-      "learning_rate": 1.826056338028169e-05,
-      "loss": 1.2729,
+      "learning_rate": 1.8253521126760566e-05,
+      "loss": 1.0821,
       "step": 4610
     },
     {
       "epoch": 64.17,
-      "learning_rate": 1.8190140845070426e-05,
-      "loss": 1.2463,
+      "learning_rate": 1.8183098591549298e-05,
+      "loss": 1.056,
       "step": 4620
     },
     {
       "epoch": 64.3,
-      "learning_rate": 1.8119718309859155e-05,
-      "loss": 1.2679,
+      "learning_rate": 1.811267605633803e-05,
+      "loss": 1.0753,
       "step": 4630
     },
     {
       "epoch": 64.44,
-      "learning_rate": 1.8049295774647886e-05,
-      "loss": 1.231,
+      "learning_rate": 1.804225352112676e-05,
+      "loss": 1.058,
       "step": 4640
     },
     {
       "epoch": 64.58,
-      "learning_rate": 1.797887323943662e-05,
-      "loss": 1.2748,
+      "learning_rate": 1.7971830985915493e-05,
+      "loss": 1.0466,
       "step": 4650
     },
     {
       "epoch": 64.72,
-      "learning_rate": 1.790845070422535e-05,
-      "loss": 1.2311,
+      "learning_rate": 1.790140845070423e-05,
+      "loss": 1.0545,
       "step": 4660
     },
     {
       "epoch": 64.86,
-      "learning_rate": 1.7838028169014085e-05,
-      "loss": 1.2795,
+      "learning_rate": 1.783098591549296e-05,
+      "loss": 1.0664,
       "step": 4670
     },
     {
       "epoch": 65.0,
-      "learning_rate": 1.7767605633802817e-05,
-      "loss": 1.2392,
+      "learning_rate": 1.7760563380281692e-05,
+      "loss": 1.0396,
       "step": 4680
     },
     {
       "epoch": 65.0,
-      "eval_loss": 2.769491195678711,
-      "eval_runtime": 130.6674,
-      "eval_samples_per_second": 4.071,
-      "eval_steps_per_second": 0.513,
-      "eval_wer": 0.9749308992736389,
+      "eval_loss": 2.5400614738464355,
+      "eval_runtime": 128.5937,
+      "eval_samples_per_second": 4.137,
+      "eval_steps_per_second": 0.521,
+      "eval_wer": 0.9806517966188854,
       "step": 4680
     },
     {
       "epoch": 65.14,
-      "learning_rate": 1.769718309859155e-05,
-      "loss": 1.2914,
+      "learning_rate": 1.7690140845070424e-05,
+      "loss": 1.1045,
       "step": 4690
     },
     {
       "epoch": 65.28,
-      "learning_rate": 1.762676056338028e-05,
-      "loss": 1.2715,
+      "learning_rate": 1.7619718309859156e-05,
+      "loss": 1.0362,
       "step": 4700
     },
     {
       "epoch": 65.42,
-      "learning_rate": 1.7556338028169013e-05,
-      "loss": 1.2494,
+      "learning_rate": 1.7549295774647888e-05,
+      "loss": 1.0587,
       "step": 4710
     },
     {
       "epoch": 65.55,
-      "learning_rate": 1.7485915492957748e-05,
-      "loss": 1.2789,
+      "learning_rate": 1.7478873239436623e-05,
+      "loss": 1.0627,
       "step": 4720
     },
     {
       "epoch": 65.69,
-      "learning_rate": 1.741549295774648e-05,
-      "loss": 1.2381,
+      "learning_rate": 1.7408450704225355e-05,
+      "loss": 1.0806,
       "step": 4730
     },
     {
       "epoch": 65.83,
-      "learning_rate": 1.7345070422535212e-05,
-      "loss": 1.3185,
+      "learning_rate": 1.7338028169014087e-05,
+      "loss": 1.0797,
       "step": 4740
     },
     {
       "epoch": 65.97,
-      "learning_rate": 1.7274647887323944e-05,
-      "loss": 1.2572,
+      "learning_rate": 1.7267605633802815e-05,
+      "loss": 1.0581,
       "step": 4750
     },
     {
       "epoch": 66.0,
-      "eval_loss": 2.867743730545044,
-      "eval_runtime": 130.6545,
-      "eval_samples_per_second": 4.072,
-      "eval_steps_per_second": 0.513,
-      "eval_wer": 0.9688886032011313,
+      "eval_loss": 2.697732448577881,
+      "eval_runtime": 127.8292,
+      "eval_samples_per_second": 4.162,
+      "eval_steps_per_second": 0.524,
+      "eval_wer": 0.9686957639647747,
       "step": 4752
     },
     {
       "epoch": 66.11,
-      "learning_rate": 1.7204225352112676e-05,
-      "loss": 1.3039,
+      "learning_rate": 1.7197183098591547e-05,
+      "loss": 1.0747,
       "step": 4760
     },
     {
       "epoch": 66.25,
-      "learning_rate": 1.7133802816901408e-05,
-      "loss": 1.2112,
+      "learning_rate": 1.7126760563380282e-05,
+      "loss": 1.0478,
       "step": 4770
     },
     {
       "epoch": 66.39,
-      "learning_rate": 1.7063380281690143e-05,
-      "loss": 1.28,
+      "learning_rate": 1.7056338028169014e-05,
+      "loss": 1.0768,
       "step": 4780
     },
     {
       "epoch": 66.53,
-      "learning_rate": 1.6992957746478875e-05,
-      "loss": 1.2529,
+      "learning_rate": 1.6985915492957746e-05,
+      "loss": 1.0504,
       "step": 4790
     },
     {
       "epoch": 66.66,
-      "learning_rate": 1.6922535211267607e-05,
-      "loss": 1.2379,
+      "learning_rate": 1.6915492957746478e-05,
+      "loss": 1.0597,
       "step": 4800
     },
     {
       "epoch": 66.8,
-      "learning_rate": 1.685211267605634e-05,
-      "loss": 1.2651,
+      "learning_rate": 1.684507042253521e-05,
+      "loss": 1.0627,
       "step": 4810
     },
     {
       "epoch": 66.94,
-      "learning_rate": 1.678169014084507e-05,
-      "loss": 1.2538,
+      "learning_rate": 1.6774647887323945e-05,
+      "loss": 1.0647,
       "step": 4820
     },
     {
       "epoch": 67.0,
-      "eval_loss": 2.8597300052642822,
-      "eval_runtime": 131.086,
-      "eval_samples_per_second": 4.058,
-      "eval_steps_per_second": 0.511,
-      "eval_wer": 0.980523237127981,
+      "eval_loss": 2.6968305110931396,
+      "eval_runtime": 127.9196,
+      "eval_samples_per_second": 4.159,
+      "eval_steps_per_second": 0.524,
+      "eval_wer": 0.969402841164749,
       "step": 4824
     },
     {
       "epoch": 67.08,
-      "learning_rate": 1.6711267605633806e-05,
-      "loss": 1.295,
+      "learning_rate": 1.6704225352112677e-05,
+      "loss": 1.0821,
       "step": 4830
     },
     {
       "epoch": 67.22,
-      "learning_rate": 1.6640845070422538e-05,
-      "loss": 1.2451,
+      "learning_rate": 1.663380281690141e-05,
+      "loss": 1.0706,
       "step": 4840
     },
     {
       "epoch": 67.36,
-      "learning_rate": 1.657042253521127e-05,
-      "loss": 1.2734,
+      "learning_rate": 1.656338028169014e-05,
+      "loss": 1.0462,
       "step": 4850
     },
     {
       "epoch": 67.5,
-      "learning_rate": 1.65e-05,
-      "loss": 1.2301,
+      "learning_rate": 1.6492957746478873e-05,
+      "loss": 1.0485,
       "step": 4860
     },
     {
       "epoch": 67.64,
-      "learning_rate": 1.6429577464788733e-05,
-      "loss": 1.2834,
+      "learning_rate": 1.6422535211267605e-05,
+      "loss": 1.0828,
       "step": 4870
     },
     {
       "epoch": 67.78,
-      "learning_rate": 1.6359154929577465e-05,
-      "loss": 1.255,
+      "learning_rate": 1.635211267605634e-05,
+      "loss": 1.0558,
       "step": 4880
     },
     {
       "epoch": 67.91,
-      "learning_rate": 1.62887323943662e-05,
-      "loss": 1.2355,
+      "learning_rate": 1.6281690140845072e-05,
+      "loss": 1.0549,
       "step": 4890
     },
     {
       "epoch": 68.0,
-      "eval_loss": 2.8299758434295654,
-      "eval_runtime": 131.2445,
-      "eval_samples_per_second": 4.054,
-      "eval_steps_per_second": 0.51,
-      "eval_wer": 0.9787876840007713,
+      "eval_loss": 2.643899917602539,
+      "eval_runtime": 127.2606,
+      "eval_samples_per_second": 4.18,
+      "eval_steps_per_second": 0.526,
+      "eval_wer": 0.9807160763643376,
       "step": 4896
     },
     {
       "epoch": 68.06,
-      "learning_rate": 1.6218309859154932e-05,
-      "loss": 1.2631,
+      "learning_rate": 1.6211267605633804e-05,
+      "loss": 1.061,
       "step": 4900
     },
     {
       "epoch": 68.19,
-      "learning_rate": 1.6147887323943664e-05,
-      "loss": 1.2278,
+      "learning_rate": 1.6140845070422536e-05,
+      "loss": 1.0335,
       "step": 4910
     },
     {
       "epoch": 68.33,
-      "learning_rate": 1.6077464788732396e-05,
-      "loss": 1.2391,
+      "learning_rate": 1.6070422535211267e-05,
+      "loss": 1.0491,
       "step": 4920
     },
     {
       "epoch": 68.47,
-      "learning_rate": 1.6007042253521128e-05,
-      "loss": 1.2209,
+      "learning_rate": 1.6000000000000003e-05,
+      "loss": 1.0531,
       "step": 4930
     },
     {
       "epoch": 68.61,
-      "learning_rate": 1.593661971830986e-05,
-      "loss": 1.2971,
+      "learning_rate": 1.5929577464788735e-05,
+      "loss": 1.0459,
       "step": 4940
     },
     {
       "epoch": 68.75,
-      "learning_rate": 1.586619718309859e-05,
-      "loss": 1.2235,
+      "learning_rate": 1.5859154929577466e-05,
+      "loss": 1.0331,
       "step": 4950
     },
     {
       "epoch": 68.89,
-      "learning_rate": 1.5795774647887323e-05,
-      "loss": 1.2682,
+      "learning_rate": 1.5788732394366198e-05,
+      "loss": 1.0607,
       "step": 4960
     },
     {
       "epoch": 69.0,
-      "eval_loss": 2.701061248779297,
-      "eval_runtime": 131.5603,
-      "eval_samples_per_second": 4.044,
-      "eval_steps_per_second": 0.509,
-      "eval_wer": 0.9741595423282123,
+      "eval_loss": 2.682176351547241,
+      "eval_runtime": 127.276,
+      "eval_samples_per_second": 4.18,
+      "eval_steps_per_second": 0.526,
+      "eval_wer": 0.977116410619014,
       "step": 4968
     },
     {
       "epoch": 69.03,
-      "learning_rate": 1.5725352112676055e-05,
-      "loss": 1.2781,
+      "learning_rate": 1.571830985915493e-05,
+      "loss": 1.0714,
       "step": 4970
     },
     {
       "epoch": 69.17,
-      "learning_rate": 1.5654929577464787e-05,
-      "loss": 1.2314,
+      "learning_rate": 1.5647887323943662e-05,
+      "loss": 1.04,
       "step": 4980
     },
     {
       "epoch": 69.3,
-      "learning_rate": 1.5584507042253522e-05,
-      "loss": 1.2574,
+      "learning_rate": 1.5577464788732397e-05,
+      "loss": 1.057,
       "step": 4990
     },
     {
       "epoch": 69.44,
-      "learning_rate": 1.5514084507042254e-05,
-      "loss": 1.247,
+      "learning_rate": 1.550704225352113e-05,
+      "loss": 1.0503,
       "step": 5000
     },
     {
       "epoch": 69.58,
-      "learning_rate": 1.5443661971830986e-05,
-      "loss": 1.2494,
+      "learning_rate": 1.543661971830986e-05,
+      "loss": 1.063,
       "step": 5010
     },
     {
       "epoch": 69.72,
-      "learning_rate": 1.5373239436619718e-05,
-      "loss": 1.1997,
+      "learning_rate": 1.5366197183098593e-05,
+      "loss": 1.0353,
       "step": 5020
     },
     {
       "epoch": 69.86,
-      "learning_rate": 1.530281690140845e-05,
-      "loss": 1.2613,
+      "learning_rate": 1.5295774647887325e-05,
+      "loss": 1.0459,
       "step": 5030
     },
     {
       "epoch": 70.0,
-      "learning_rate": 1.5232394366197183e-05,
-      "loss": 1.2252,
+      "learning_rate": 1.5225352112676058e-05,
+      "loss": 1.05,
       "step": 5040
     },
     {
       "epoch": 70.0,
-      "eval_loss": 2.8321807384490967,
-      "eval_runtime": 131.0272,
-      "eval_samples_per_second": 4.06,
-      "eval_steps_per_second": 0.511,
-      "eval_wer": 0.9722311499646461,
+      "eval_loss": 2.7011446952819824,
+      "eval_runtime": 127.6274,
+      "eval_samples_per_second": 4.168,
+      "eval_steps_per_second": 0.525,
+      "eval_wer": 0.9607250755287009,
       "step": 5040
     },
     {
       "epoch": 70.14,
-      "learning_rate": 1.5161971830985915e-05,
-      "loss": 1.2647,
+      "learning_rate": 1.5154929577464789e-05,
+      "loss": 1.0781,
       "step": 5050
     },
     {
       "epoch": 70.28,
-      "learning_rate": 1.5091549295774649e-05,
-      "loss": 1.2388,
+      "learning_rate": 1.508450704225352e-05,
+      "loss": 1.0427,
       "step": 5060
     },
     {
       "epoch": 70.42,
-      "learning_rate": 1.502112676056338e-05,
-      "loss": 1.2413,
+      "learning_rate": 1.5014084507042252e-05,
+      "loss": 1.051,
       "step": 5070
     },
     {
       "epoch": 70.55,
-      "learning_rate": 1.4950704225352113e-05,
-      "loss": 1.2658,
+      "learning_rate": 1.4943661971830986e-05,
+      "loss": 1.0409,
       "step": 5080
     },
     {
       "epoch": 70.69,
-      "learning_rate": 1.4880281690140846e-05,
-      "loss": 1.1942,
+      "learning_rate": 1.4873239436619718e-05,
+      "loss": 1.0442,
       "step": 5090
     },
     {
       "epoch": 70.83,
-      "learning_rate": 1.4809859154929578e-05,
-      "loss": 1.2409,
+      "learning_rate": 1.4802816901408451e-05,
+      "loss": 1.0474,
       "step": 5100
     },
     {
       "epoch": 70.97,
-      "learning_rate": 1.4739436619718312e-05,
-      "loss": 1.2085,
+      "learning_rate": 1.4732394366197183e-05,
+      "loss": 1.042,
       "step": 5110
     },
     {
       "epoch": 71.0,
-      "eval_loss": 2.740072727203369,
-      "eval_runtime": 130.8975,
-      "eval_samples_per_second": 4.064,
-      "eval_steps_per_second": 0.512,
-      "eval_wer": 0.9710741145465064,
+      "eval_loss": 2.576570749282837,
+      "eval_runtime": 128.5974,
+      "eval_samples_per_second": 4.137,
+      "eval_steps_per_second": 0.521,
+      "eval_wer": 0.9713312335283152,
       "step": 5112
     },
     {
       "epoch": 71.11,
-      "learning_rate": 1.4669014084507044e-05,
-      "loss": 1.2891,
+      "learning_rate": 1.4661971830985915e-05,
+      "loss": 1.0515,
       "step": 5120
     },
     {
       "epoch": 71.25,
-      "learning_rate": 1.4598591549295775e-05,
-      "loss": 1.2077,
+      "learning_rate": 1.4591549295774649e-05,
+      "loss": 1.0315,
       "step": 5130
     },
     {
       "epoch": 71.39,
-      "learning_rate": 1.4528169014084509e-05,
-      "loss": 1.2418,
+      "learning_rate": 1.452112676056338e-05,
+      "loss": 1.0655,
       "step": 5140
     },
     {
       "epoch": 71.53,
-      "learning_rate": 1.4457746478873241e-05,
-      "loss": 1.2247,
+      "learning_rate": 1.4450704225352112e-05,
+      "loss": 1.0249,
       "step": 5150
     },
     {
       "epoch": 71.66,
-      "learning_rate": 1.4387323943661973e-05,
-      "loss": 1.2343,
+      "learning_rate": 1.4380281690140846e-05,
+      "loss": 1.0413,
       "step": 5160
     },
     {
       "epoch": 71.8,
-      "learning_rate": 1.4316901408450706e-05,
-      "loss": 1.2643,
+      "learning_rate": 1.4309859154929578e-05,
+      "loss": 1.055,
       "step": 5170
     },
     {
       "epoch": 71.94,
-      "learning_rate": 1.4246478873239438e-05,
-      "loss": 1.2412,
+      "learning_rate": 1.423943661971831e-05,
+      "loss": 1.042,
       "step": 5180
     },
     {
       "epoch": 72.0,
-      "eval_loss": 2.6227338314056396,
-      "eval_runtime": 130.7815,
-      "eval_samples_per_second": 4.068,
-      "eval_steps_per_second": 0.512,
-      "eval_wer": 0.9762807739281353,
+      "eval_loss": 2.572017192840576,
+      "eval_runtime": 127.0339,
+      "eval_samples_per_second": 4.188,
+      "eval_steps_per_second": 0.527,
+      "eval_wer": 0.9747380600372823,
       "step": 5184
     },
     {
       "epoch": 72.08,
-      "learning_rate": 1.417605633802817e-05,
-      "loss": 1.2857,
+      "learning_rate": 1.4169014084507043e-05,
+      "loss": 1.0803,
       "step": 5190
     },
     {
       "epoch": 72.22,
-      "learning_rate": 1.4105633802816904e-05,
-      "loss": 1.2068,
+      "learning_rate": 1.4098591549295775e-05,
+      "loss": 1.0462,
       "step": 5200
     },
     {
       "epoch": 72.36,
-      "learning_rate": 1.4035211267605636e-05,
-      "loss": 1.2536,
+      "learning_rate": 1.4028169014084509e-05,
+      "loss": 1.0362,
       "step": 5210
     },
     {
       "epoch": 72.5,
-      "learning_rate": 1.3964788732394367e-05,
-      "loss": 1.2114,
+      "learning_rate": 1.395774647887324e-05,
+      "loss": 1.0291,
       "step": 5220
     },
     {
       "epoch": 72.64,
-      "learning_rate": 1.3894366197183098e-05,
-      "loss": 1.2595,
+      "learning_rate": 1.3887323943661972e-05,
+      "loss": 1.0326,
       "step": 5230
     },
     {
       "epoch": 72.78,
-      "learning_rate": 1.3823943661971831e-05,
-      "loss": 1.209,
+      "learning_rate": 1.3816901408450706e-05,
+      "loss": 1.0279,
       "step": 5240
     },
     {
       "epoch": 72.91,
-      "learning_rate": 1.3753521126760563e-05,
-      "loss": 1.2148,
+      "learning_rate": 1.3746478873239438e-05,
+      "loss": 1.0594,
       "step": 5250
     },
     {
       "epoch": 73.0,
-      "eval_loss": 2.714576244354248,
-      "eval_runtime": 131.252,
-      "eval_samples_per_second": 4.053,
-      "eval_steps_per_second": 0.51,
-      "eval_wer": 0.9689528829465835,
+      "eval_loss": 2.717604160308838,
+      "eval_runtime": 126.8003,
+      "eval_samples_per_second": 4.196,
+      "eval_steps_per_second": 0.528,
+      "eval_wer": 0.9704313170919843,
       "step": 5256
     },
     {
       "epoch": 73.06,
-      "learning_rate": 1.3683098591549295e-05,
-      "loss": 1.2858,
+      "learning_rate": 1.367605633802817e-05,
+      "loss": 1.051,
       "step": 5260
     },
     {
       "epoch": 73.19,
-      "learning_rate": 1.3612676056338028e-05,
-      "loss": 1.2126,
+      "learning_rate": 1.3605633802816903e-05,
+      "loss": 1.0259,
       "step": 5270
     },
     {
       "epoch": 73.33,
-      "learning_rate": 1.354225352112676e-05,
-      "loss": 1.257,
+      "learning_rate": 1.3535211267605635e-05,
+      "loss": 1.0576,
       "step": 5280
     },
     {
       "epoch": 73.47,
-      "learning_rate": 1.3471830985915492e-05,
-      "loss": 1.2102,
+      "learning_rate": 1.3464788732394367e-05,
+      "loss": 1.0293,
       "step": 5290
     },
     {
       "epoch": 73.61,
-      "learning_rate": 1.3401408450704226e-05,
-      "loss": 1.2511,
+      "learning_rate": 1.33943661971831e-05,
+      "loss": 1.0324,
       "step": 5300
     },
     {
       "epoch": 73.75,
-      "learning_rate": 1.3330985915492958e-05,
-      "loss": 1.2035,
+      "learning_rate": 1.3323943661971833e-05,
+      "loss": 1.0203,
       "step": 5310
     },
     {
       "epoch": 73.89,
-      "learning_rate": 1.326056338028169e-05,
-      "loss": 1.2411,
+      "learning_rate": 1.3253521126760564e-05,
+      "loss": 1.0425,
       "step": 5320
     },
     {
       "epoch": 74.0,
-      "eval_loss": 2.996312141418457,
-      "eval_runtime": 131.4356,
-      "eval_samples_per_second": 4.048,
-      "eval_steps_per_second": 0.51,
-      "eval_wer": 0.966317413383043,
+      "eval_loss": 2.7458221912384033,
+      "eval_runtime": 127.673,
+      "eval_samples_per_second": 4.167,
+      "eval_steps_per_second": 0.525,
+      "eval_wer": 0.9614321527286752,
       "step": 5328
     },
     {
       "epoch": 74.03,
-      "learning_rate": 1.3190140845070423e-05,
-      "loss": 1.2552,
+      "learning_rate": 1.3183098591549298e-05,
+      "loss": 1.0546,
       "step": 5330
     },
     {
       "epoch": 74.17,
-      "learning_rate": 1.3119718309859155e-05,
-      "loss": 1.2058,
+      "learning_rate": 1.3112676056338028e-05,
+      "loss": 1.0197,
       "step": 5340
     },
     {
       "epoch": 74.3,
-      "learning_rate": 1.3049295774647889e-05,
-      "loss": 1.2041,
+      "learning_rate": 1.304225352112676e-05,
+      "loss": 1.0243,
       "step": 5350
     },
     {
       "epoch": 74.44,
-      "learning_rate": 1.297887323943662e-05,
-      "loss": 1.2125,
+      "learning_rate": 1.2971830985915492e-05,
+      "loss": 1.0373,
       "step": 5360
     },
     {
       "epoch": 74.58,
-      "learning_rate": 1.2908450704225352e-05,
-      "loss": 1.2323,
+      "learning_rate": 1.2901408450704226e-05,
+      "loss": 1.0336,
       "step": 5370
     },
     {
       "epoch": 74.72,
-      "learning_rate": 1.2838028169014086e-05,
-      "loss": 1.2222,
+      "learning_rate": 1.2830985915492957e-05,
+      "loss": 1.0304,
       "step": 5380
     },
     {
       "epoch": 74.86,
-      "learning_rate": 1.2767605633802818e-05,
-      "loss": 1.2527,
+      "learning_rate": 1.276056338028169e-05,
+      "loss": 1.0448,
       "step": 5390
     },
     {
       "epoch": 75.0,
-      "learning_rate": 1.269718309859155e-05,
-      "loss": 1.2012,
+      "learning_rate": 1.2690140845070423e-05,
+      "loss": 1.0199,
       "step": 5400
     },
     {
       "epoch": 75.0,
-      "eval_loss": 2.7940216064453125,
-      "eval_runtime": 130.9397,
-      "eval_samples_per_second": 4.063,
-      "eval_steps_per_second": 0.512,
-      "eval_wer": 0.9655460564376165,
+      "eval_loss": 2.5905630588531494,
+      "eval_runtime": 127.2257,
+      "eval_samples_per_second": 4.182,
+      "eval_steps_per_second": 0.527,
+      "eval_wer": 0.9987144050909559,
       "step": 5400
     },
     {
       "epoch": 75.14,
-      "learning_rate": 1.2626760563380283e-05,
-      "loss": 1.2865,
+      "learning_rate": 1.2619718309859155e-05,
+      "loss": 1.0716,
       "step": 5410
     },
     {
       "epoch": 75.28,
-      "learning_rate": 1.2556338028169015e-05,
-      "loss": 1.2463,
+      "learning_rate": 1.2549295774647887e-05,
+      "loss": 1.0184,
       "step": 5420
     },
     {
       "epoch": 75.42,
-      "learning_rate": 1.2485915492957747e-05,
-      "loss": 1.2363,
+      "learning_rate": 1.247887323943662e-05,
+      "loss": 1.0305,
       "step": 5430
     },
     {
       "epoch": 75.55,
-      "learning_rate": 1.241549295774648e-05,
-      "loss": 1.2467,
+      "learning_rate": 1.2408450704225352e-05,
+      "loss": 1.0315,
       "step": 5440
     },
     {
       "epoch": 75.69,
-      "learning_rate": 1.234507042253521e-05,
-      "loss": 1.2138,
+      "learning_rate": 1.2338028169014084e-05,
+      "loss": 1.0329,
       "step": 5450
     },
     {
       "epoch": 75.83,
-      "learning_rate": 1.2274647887323944e-05,
-      "loss": 1.2144,
+      "learning_rate": 1.2267605633802817e-05,
+      "loss": 1.0248,
       "step": 5460
     },
     {
       "epoch": 75.97,
-      "learning_rate": 1.2204225352112676e-05,
-      "loss": 1.1969,
+      "learning_rate": 1.219718309859155e-05,
+      "loss": 1.0198,
       "step": 5470
     },
     {
       "epoch": 76.0,
-      "eval_loss": 2.621224880218506,
-      "eval_runtime": 131.068,
-      "eval_samples_per_second": 4.059,
-      "eval_steps_per_second": 0.511,
-      "eval_wer": 0.9790448029825802,
+      "eval_loss": 2.553414821624756,
+      "eval_runtime": 128.3399,
+      "eval_samples_per_second": 4.145,
+      "eval_steps_per_second": 0.522,
+      "eval_wer": 1.0087420453815004,
       "step": 5472
     },
     {
       "epoch": 76.11,
-      "learning_rate": 1.2133802816901408e-05,
-      "loss": 1.276,
+      "learning_rate": 1.2126760563380283e-05,
+      "loss": 1.0646,
       "step": 5480
     },
     {
       "epoch": 76.25,
-      "learning_rate": 1.2063380281690142e-05,
-      "loss": 1.2038,
+      "learning_rate": 1.2056338028169015e-05,
+      "loss": 1.0279,
       "step": 5490
     },
     {
       "epoch": 76.39,
-      "learning_rate": 1.1992957746478873e-05,
-      "loss": 1.2469,
+      "learning_rate": 1.1985915492957747e-05,
+      "loss": 1.0421,
       "step": 5500
     },
     {
       "epoch": 76.53,
-      "learning_rate": 1.1922535211267605e-05,
-      "loss": 1.2184,
+      "learning_rate": 1.191549295774648e-05,
+      "loss": 1.0234,
       "step": 5510
     },
     {
       "epoch": 76.66,
-      "learning_rate": 1.1852112676056339e-05,
-      "loss": 1.216,
+      "learning_rate": 1.1845070422535212e-05,
+      "loss": 1.0396,
       "step": 5520
     },
     {
       "epoch": 76.8,
-      "learning_rate": 1.178169014084507e-05,
-      "loss": 1.2265,
+      "learning_rate": 1.1774647887323944e-05,
+      "loss": 1.0249,
       "step": 5530
     },
     {
       "epoch": 76.94,
-      "learning_rate": 1.1711267605633804e-05,
-      "loss": 1.2004,
+      "learning_rate": 1.1704225352112678e-05,
+      "loss": 1.0193,
       "step": 5540
     },
     {
       "epoch": 77.0,
-      "eval_loss": 2.8644309043884277,
-      "eval_runtime": 131.1358,
-      "eval_samples_per_second": 4.057,
-      "eval_steps_per_second": 0.511,
-      "eval_wer": 0.9710741145465064,
+      "eval_loss": 2.5420749187469482,
+      "eval_runtime": 127.1146,
+      "eval_samples_per_second": 4.185,
+      "eval_steps_per_second": 0.527,
+      "eval_wer": 0.9933149064729704,
       "step": 5544
     },
     {
       "epoch": 77.08,
-      "learning_rate": 1.1640845070422536e-05,
-      "loss": 1.2794,
+      "learning_rate": 1.163380281690141e-05,
+      "loss": 1.0387,
       "step": 5550
     },
     {
       "epoch": 77.22,
-      "learning_rate": 1.1570422535211268e-05,
-      "loss": 1.1772,
+      "learning_rate": 1.1563380281690141e-05,
+      "loss": 1.015,
       "step": 5560
     },
     {
       "epoch": 77.36,
-      "learning_rate": 1.1500000000000002e-05,
-      "loss": 1.2421,
+      "learning_rate": 1.1492957746478873e-05,
+      "loss": 1.0254,
       "step": 5570
     },
     {
       "epoch": 77.5,
-      "learning_rate": 1.1429577464788734e-05,
-      "loss": 1.1951,
+      "learning_rate": 1.1422535211267605e-05,
+      "loss": 1.012,
       "step": 5580
     },
     {
       "epoch": 77.64,
-      "learning_rate": 1.1359154929577465e-05,
-      "loss": 1.2643,
+      "learning_rate": 1.1352112676056339e-05,
+      "loss": 1.0334,
       "step": 5590
     },
     {
       "epoch": 77.78,
-      "learning_rate": 1.1288732394366197e-05,
-      "loss": 1.232,
+      "learning_rate": 1.128169014084507e-05,
+      "loss": 1.0393,
       "step": 5600
     },
     {
       "epoch": 77.91,
-      "learning_rate": 1.121830985915493e-05,
-      "loss": 1.2247,
+      "learning_rate": 1.1211267605633802e-05,
+      "loss": 1.0379,
       "step": 5610
     },
     {
       "epoch": 78.0,
-      "eval_loss": 2.612703561782837,
-      "eval_runtime": 131.1043,
-      "eval_samples_per_second": 4.058,
-      "eval_steps_per_second": 0.511,
-      "eval_wer": 0.9620749501831972,
+      "eval_loss": 2.5139431953430176,
+      "eval_runtime": 127.3765,
+      "eval_samples_per_second": 4.177,
+      "eval_steps_per_second": 0.526,
+      "eval_wer": 0.9993572025454779,
       "step": 5616
     },
     {
       "epoch": 78.06,
-      "learning_rate": 1.1147887323943663e-05,
-      "loss": 1.2683,
+      "learning_rate": 1.1140845070422536e-05,
+      "loss": 1.049,
       "step": 5620
     },
     {
       "epoch": 78.19,
-      "learning_rate": 1.1077464788732395e-05,
-      "loss": 1.198,
+      "learning_rate": 1.1070422535211268e-05,
+      "loss": 1.014,
       "step": 5630
     },
     {
       "epoch": 78.33,
-      "learning_rate": 1.1007042253521127e-05,
-      "loss": 1.2199,
+      "learning_rate": 1.1000000000000001e-05,
+      "loss": 1.0155,
       "step": 5640
     },
     {
       "epoch": 78.47,
-      "learning_rate": 1.093661971830986e-05,
-      "loss": 1.1778,
+      "learning_rate": 1.0929577464788733e-05,
+      "loss": 1.0165,
       "step": 5650
     },
     {
       "epoch": 78.61,
-      "learning_rate": 1.0866197183098592e-05,
-      "loss": 1.2063,
+      "learning_rate": 1.0859154929577465e-05,
+      "loss": 1.0086,
       "step": 5660
     },
     {
       "epoch": 78.75,
-      "learning_rate": 1.0795774647887324e-05,
-      "loss": 1.1895,
+      "learning_rate": 1.0788732394366199e-05,
+      "loss": 1.0145,
       "step": 5670
     },
     {
       "epoch": 78.89,
-      "learning_rate": 1.0725352112676057e-05,
-      "loss": 1.2791,
+      "learning_rate": 1.071830985915493e-05,
+      "loss": 1.025,
       "step": 5680
     },
     {
       "epoch": 79.0,
-      "eval_loss": 2.653787136077881,
-      "eval_runtime": 131.0838,
-      "eval_samples_per_second": 4.058,
-      "eval_steps_per_second": 0.511,
-      "eval_wer": 0.9719097512373851,
+      "eval_loss": 2.4849629402160645,
+      "eval_runtime": 127.0424,
+      "eval_samples_per_second": 4.188,
+      "eval_steps_per_second": 0.527,
+      "eval_wer": 1.0313042360352254,
       "step": 5688
     },
     {
       "epoch": 79.03,
-      "learning_rate": 1.065492957746479e-05,
-      "loss": 1.2484,
+      "learning_rate": 1.0647887323943662e-05,
+      "loss": 1.0454,
       "step": 5690
     },
     {
       "epoch": 79.17,
-      "learning_rate": 1.0584507042253521e-05,
-      "loss": 1.2304,
+      "learning_rate": 1.0577464788732396e-05,
+      "loss": 1.0163,
       "step": 5700
     },
     {
       "epoch": 79.3,
-      "learning_rate": 1.0514084507042255e-05,
-      "loss": 1.2353,
+      "learning_rate": 1.0507042253521126e-05,
+      "loss": 1.0109,
       "step": 5710
     },
     {
       "epoch": 79.44,
-      "learning_rate": 1.0443661971830987e-05,
-      "loss": 1.1799,
+      "learning_rate": 1.043661971830986e-05,
+      "loss": 1.0111,
       "step": 5720
     },
     {
       "epoch": 79.58,
-      "learning_rate": 1.037323943661972e-05,
-      "loss": 1.2419,
+      "learning_rate": 1.0366197183098592e-05,
+      "loss": 1.0166,
       "step": 5730
     },
     {
       "epoch": 79.72,
-      "learning_rate": 1.030281690140845e-05,
-      "loss": 1.2154,
+      "learning_rate": 1.0295774647887324e-05,
+      "loss": 1.0286,
       "step": 5740
     },
     {
       "epoch": 79.86,
-      "learning_rate": 1.0232394366197182e-05,
-      "loss": 1.2384,
+      "learning_rate": 1.0225352112676057e-05,
+      "loss": 1.0119,
       "step": 5750
     },
     {
       "epoch": 80.0,
-      "learning_rate": 1.0161971830985916e-05,
-      "loss": 1.1867,
+      "learning_rate": 1.0154929577464789e-05,
+      "loss": 1.0054,
       "step": 5760
     },
     {
       "epoch": 80.0,
-      "eval_loss": 2.7778642177581787,
-      "eval_runtime": 129.0244,
-      "eval_samples_per_second": 4.123,
-      "eval_steps_per_second": 0.519,
-      "eval_wer": 0.9586681236742303,
+      "eval_loss": 2.580329179763794,
+      "eval_runtime": 127.0215,
+      "eval_samples_per_second": 4.188,
+      "eval_steps_per_second": 0.527,
+      "eval_wer": 0.9814231535643119,
       "step": 5760
     },
     {
       "epoch": 80.14,
-      "learning_rate": 1.0091549295774648e-05,
-      "loss": 1.259,
+      "learning_rate": 1.0084507042253521e-05,
+      "loss": 1.0434,
       "step": 5770
     },
     {
       "epoch": 80.28,
-      "learning_rate": 1.0021126760563381e-05,
-      "loss": 1.2021,
+      "learning_rate": 1.0014084507042254e-05,
+      "loss": 1.0142,
       "step": 5780
     },
     {
       "epoch": 80.42,
-      "learning_rate": 9.950704225352113e-06,
-      "loss": 1.2074,
+      "learning_rate": 9.943661971830986e-06,
+      "loss": 1.0268,
       "step": 5790
     },
     {
       "epoch": 80.55,
-      "learning_rate": 9.880281690140845e-06,
-      "loss": 1.2142,
+      "learning_rate": 9.873239436619718e-06,
+      "loss": 1.0159,
       "step": 5800
     },
     {
       "epoch": 80.69,
-      "learning_rate": 9.809859154929579e-06,
-      "loss": 1.1929,
+      "learning_rate": 9.802816901408452e-06,
+      "loss": 1.0417,
       "step": 5810
     },
     {
       "epoch": 80.83,
-      "learning_rate": 9.73943661971831e-06,
-      "loss": 1.2238,
+      "learning_rate": 9.732394366197184e-06,
+      "loss": 1.0246,
       "step": 5820
     },
     {
       "epoch": 80.97,
-      "learning_rate": 9.669014084507042e-06,
-      "loss": 1.1992,
+      "learning_rate": 9.661971830985917e-06,
+      "loss": 1.0218,
       "step": 5830
     },
     {
       "epoch": 81.0,
-      "eval_loss": 2.829700231552124,
-      "eval_runtime": 128.7334,
-      "eval_samples_per_second": 4.133,
-      "eval_steps_per_second": 0.52,
-      "eval_wer": 0.9643247412740246,
+      "eval_loss": 2.569614887237549,
+      "eval_runtime": 126.834,
+      "eval_samples_per_second": 4.194,
+      "eval_steps_per_second": 0.528,
+      "eval_wer": 0.986694092691393,
       "step": 5832
     },
     {
       "epoch": 81.11,
-      "learning_rate": 9.598591549295776e-06,
-      "loss": 1.2484,
+      "learning_rate": 9.591549295774649e-06,
+      "loss": 1.038,
       "step": 5840
     },
     {
       "epoch": 81.25,
-      "learning_rate": 9.528169014084508e-06,
-      "loss": 1.2018,
+      "learning_rate": 9.521126760563381e-06,
+      "loss": 1.0142,
       "step": 5850
     },
     {
       "epoch": 81.39,
-      "learning_rate": 9.45774647887324e-06,
-      "loss": 1.2252,
+      "learning_rate": 9.450704225352113e-06,
+      "loss": 1.0117,
       "step": 5860
     },
     {
       "epoch": 81.53,
-      "learning_rate": 9.387323943661973e-06,
-      "loss": 1.1984,
+      "learning_rate": 9.380281690140845e-06,
+      "loss": 1.0029,
       "step": 5870
     },
     {
       "epoch": 81.66,
-      "learning_rate": 9.316901408450705e-06,
-      "loss": 1.198,
+      "learning_rate": 9.309859154929578e-06,
+      "loss": 1.0172,
       "step": 5880
     },
     {
       "epoch": 81.8,
-      "learning_rate": 9.246478873239437e-06,
-      "loss": 1.2184,
+      "learning_rate": 9.23943661971831e-06,
+      "loss": 1.0007,
       "step": 5890
     },
     {
       "epoch": 81.94,
-      "learning_rate": 9.176056338028169e-06,
-      "loss": 1.1833,
+      "learning_rate": 9.169014084507042e-06,
+      "loss": 1.0177,
       "step": 5900
     },
     {
       "epoch": 82.0,
-      "eval_loss": 2.850757122039795,
-      "eval_runtime": 128.577,
-      "eval_samples_per_second": 4.138,
-      "eval_steps_per_second": 0.521,
-      "eval_wer": 0.9722954297100983,
+      "eval_loss": 2.601060390472412,
+      "eval_runtime": 126.8423,
+      "eval_samples_per_second": 4.194,
+      "eval_steps_per_second": 0.528,
+      "eval_wer": 1.006492254290673,
       "step": 5904
     },
     {
       "epoch": 82.08,
-      "learning_rate": 9.1056338028169e-06,
-      "loss": 1.2571,
+      "learning_rate": 9.098591549295776e-06,
+      "loss": 1.0443,
       "step": 5910
     },
     {
       "epoch": 82.22,
-      "learning_rate": 9.035211267605634e-06,
-      "loss": 1.1955,
+      "learning_rate": 9.028169014084507e-06,
+      "loss": 1.0218,
       "step": 5920
     },
     {
       "epoch": 82.36,
-      "learning_rate": 8.964788732394366e-06,
-      "loss": 1.2139,
+      "learning_rate": 8.95774647887324e-06,
+      "loss": 0.9952,
       "step": 5930
     },
     {
       "epoch": 82.5,
-      "learning_rate": 8.894366197183098e-06,
-      "loss": 1.1704,
+      "learning_rate": 8.887323943661973e-06,
+      "loss": 1.0065,
       "step": 5940
     },
     {
       "epoch": 82.64,
-      "learning_rate": 8.823943661971832e-06,
-      "loss": 1.2177,
+      "learning_rate": 8.816901408450705e-06,
+      "loss": 1.0205,
       "step": 5950
     },
     {
       "epoch": 82.78,
-      "learning_rate": 8.753521126760563e-06,
-      "loss": 1.2016,
+      "learning_rate": 8.746478873239437e-06,
+      "loss": 1.0082,
       "step": 5960
     },
     {
       "epoch": 82.91,
-      "learning_rate": 8.683098591549297e-06,
-      "loss": 1.2092,
+      "learning_rate": 8.67605633802817e-06,
+      "loss": 1.0094,
       "step": 5970
     },
     {
       "epoch": 83.0,
-      "eval_loss": 2.8260996341705322,
-      "eval_runtime": 128.9666,
-      "eval_samples_per_second": 4.125,
-      "eval_steps_per_second": 0.52,
-      "eval_wer": 0.9690171626920357,
+      "eval_loss": 2.6165919303894043,
+      "eval_runtime": 127.0294,
+      "eval_samples_per_second": 4.188,
+      "eval_steps_per_second": 0.527,
+      "eval_wer": 0.9855370572732532,
       "step": 5976
     },
     {
       "epoch": 83.06,
-      "learning_rate": 8.612676056338029e-06,
-      "loss": 1.2366,
+      "learning_rate": 8.605633802816902e-06,
+      "loss": 1.0319,
       "step": 5980
     },
     {
       "epoch": 83.19,
-      "learning_rate": 8.54225352112676e-06,
-      "loss": 1.1923,
+      "learning_rate": 8.535211267605634e-06,
+      "loss": 1.0136,
       "step": 5990
     },
     {
       "epoch": 83.33,
-      "learning_rate": 8.471830985915494e-06,
-      "loss": 1.205,
+      "learning_rate": 8.464788732394368e-06,
+      "loss": 1.0105,
       "step": 6000
     },
     {
       "epoch": 83.47,
-      "learning_rate": 8.401408450704226e-06,
-      "loss": 1.1998,
+      "learning_rate": 8.394366197183098e-06,
+      "loss": 1.0101,
       "step": 6010
     },
     {
       "epoch": 83.61,
-      "learning_rate": 8.330985915492958e-06,
-      "loss": 1.2067,
+      "learning_rate": 8.323943661971831e-06,
+      "loss": 1.0045,
       "step": 6020
     },
     {
       "epoch": 83.75,
-      "learning_rate": 8.260563380281692e-06,
-      "loss": 1.1848,
+      "learning_rate": 8.253521126760563e-06,
+      "loss": 1.0018,
       "step": 6030
     },
     {
       "epoch": 83.89,
-      "learning_rate": 8.190140845070422e-06,
-      "loss": 1.2215,
+      "learning_rate": 8.183098591549295e-06,
+      "loss": 1.0202,
       "step": 6040
     },
     {
       "epoch": 84.0,
-      "eval_loss": 2.599538564682007,
-      "eval_runtime": 128.5443,
-      "eval_samples_per_second": 4.139,
-      "eval_steps_per_second": 0.521,
-      "eval_wer": 0.9717169120010285,
+      "eval_loss": 2.555656671524048,
+      "eval_runtime": 127.4032,
+      "eval_samples_per_second": 4.176,
+      "eval_steps_per_second": 0.526,
+      "eval_wer": 1.0204409590538022,
       "step": 6048
     },
     {
       "epoch": 84.03,
-      "learning_rate": 8.119718309859155e-06,
-      "loss": 1.229,
+      "learning_rate": 8.112676056338029e-06,
+      "loss": 1.042,
       "step": 6050
     },
     {
       "epoch": 84.17,
-      "learning_rate": 8.049295774647887e-06,
-      "loss": 1.1939,
+      "learning_rate": 8.04225352112676e-06,
+      "loss": 1.0202,
       "step": 6060
     },
     {
       "epoch": 84.3,
-      "learning_rate": 7.97887323943662e-06,
-      "loss": 1.2002,
+      "learning_rate": 7.971830985915494e-06,
+      "loss": 0.9892,
       "step": 6070
     },
     {
       "epoch": 84.44,
-      "learning_rate": 7.908450704225353e-06,
-      "loss": 1.1931,
+      "learning_rate": 7.901408450704226e-06,
+      "loss": 1.0204,
       "step": 6080
     },
     {
       "epoch": 84.58,
-      "learning_rate": 7.838028169014085e-06,
-      "loss": 1.1964,
+      "learning_rate": 7.830985915492958e-06,
+      "loss": 0.9938,
       "step": 6090
     },
     {
       "epoch": 84.72,
-      "learning_rate": 7.767605633802817e-06,
-      "loss": 1.169,
+      "learning_rate": 7.760563380281691e-06,
+      "loss": 1.0036,
       "step": 6100
     },
     {
       "epoch": 84.86,
-      "learning_rate": 7.69718309859155e-06,
-      "loss": 1.208,
+      "learning_rate": 7.690140845070423e-06,
+      "loss": 1.0066,
       "step": 6110
     },
     {
       "epoch": 85.0,
-      "learning_rate": 7.626760563380282e-06,
-      "loss": 1.1966,
+      "learning_rate": 7.619718309859156e-06,
+      "loss": 1.0148,
       "step": 6120
     },
     {
       "epoch": 85.0,
-      "eval_loss": 2.7658963203430176,
-      "eval_runtime": 128.9084,
-      "eval_samples_per_second": 4.127,
-      "eval_steps_per_second": 0.52,
-      "eval_wer": 0.9677315677829916,
+      "eval_loss": 2.611813545227051,
+      "eval_runtime": 127.4222,
+      "eval_samples_per_second": 4.175,
+      "eval_steps_per_second": 0.526,
+      "eval_wer": 1.0032782670180627,
       "step": 6120
     },
     {
       "epoch": 85.14,
-      "learning_rate": 7.556338028169015e-06,
-      "loss": 1.242,
+      "learning_rate": 7.549295774647888e-06,
+      "loss": 1.0309,
       "step": 6130
     },
     {
       "epoch": 85.28,
-      "learning_rate": 7.485915492957747e-06,
-      "loss": 1.2057,
+      "learning_rate": 7.478873239436621e-06,
+      "loss": 0.9985,
       "step": 6140
     },
     {
       "epoch": 85.42,
-      "learning_rate": 7.415492957746479e-06,
-      "loss": 1.2159,
+      "learning_rate": 7.408450704225353e-06,
+      "loss": 1.0315,
       "step": 6150
     },
     {
       "epoch": 85.55,
-      "learning_rate": 7.345070422535212e-06,
-      "loss": 1.2113,
+      "learning_rate": 7.338028169014084e-06,
+      "loss": 1.015,
       "step": 6160
     },
     {
       "epoch": 85.69,
-      "learning_rate": 7.274647887323945e-06,
-      "loss": 1.1814,
+      "learning_rate": 7.267605633802817e-06,
+      "loss": 1.0206,
       "step": 6170
     },
     {
       "epoch": 85.83,
-      "learning_rate": 7.2042253521126775e-06,
-      "loss": 1.2145,
+      "learning_rate": 7.197183098591549e-06,
+      "loss": 1.012,
       "step": 6180
     },
     {
       "epoch": 85.97,
-      "learning_rate": 7.1338028169014085e-06,
-      "loss": 1.1774,
+      "learning_rate": 7.126760563380282e-06,
+      "loss": 1.0117,
       "step": 6190
     },
     {
       "epoch": 86.0,
-      "eval_loss": 2.628901243209839,
-      "eval_runtime": 131.7057,
-      "eval_samples_per_second": 4.039,
-      "eval_steps_per_second": 0.509,
-      "eval_wer": 0.9703027576010799,
+      "eval_loss": 2.5671329498291016,
+      "eval_runtime": 127.179,
+      "eval_samples_per_second": 4.183,
+      "eval_steps_per_second": 0.527,
+      "eval_wer": 1.0119560326541106,
       "step": 6192
     },
     {
       "epoch": 86.11,
-      "learning_rate": 7.06338028169014e-06,
-      "loss": 1.2575,
+      "learning_rate": 7.056338028169014e-06,
+      "loss": 1.0274,
       "step": 6200
     },
     {
       "epoch": 86.25,
-      "learning_rate": 6.992957746478873e-06,
-      "loss": 1.1787,
+      "learning_rate": 6.985915492957746e-06,
+      "loss": 0.9968,
       "step": 6210
     },
     {
       "epoch": 86.39,
-      "learning_rate": 6.922535211267606e-06,
-      "loss": 1.2204,
+      "learning_rate": 6.915492957746479e-06,
+      "loss": 1.0117,
       "step": 6220
     },
     {
       "epoch": 86.53,
-      "learning_rate": 6.8521126760563385e-06,
-      "loss": 1.1907,
+      "learning_rate": 6.845070422535212e-06,
+      "loss": 0.9855,
       "step": 6230
     },
     {
       "epoch": 86.66,
-      "learning_rate": 6.78169014084507e-06,
-      "loss": 1.2021,
+      "learning_rate": 6.7746478873239444e-06,
+      "loss": 0.9979,
       "step": 6240
     },
     {
       "epoch": 86.8,
-      "learning_rate": 6.711267605633803e-06,
-      "loss": 1.1983,
+      "learning_rate": 6.704225352112676e-06,
+      "loss": 1.008,
       "step": 6250
     },
     {
       "epoch": 86.94,
-      "learning_rate": 6.640845070422536e-06,
-      "loss": 1.2002,
+      "learning_rate": 6.633802816901409e-06,
+      "loss": 1.0195,
       "step": 6260
     },
     {
       "epoch": 87.0,
-      "eval_loss": 2.719727039337158,
-      "eval_runtime": 128.6644,
-      "eval_samples_per_second": 4.135,
-      "eval_steps_per_second": 0.521,
-      "eval_wer": 0.9775020890917272,
+      "eval_loss": 2.5443294048309326,
+      "eval_runtime": 127.4113,
+      "eval_samples_per_second": 4.175,
+      "eval_steps_per_second": 0.526,
+      "eval_wer": 1.0041139037089413,
       "step": 6264
     },
     {
       "epoch": 87.08,
       "learning_rate": 6.570422535211268e-06,
-      "loss": 1.2199,
+      "loss": 1.0299,
       "step": 6270
     },
     {
       "epoch": 87.22,
       "learning_rate": 6.5000000000000004e-06,
-      "loss": 1.1853,
+      "loss": 1.0009,
       "step": 6280
     },
     {
       "epoch": 87.36,
       "learning_rate": 6.429577464788733e-06,
-      "loss": 1.2159,
+      "loss": 1.0081,
       "step": 6290
     },
     {
       "epoch": 87.5,
       "learning_rate": 6.359154929577466e-06,
-      "loss": 1.1772,
+      "loss": 0.9918,
       "step": 6300
     },
     {
       "epoch": 87.64,
       "learning_rate": 6.288732394366198e-06,
-      "loss": 1.2092,
+      "loss": 1.0123,
       "step": 6310
     },
     {
       "epoch": 87.78,
       "learning_rate": 6.21830985915493e-06,
-      "loss": 1.1923,
+      "loss": 1.0097,
       "step": 6320
     },
     {
       "epoch": 87.91,
       "learning_rate": 6.147887323943662e-06,
-      "loss": 1.2035,
+      "loss": 1.0114,
       "step": 6330
     },
     {
       "epoch": 88.0,
-      "eval_loss": 2.6153945922851562,
-      "eval_runtime": 128.668,
-      "eval_samples_per_second": 4.135,
-      "eval_steps_per_second": 0.521,
-      "eval_wer": 0.9778234878189882,
+      "eval_loss": 2.5627329349517822,
+      "eval_runtime": 127.3746,
+      "eval_samples_per_second": 4.177,
+      "eval_steps_per_second": 0.526,
+      "eval_wer": 1.004885260654368,
       "step": 6336
     },
     {
       "epoch": 88.06,
       "learning_rate": 6.077464788732395e-06,
-      "loss": 1.2339,
+      "loss": 1.0348,
       "step": 6340
     },
     {
       "epoch": 88.19,
       "learning_rate": 6.007042253521127e-06,
-      "loss": 1.1986,
+      "loss": 1.0066,
       "step": 6350
     },
     {
       "epoch": 88.33,
       "learning_rate": 5.93661971830986e-06,
-      "loss": 1.1895,
+      "loss": 1.0026,
       "step": 6360
     },
     {
       "epoch": 88.47,
       "learning_rate": 5.8661971830985916e-06,
-      "loss": 1.1787,
+      "loss": 1.0117,
       "step": 6370
     },
     {
       "epoch": 88.61,
       "learning_rate": 5.795774647887324e-06,
-      "loss": 1.202,
+      "loss": 0.988,
       "step": 6380
     },
     {
       "epoch": 88.75,
       "learning_rate": 5.725352112676056e-06,
-      "loss": 1.1641,
+      "loss": 0.9868,
       "step": 6390
     },
     {
       "epoch": 88.89,
       "learning_rate": 5.654929577464789e-06,
-      "loss": 1.2219,
+      "loss": 1.0074,
       "step": 6400
     },
     {
       "epoch": 89.0,
-      "eval_loss": 2.6426045894622803,
-      "eval_runtime": 130.6768,
-      "eval_samples_per_second": 4.071,
-      "eval_steps_per_second": 0.513,
-      "eval_wer": 0.9760879346917786,
+      "eval_loss": 2.566972494125366,
+      "eval_runtime": 127.5311,
+      "eval_samples_per_second": 4.172,
+      "eval_steps_per_second": 0.525,
+      "eval_wer": 1.0254547791990745,
       "step": 6408
     },
     {
       "epoch": 89.03,
       "learning_rate": 5.584507042253522e-06,
-      "loss": 1.2101,
+      "loss": 1.0154,
       "step": 6410
     },
     {
       "epoch": 89.17,
       "learning_rate": 5.514084507042254e-06,
-      "loss": 1.1915,
+      "loss": 1.0078,
       "step": 6420
     },
     {
       "epoch": 89.3,
       "learning_rate": 5.443661971830986e-06,
-      "loss": 1.1973,
+      "loss": 0.9927,
       "step": 6430
     },
     {
       "epoch": 89.44,
       "learning_rate": 5.373239436619719e-06,
-      "loss": 1.1821,
+      "loss": 1.0067,
       "step": 6440
     },
     {
       "epoch": 89.58,
       "learning_rate": 5.302816901408451e-06,
-      "loss": 1.1799,
+      "loss": 0.9898,
       "step": 6450
     },
     {
       "epoch": 89.72,
       "learning_rate": 5.2323943661971835e-06,
-      "loss": 1.181,
+      "loss": 1.0117,
       "step": 6460
     },
     {
       "epoch": 89.86,
       "learning_rate": 5.161971830985915e-06,
-      "loss": 1.2013,
+      "loss": 1.0002,
       "step": 6470
     },
     {
       "epoch": 90.0,
       "learning_rate": 5.091549295774648e-06,
-      "loss": 1.1724,
+      "loss": 0.9883,
       "step": 6480
     },
     {
       "epoch": 90.0,
-      "eval_loss": 2.599557399749756,
-      "eval_runtime": 132.1227,
-      "eval_samples_per_second": 4.027,
-      "eval_steps_per_second": 0.507,
-      "eval_wer": 0.9801375586552677,
+      "eval_loss": 2.5338053703308105,
+      "eval_runtime": 128.1248,
+      "eval_samples_per_second": 4.152,
+      "eval_steps_per_second": 0.523,
+      "eval_wer": 1.030597158835251,
       "step": 6480
     },
     {
       "epoch": 90.14,
       "learning_rate": 5.021126760563381e-06,
-      "loss": 1.2122,
+      "loss": 1.0219,
       "step": 6490
     },
     {
       "epoch": 90.28,
       "learning_rate": 4.950704225352113e-06,
-      "loss": 1.1918,
+      "loss": 0.9923,
       "step": 6500
     },
     {
       "epoch": 90.42,
       "learning_rate": 4.8802816901408454e-06,
-      "loss": 1.1766,
+      "loss": 0.99,
       "step": 6510
     },
     {
       "epoch": 90.55,
       "learning_rate": 4.809859154929577e-06,
-      "loss": 1.1855,
+      "loss": 0.9883,
       "step": 6520
     },
     {
       "epoch": 90.69,
       "learning_rate": 4.73943661971831e-06,
-      "loss": 1.1743,
+      "loss": 0.9859,
       "step": 6530
     },
     {
       "epoch": 90.83,
       "learning_rate": 4.669014084507043e-06,
-      "loss": 1.2096,
+      "loss": 0.9924,
       "step": 6540
     },
     {
       "epoch": 90.97,
       "learning_rate": 4.598591549295775e-06,
-      "loss": 1.1832,
+      "loss": 1.0112,
       "step": 6550
     },
     {
       "epoch": 91.0,
-      "eval_loss": 2.731451988220215,
-      "eval_runtime": 131.0173,
-      "eval_samples_per_second": 4.061,
-      "eval_steps_per_second": 0.511,
-      "eval_wer": 0.9742881018191168,
+      "eval_loss": 2.561525821685791,
+      "eval_runtime": 127.4376,
+      "eval_samples_per_second": 4.175,
+      "eval_steps_per_second": 0.526,
+      "eval_wer": 1.014205823744938,
       "step": 6552
     },
     {
       "epoch": 91.11,
       "learning_rate": 4.528169014084507e-06,
-      "loss": 1.2037,
+      "loss": 1.0043,
       "step": 6560
     },
     {
       "epoch": 91.25,
       "learning_rate": 4.45774647887324e-06,
-      "loss": 1.1731,
+      "loss": 0.9678,
       "step": 6570
     },
     {
       "epoch": 91.39,
       "learning_rate": 4.387323943661972e-06,
-      "loss": 1.2071,
+      "loss": 1.0007,
       "step": 6580
     },
     {
       "epoch": 91.53,
       "learning_rate": 4.316901408450704e-06,
-      "loss": 1.1664,
+      "loss": 0.977,
       "step": 6590
     },
     {
       "epoch": 91.66,
       "learning_rate": 4.2464788732394366e-06,
-      "loss": 1.1692,
+      "loss": 0.9891,
       "step": 6600
     },
     {
       "epoch": 91.8,
       "learning_rate": 4.176056338028169e-06,
-      "loss": 1.191,
+      "loss": 0.9976,
       "step": 6610
     },
     {
       "epoch": 91.94,
       "learning_rate": 4.105633802816901e-06,
-      "loss": 1.1759,
+      "loss": 0.9986,
       "step": 6620
     },
     {
       "epoch": 92.0,
-      "eval_loss": 2.5726349353790283,
-      "eval_runtime": 131.726,
-      "eval_samples_per_second": 4.039,
-      "eval_steps_per_second": 0.509,
-      "eval_wer": 0.9965288937455807,
+      "eval_loss": 2.5565598011016846,
+      "eval_runtime": 127.7934,
+      "eval_samples_per_second": 4.163,
+      "eval_steps_per_second": 0.524,
+      "eval_wer": 1.0415247155621263,
       "step": 6624
     },
     {
       "epoch": 92.08,
       "learning_rate": 4.035211267605634e-06,
-      "loss": 1.2287,
+      "loss": 1.0183,
       "step": 6630
     },
     {
       "epoch": 92.22,
       "learning_rate": 3.964788732394367e-06,
-      "loss": 1.1849,
+      "loss": 0.9992,
       "step": 6640
     },
     {
       "epoch": 92.36,
       "learning_rate": 3.894366197183099e-06,
-      "loss": 1.2151,
+      "loss": 0.9967,
       "step": 6650
     },
     {
       "epoch": 92.5,
       "learning_rate": 3.823943661971831e-06,
-      "loss": 1.1783,
+      "loss": 1.0078,
       "step": 6660
     },
     {
       "epoch": 92.64,
       "learning_rate": 3.753521126760563e-06,
-      "loss": 1.1818,
+      "loss": 0.9884,
       "step": 6670
     },
     {
       "epoch": 92.78,
       "learning_rate": 3.683098591549296e-06,
-      "loss": 1.1783,
+      "loss": 0.9895,
       "step": 6680
     },
     {
       "epoch": 92.91,
       "learning_rate": 3.612676056338028e-06,
-      "loss": 1.1638,
+      "loss": 0.9939,
       "step": 6690
     },
     {
       "epoch": 93.0,
-      "eval_loss": 2.6569957733154297,
-      "eval_runtime": 131.2307,
-      "eval_samples_per_second": 4.054,
-      "eval_steps_per_second": 0.511,
-      "eval_wer": 0.9865012534550364,
+      "eval_loss": 2.572765588760376,
+      "eval_runtime": 127.6748,
+      "eval_samples_per_second": 4.167,
+      "eval_steps_per_second": 0.525,
+      "eval_wer": 1.028733046217137,
       "step": 6696
     },
     {
       "epoch": 93.06,
       "learning_rate": 3.542253521126761e-06,
-      "loss": 1.2044,
+      "loss": 1.0046,
       "step": 6700
     },
     {
       "epoch": 93.19,
       "learning_rate": 3.471830985915493e-06,
-      "loss": 1.1595,
+      "loss": 0.984,
       "step": 6710
     },
     {
       "epoch": 93.33,
       "learning_rate": 3.401408450704226e-06,
-      "loss": 1.188,
+      "loss": 0.9735,
       "step": 6720
     },
     {
       "epoch": 93.47,
       "learning_rate": 3.330985915492958e-06,
-      "loss": 1.1614,
+      "loss": 0.9812,
       "step": 6730
     },
     {
       "epoch": 93.61,
       "learning_rate": 3.26056338028169e-06,
-      "loss": 1.1906,
+      "loss": 0.9914,
       "step": 6740
     },
     {
       "epoch": 93.75,
       "learning_rate": 3.1901408450704223e-06,
-      "loss": 1.1665,
+      "loss": 0.9929,
       "step": 6750
     },
     {
       "epoch": 93.89,
       "learning_rate": 3.119718309859155e-06,
-      "loss": 1.1872,
+      "loss": 0.9954,
       "step": 6760
     },
     {
       "epoch": 94.0,
-      "eval_loss": 2.6414103507995605,
-      "eval_runtime": 131.1073,
-      "eval_samples_per_second": 4.058,
-      "eval_steps_per_second": 0.511,
-      "eval_wer": 0.9947933406183711,
+      "eval_loss": 2.5616886615753174,
+      "eval_runtime": 127.5178,
+      "eval_samples_per_second": 4.172,
+      "eval_steps_per_second": 0.525,
+      "eval_wer": 1.0138201452722246,
       "step": 6768
     },
     {
       "epoch": 94.03,
       "learning_rate": 3.0492957746478873e-06,
-      "loss": 1.2089,
+      "loss": 1.0028,
       "step": 6770
     },
     {
       "epoch": 94.17,
       "learning_rate": 2.97887323943662e-06,
-      "loss": 1.1699,
+      "loss": 0.9896,
       "step": 6780
     },
     {
       "epoch": 94.3,
       "learning_rate": 2.9084507042253524e-06,
-      "loss": 1.1739,
+      "loss": 0.9781,
       "step": 6790
     },
     {
       "epoch": 94.44,
       "learning_rate": 2.8380281690140847e-06,
-      "loss": 1.1663,
+      "loss": 0.9773,
       "step": 6800
     },
     {
       "epoch": 94.58,
       "learning_rate": 2.767605633802817e-06,
-      "loss": 1.1771,
+      "loss": 0.9818,
       "step": 6810
     },
     {
       "epoch": 94.72,
       "learning_rate": 2.6971830985915497e-06,
-      "loss": 1.1685,
+      "loss": 0.997,
       "step": 6820
     },
     {
       "epoch": 94.86,
       "learning_rate": 2.6267605633802816e-06,
-      "loss": 1.2045,
+      "loss": 0.9835,
       "step": 6830
     },
     {
       "epoch": 95.0,
       "learning_rate": 2.5563380281690143e-06,
-      "loss": 1.144,
+      "loss": 0.9643,
       "step": 6840
     },
     {
       "epoch": 95.0,
-      "eval_loss": 2.6263763904571533,
-      "eval_runtime": 131.0288,
-      "eval_samples_per_second": 4.06,
-      "eval_steps_per_second": 0.511,
-      "eval_wer": 0.9863084142186797,
+      "eval_loss": 2.5889523029327393,
+      "eval_runtime": 127.4941,
+      "eval_samples_per_second": 4.173,
+      "eval_steps_per_second": 0.526,
+      "eval_wer": 1.014527222472199,
       "step": 6840
     },
     {
       "epoch": 95.14,
       "learning_rate": 2.4859154929577466e-06,
-      "loss": 1.2542,
+      "loss": 1.0048,
       "step": 6850
     },
     {
       "epoch": 95.28,
       "learning_rate": 2.4154929577464793e-06,
-      "loss": 1.1588,
+      "loss": 0.9637,
       "step": 6860
     },
     {
       "epoch": 95.42,
       "learning_rate": 2.345070422535211e-06,
-      "loss": 1.1804,
+      "loss": 0.9962,
       "step": 6870
     },
     {
       "epoch": 95.55,
       "learning_rate": 2.274647887323944e-06,
-      "loss": 1.1847,
+      "loss": 0.9706,
       "step": 6880
     },
     {
       "epoch": 95.69,
       "learning_rate": 2.204225352112676e-06,
-      "loss": 1.1723,
+      "loss": 0.9884,
       "step": 6890
     },
     {
       "epoch": 95.83,
       "learning_rate": 2.1338028169014085e-06,
-      "loss": 1.1867,
+      "loss": 0.9861,
       "step": 6900
     },
     {
       "epoch": 95.97,
       "learning_rate": 2.063380281690141e-06,
-      "loss": 1.1636,
+      "loss": 0.9892,
       "step": 6910
     },
     {
       "epoch": 96.0,
-      "eval_loss": 2.5819830894470215,
-      "eval_runtime": 130.9251,
-      "eval_samples_per_second": 4.063,
-      "eval_steps_per_second": 0.512,
-      "eval_wer": 0.9917721925821174,
+      "eval_loss": 2.5917582511901855,
+      "eval_runtime": 128.3033,
+      "eval_samples_per_second": 4.146,
+      "eval_steps_per_second": 0.522,
+      "eval_wer": 1.0118917529086584,
       "step": 6912
     },
     {
       "epoch": 96.11,
       "learning_rate": 1.9929577464788735e-06,
-      "loss": 1.2053,
+      "loss": 0.9988,
       "step": 6920
     },
     {
       "epoch": 96.25,
       "learning_rate": 1.922535211267606e-06,
-      "loss": 1.1621,
+      "loss": 0.9786,
       "step": 6930
     },
     {
       "epoch": 96.39,
       "learning_rate": 1.8521126760563383e-06,
-      "loss": 1.1944,
+      "loss": 0.9927,
       "step": 6940
     },
     {
       "epoch": 96.53,
       "learning_rate": 1.7816901408450704e-06,
-      "loss": 1.1547,
+      "loss": 0.9801,
       "step": 6950
     },
     {
       "epoch": 96.66,
       "learning_rate": 1.711267605633803e-06,
-      "loss": 1.1609,
+      "loss": 0.9774,
       "step": 6960
     },
     {
       "epoch": 96.8,
       "learning_rate": 1.6408450704225354e-06,
-      "loss": 1.1928,
+      "loss": 0.9754,
       "step": 6970
     },
     {
       "epoch": 96.94,
       "learning_rate": 1.5704225352112675e-06,
-      "loss": 1.1714,
+      "loss": 0.983,
       "step": 6980
     },
     {
       "epoch": 97.0,
-      "eval_loss": 2.596989870071411,
-      "eval_runtime": 130.9899,
-      "eval_samples_per_second": 4.061,
-      "eval_steps_per_second": 0.511,
-      "eval_wer": 0.991322234363952,
+      "eval_loss": 2.5861523151397705,
+      "eval_runtime": 127.885,
+      "eval_samples_per_second": 4.16,
+      "eval_steps_per_second": 0.524,
+      "eval_wer": 1.0175483705084527,
       "step": 6984
     },
     {
       "epoch": 97.08,
       "learning_rate": 1.5e-06,
-      "loss": 1.211,
+      "loss": 0.9888,
       "step": 6990
     },
     {
       "epoch": 97.22,
       "learning_rate": 1.4295774647887325e-06,
-      "loss": 1.1678,
+      "loss": 0.9668,
       "step": 7000
     },
     {
       "epoch": 97.36,
       "learning_rate": 1.3591549295774648e-06,
-      "loss": 1.185,
+      "loss": 0.9922,
       "step": 7010
     },
     {
       "epoch": 97.5,
       "learning_rate": 1.2887323943661974e-06,
-      "loss": 1.1638,
+      "loss": 0.978,
       "step": 7020
     },
     {
       "epoch": 97.64,
       "learning_rate": 1.2183098591549297e-06,
-      "loss": 1.1858,
+      "loss": 0.977,
       "step": 7030
     },
     {
       "epoch": 97.78,
       "learning_rate": 1.1478873239436622e-06,
-      "loss": 1.1571,
+      "loss": 0.9724,
       "step": 7040
     },
     {
       "epoch": 97.91,
       "learning_rate": 1.0774647887323945e-06,
-      "loss": 1.1943,
+      "loss": 0.988,
       "step": 7050
     },
     {
       "epoch": 98.0,
-      "eval_loss": 2.6308441162109375,
-      "eval_runtime": 131.2345,
-      "eval_samples_per_second": 4.054,
-      "eval_steps_per_second": 0.511,
-      "eval_wer": 0.9894581217458379,
+      "eval_loss": 2.587308645248413,
+      "eval_runtime": 128.6966,
+      "eval_samples_per_second": 4.134,
+      "eval_steps_per_second": 0.521,
+      "eval_wer": 1.0146557819631035,
       "step": 7056
     },
     {
       "epoch": 98.06,
       "learning_rate": 1.007042253521127e-06,
-      "loss": 1.194,
+      "loss": 0.9851,
       "step": 7060
     },
     {
       "epoch": 98.19,
       "learning_rate": 9.366197183098592e-07,
-      "loss": 1.1703,
+      "loss": 0.9803,
       "step": 7070
     },
     {
       "epoch": 98.33,
       "learning_rate": 8.661971830985915e-07,
-      "loss": 1.1724,
+      "loss": 0.9714,
       "step": 7080
     },
     {
       "epoch": 98.47,
       "learning_rate": 7.95774647887324e-07,
-      "loss": 1.1557,
+      "loss": 0.9784,
       "step": 7090
     },
     {
       "epoch": 98.61,
       "learning_rate": 7.253521126760564e-07,
-      "loss": 1.1746,
+      "loss": 0.9682,
       "step": 7100
     },
     {
       "epoch": 98.75,
       "learning_rate": 6.549295774647888e-07,
-      "loss": 1.1476,
+      "loss": 0.9569,
       "step": 7110
     },
     {
       "epoch": 98.89,
       "learning_rate": 5.845070422535212e-07,
-      "loss": 1.185,
+      "loss": 0.9908,
       "step": 7120
     },
     {
       "epoch": 99.0,
-      "eval_loss": 2.637880325317383,
-      "eval_runtime": 131.2546,
-      "eval_samples_per_second": 4.053,
-      "eval_steps_per_second": 0.51,
-      "eval_wer": 0.9898438002185511,
+      "eval_loss": 2.5973317623138428,
+      "eval_runtime": 127.2594,
+      "eval_samples_per_second": 4.18,
+      "eval_steps_per_second": 0.526,
+      "eval_wer": 1.0072636112360995,
       "step": 7128
     },
     {
       "epoch": 99.03,
       "learning_rate": 5.140845070422535e-07,
-      "loss": 1.2175,
+      "loss": 1.0143,
       "step": 7130
     },
     {
       "epoch": 99.17,
       "learning_rate": 4.436619718309859e-07,
-      "loss": 1.1628,
+      "loss": 0.9764,
       "step": 7140
     },
     {
       "epoch": 99.3,
       "learning_rate": 3.7323943661971836e-07,
-      "loss": 1.1645,
+      "loss": 0.9699,
       "step": 7150
     },
     {
       "epoch": 99.44,
       "learning_rate": 3.0281690140845076e-07,
-      "loss": 1.1748,
+      "loss": 0.9769,
       "step": 7160
     },
     {
       "epoch": 99.58,
       "learning_rate": 2.323943661971831e-07,
-      "loss": 1.164,
+      "loss": 0.9682,
       "step": 7170
     },
     {
       "epoch": 99.72,
       "learning_rate": 1.619718309859155e-07,
-      "loss": 1.1703,
+      "loss": 0.9809,
       "step": 7180
     },
     {
       "epoch": 99.86,
       "learning_rate": 9.154929577464789e-08,
-      "loss": 1.183,
+      "loss": 0.9763,
       "step": 7190
     },
     {
       "epoch": 100.0,
       "learning_rate": 2.1126760563380285e-08,
-      "loss": 1.1569,
+      "loss": 0.9696,
       "step": 7200
     },
     {
       "epoch": 100.0,
-      "eval_loss": 2.6330785751342773,
-      "eval_runtime": 131.615,
-      "eval_samples_per_second": 4.042,
-      "eval_steps_per_second": 0.509,
-      "eval_wer": 0.9906151571639776,
+      "eval_loss": 2.5937769412994385,
+      "eval_runtime": 128.9867,
+      "eval_samples_per_second": 4.124,
+      "eval_steps_per_second": 0.519,
+      "eval_wer": 1.0156199781448865,
       "step": 7200
     },
     {
       "epoch": 100.0,
       "step": 7200,
       "total_flos": 2.7202925310621965e+20,
-      "train_loss": 1.475777028269238,
-      "train_runtime": 191216.7272,
-      "train_samples_per_second": 2.415,
+      "train_loss": 1.1222029450204638,
+      "train_runtime": 188806.8283,
+      "train_samples_per_second": 2.445,
       "train_steps_per_second": 0.038
     }
   ],