large_sami_cont_pt_22k_finetuned_pseudolabels

README.md

@@ -16,9 +16,9 @@ should probably proofread and complete it, then remove this comment. -->
 
 This model was trained from scratch on an unknown dataset.
 It achieves the following results on the evaluation set:
-- Loss: 447.4113
-- Wer: 0.4189
-- Cer: 0.1464
+- Loss: 204.4745
+- Wer: 0.3799
+- Cer: 0.1198
 
 ## Model description
 

all_results.json

@@ -1,16 +1,16 @@
 {
     "epoch": 60.0,
-    "eval_cer": 0.11855950205009139,
-    "eval_loss": 223.1608123779297,
-    "eval_runtime": 25.7763,
+    "eval_cer": 0.11979449686311318,
+    "eval_loss": 204.47450256347656,
+    "eval_runtime": 26.0878,
     "eval_samples": 890,
-    "eval_samples_per_second": 34.528,
-    "eval_steps_per_second": 4.345,
-    "eval_wer": 0.3813131313131313,
+    "eval_samples_per_second": 34.116,
+    "eval_steps_per_second": 4.293,
+    "eval_wer": 0.37989267676767674,
     "total_flos": 1.8440987587856836e+20,
-    "train_loss": 306.16538580246913,
-    "train_runtime": 47401.1605,
+    "train_loss": 277.69142749927664,
+    "train_runtime": 46658.0373,
     "train_samples": 17267,
-    "train_samples_per_second": 21.856,
-    "train_steps_per_second": 1.367
+    "train_samples_per_second": 22.205,
+    "train_steps_per_second": 1.389
 }

eval--1hr test set_results.json

@@ -1,10 +1,10 @@
 {
     "epoch": 60.0,
-    "eval_cer": 0.11855950205009139,
-    "eval_loss": 223.1608123779297,
-    "eval_runtime": 25.7763,
+    "eval_cer": 0.11979449686311318,
+    "eval_loss": 204.47450256347656,
+    "eval_runtime": 26.0878,
     "eval_samples": 890,
-    "eval_samples_per_second": 34.528,
-    "eval_steps_per_second": 4.345,
-    "eval_wer": 0.3813131313131313
+    "eval_samples_per_second": 34.116,
+    "eval_steps_per_second": 4.293,
+    "eval_wer": 0.37989267676767674
 }

train_results.json

@@ -1,9 +1,9 @@
 {
     "epoch": 60.0,
     "total_flos": 1.8440987587856836e+20,
-    "train_loss": 306.16538580246913,
-    "train_runtime": 47401.1605,
+    "train_loss": 277.69142749927664,
+    "train_runtime": 46658.0373,
     "train_samples": 17267,
-    "train_samples_per_second": 21.856,
-    "train_steps_per_second": 1.367
+    "train_samples_per_second": 22.205,
+    "train_steps_per_second": 1.389
 }

trainer_state.json

@@ -1,6 +1,6 @@
 {
-  "best_metric": 0.3817866161616162,
-  "best_model_checkpoint": "/scratch/elec/puhe/p/palp3/sami_ASR/large_model_output/large-sami-22k/outputs/checkpoint-2160",
+  "best_metric": 0.3803661616161616,
+  "best_model_checkpoint": "/scratch/elec/puhe/p/palp3/sami_ASR/large_model_output/large-sami-cont-pt-22k/outputs/checkpoint-3240",
   "epoch": 60.0,
   "eval_steps": 500,
   "global_step": 64800,
@@ -10,1032 +10,1032 @@
   "log_history": [
     {
       "epoch": 1.0,
-      "grad_norm": 843.546142578125,
-      "learning_rate": 3.277777777777778e-05,
-      "loss": 3985.5875,
+      "grad_norm": 119.4698486328125,
+      "learning_rate": 3.290123456790124e-05,
+      "loss": 5958.8606,
       "step": 1080
     },
     {
       "epoch": 1.0,
-      "eval_cer": 0.16660080027663884,
-      "eval_loss": 283.68731689453125,
-      "eval_runtime": 26.5521,
-      "eval_samples_per_second": 33.519,
-      "eval_steps_per_second": 4.218,
-      "eval_wer": 0.5932765151515151,
+      "eval_cer": 1.0,
+      "eval_loss": 1153.3079833984375,
+      "eval_runtime": 25.943,
+      "eval_samples_per_second": 34.306,
+      "eval_steps_per_second": 4.317,
+      "eval_wer": 1.0,
       "step": 1080
     },
     {
       "epoch": 2.0,
-      "grad_norm": 152.15969848632812,
-      "learning_rate": 6.611111111111111e-05,
-      "loss": 665.3671,
+      "grad_norm": 67.0517578125,
+      "learning_rate": 6.623456790123457e-05,
+      "loss": 2592.9389,
       "step": 2160
     },
     {
       "epoch": 2.0,
-      "eval_cer": 0.11860890184261226,
-      "eval_loss": 223.17312622070312,
-      "eval_runtime": 26.9667,
-      "eval_samples_per_second": 33.004,
-      "eval_steps_per_second": 4.153,
-      "eval_wer": 0.3817866161616162,
+      "eval_cer": 0.15714074000889197,
+      "eval_loss": 283.1771545410156,
+      "eval_runtime": 26.1688,
+      "eval_samples_per_second": 34.01,
+      "eval_steps_per_second": 4.28,
+      "eval_wer": 0.49510732323232326,
       "step": 2160
     },
     {
       "epoch": 3.0,
-      "grad_norm": 69.1477279663086,
-      "learning_rate": 9.944444444444444e-05,
-      "loss": 481.9958,
+      "grad_norm": 232.4710693359375,
+      "learning_rate": 9.956790123456791e-05,
+      "loss": 537.566,
       "step": 3240
     },
     {
       "epoch": 3.0,
-      "eval_cer": 0.1353801314034481,
-      "eval_loss": 223.3375244140625,
-      "eval_runtime": 26.3476,
-      "eval_samples_per_second": 33.779,
-      "eval_steps_per_second": 4.251,
-      "eval_wer": 0.4375,
+      "eval_cer": 0.1192510991453836,
+      "eval_loss": 201.5854949951172,
+      "eval_runtime": 26.0419,
+      "eval_samples_per_second": 34.176,
+      "eval_steps_per_second": 4.301,
+      "eval_wer": 0.3803661616161616,
       "step": 3240
     },
     {
       "epoch": 4.0,
-      "grad_norm": 252.46255493164062,
-      "learning_rate": 0.00013277777777777778,
-      "loss": 442.5208,
+      "grad_norm": 27.231971740722656,
+      "learning_rate": 0.00013290123456790123,
+      "loss": 384.8012,
       "step": 4320
     },
     {
       "epoch": 4.0,
-      "eval_cer": 0.14442029343476756,
-      "eval_loss": 236.09030151367188,
-      "eval_runtime": 26.3227,
-      "eval_samples_per_second": 33.811,
-      "eval_steps_per_second": 4.255,
-      "eval_wer": 0.42329545454545453,
+      "eval_cer": 0.11917699945660228,
+      "eval_loss": 204.7545623779297,
+      "eval_runtime": 26.024,
+      "eval_samples_per_second": 34.199,
+      "eval_steps_per_second": 4.304,
+      "eval_wer": 0.38273358585858586,
       "step": 4320
     },
     {
       "epoch": 5.0,
-      "grad_norm": 23.017539978027344,
-      "learning_rate": 0.0001661111111111111,
-      "loss": 421.2154,
+      "grad_norm": 16.580591201782227,
+      "learning_rate": 0.00016623456790123455,
+      "loss": 329.1704,
       "step": 5400
     },
     {
       "epoch": 5.0,
-      "eval_cer": 0.15635034332855802,
-      "eval_loss": 304.9386901855469,
-      "eval_runtime": 27.9275,
-      "eval_samples_per_second": 31.868,
-      "eval_steps_per_second": 4.01,
-      "eval_wer": 0.46259469696969696,
+      "eval_cer": 0.13298424146618584,
+      "eval_loss": 222.9210662841797,
+      "eval_runtime": 29.4239,
+      "eval_samples_per_second": 30.248,
+      "eval_steps_per_second": 3.806,
+      "eval_wer": 0.40041035353535354,
       "step": 5400
     },
     {
       "epoch": 6.0,
-      "grad_norm": 154.2694549560547,
-      "learning_rate": 0.00019944444444444445,
-      "loss": 431.8639,
+      "grad_norm": 49.58544158935547,
+      "learning_rate": 0.00019956790123456793,
+      "loss": 297.2885,
       "step": 6480
     },
     {
       "epoch": 6.0,
-      "eval_cer": 0.17932124685076323,
-      "eval_loss": 298.5758056640625,
-      "eval_runtime": 26.8301,
-      "eval_samples_per_second": 33.172,
-      "eval_steps_per_second": 4.174,
-      "eval_wer": 0.5198863636363636,
+      "eval_cer": 0.13730672331176208,
+      "eval_loss": 234.79901123046875,
+      "eval_runtime": 28.052,
+      "eval_samples_per_second": 31.727,
+      "eval_steps_per_second": 3.993,
+      "eval_wer": 0.4163510101010101,
       "step": 6480
     },
     {
       "epoch": 7.0,
-      "grad_norm": 252.90023803710938,
-      "learning_rate": 0.0002327777777777778,
-      "loss": 444.0917,
+      "grad_norm": 301.71343994140625,
+      "learning_rate": 0.00023290123456790125,
+      "loss": 289.8471,
       "step": 7560
     },
     {
       "epoch": 7.0,
-      "eval_cer": 0.198809465000247,
-      "eval_loss": 347.2078857421875,
-      "eval_runtime": 26.7772,
-      "eval_samples_per_second": 33.237,
-      "eval_steps_per_second": 4.183,
-      "eval_wer": 0.55697601010101,
+      "eval_cer": 0.1443708936422467,
+      "eval_loss": 269.34783935546875,
+      "eval_runtime": 28.1338,
+      "eval_samples_per_second": 31.634,
+      "eval_steps_per_second": 3.981,
+      "eval_wer": 0.4202967171717172,
       "step": 7560
     },
     {
       "epoch": 8.0,
-      "grad_norm": 205.64749145507812,
-      "learning_rate": 0.0002661111111111111,
-      "loss": 451.4575,
+      "grad_norm": 198.65408325195312,
+      "learning_rate": 0.00026623456790123457,
+      "loss": 279.9123,
       "step": 8640
     },
     {
       "epoch": 8.0,
-      "eval_cer": 0.20920812132589042,
-      "eval_loss": 375.0052795410156,
-      "eval_runtime": 26.6502,
-      "eval_samples_per_second": 33.396,
-      "eval_steps_per_second": 4.203,
-      "eval_wer": 0.5730744949494949,
+      "eval_cer": 0.1378007212369708,
+      "eval_loss": 275.84033203125,
+      "eval_runtime": 28.0629,
+      "eval_samples_per_second": 31.714,
+      "eval_steps_per_second": 3.991,
+      "eval_wer": 0.4045138888888889,
       "step": 8640
     },
     {
       "epoch": 9.0,
-      "grad_norm": 186.66598510742188,
-      "learning_rate": 0.00029944444444444446,
-      "loss": 445.6943,
+      "grad_norm": 12.690323829650879,
+      "learning_rate": 0.0002995679012345679,
+      "loss": 266.2481,
       "step": 9720
     },
     {
       "epoch": 9.0,
-      "eval_cer": 0.24057698957664378,
-      "eval_loss": 360.6094665527344,
-      "eval_runtime": 26.557,
-      "eval_samples_per_second": 33.513,
-      "eval_steps_per_second": 4.217,
-      "eval_wer": 0.6164772727272727,
+      "eval_cer": 0.15694314083880848,
+      "eval_loss": 283.7547607421875,
+      "eval_runtime": 27.7523,
+      "eval_samples_per_second": 32.069,
+      "eval_steps_per_second": 4.036,
+      "eval_wer": 0.4548611111111111,
       "step": 9720
     },
     {
       "epoch": 10.0,
-      "grad_norm": 148.4665985107422,
-      "learning_rate": 0.0003327777777777778,
-      "loss": 457.5267,
+      "grad_norm": 159.38430786132812,
+      "learning_rate": 0.0003329012345679012,
+      "loss": 262.1636,
       "step": 10800
     },
     {
       "epoch": 10.0,
-      "eval_cer": 0.252210640715309,
-      "eval_loss": 413.63214111328125,
-      "eval_runtime": 27.5043,
-      "eval_samples_per_second": 32.359,
-      "eval_steps_per_second": 4.072,
-      "eval_wer": 0.6510416666666666,
+      "eval_cer": 0.14177740453490095,
+      "eval_loss": 295.8691101074219,
+      "eval_runtime": 26.2783,
+      "eval_samples_per_second": 33.868,
+      "eval_steps_per_second": 4.262,
+      "eval_wer": 0.41761363636363635,
       "step": 10800
     },
     {
       "epoch": 11.0,
-      "grad_norm": 83.87371063232422,
-      "learning_rate": 0.00036608024691358026,
-      "loss": 468.0921,
+      "grad_norm": 83.4280014038086,
+      "learning_rate": 0.0003662037037037037,
+      "loss": 261.9983,
       "step": 11880
     },
     {
       "epoch": 11.0,
-      "eval_cer": 0.2446771723558761,
-      "eval_loss": 405.5586853027344,
-      "eval_runtime": 26.521,
-      "eval_samples_per_second": 33.558,
-      "eval_steps_per_second": 4.223,
-      "eval_wer": 0.6422032828282829,
+      "eval_cer": 0.16474830805710616,
+      "eval_loss": 317.6566162109375,
+      "eval_runtime": 27.4165,
+      "eval_samples_per_second": 32.462,
+      "eval_steps_per_second": 4.085,
+      "eval_wer": 0.4649621212121212,
       "step": 11880
     },
     {
       "epoch": 12.0,
-      "grad_norm": 115.64622497558594,
-      "learning_rate": 0.0003994135802469136,
-      "loss": 469.1014,
+      "grad_norm": 140.5787811279297,
+      "learning_rate": 0.00039953703703703706,
+      "loss": 250.5207,
       "step": 12960
     },
     {
       "epoch": 12.0,
-      "eval_cer": 0.24956775181544238,
-      "eval_loss": 423.7670593261719,
-      "eval_runtime": 27.3392,
-      "eval_samples_per_second": 32.554,
-      "eval_steps_per_second": 4.097,
-      "eval_wer": 0.6369949494949495,
+      "eval_cer": 0.1583510349256533,
+      "eval_loss": 293.32470703125,
+      "eval_runtime": 26.0661,
+      "eval_samples_per_second": 34.144,
+      "eval_steps_per_second": 4.297,
+      "eval_wer": 0.4670138888888889,
       "step": 12960
     },
     {
       "epoch": 13.0,
-      "grad_norm": 154.60157775878906,
-      "learning_rate": 0.0004327469135802469,
-      "loss": 472.7028,
+      "grad_norm": 11.17358684539795,
+      "learning_rate": 0.00043287037037037035,
+      "loss": 253.3931,
       "step": 14040
     },
     {
       "epoch": 13.0,
-      "eval_cer": 0.25208714123400683,
-      "eval_loss": 476.42498779296875,
-      "eval_runtime": 26.8242,
-      "eval_samples_per_second": 33.179,
-      "eval_steps_per_second": 4.175,
-      "eval_wer": 0.6583017676767676,
+      "eval_cer": 0.1578570370004446,
+      "eval_loss": 386.1613464355469,
+      "eval_runtime": 25.6588,
+      "eval_samples_per_second": 34.686,
+      "eval_steps_per_second": 4.365,
+      "eval_wer": 0.4613320707070707,
       "step": 14040
     },
     {
       "epoch": 14.0,
-      "grad_norm": 35.594207763671875,
-      "learning_rate": 0.00046608024691358025,
-      "loss": 477.4817,
+      "grad_norm": 35.230567932128906,
+      "learning_rate": 0.00046620370370370376,
+      "loss": 256.4585,
       "step": 15120
     },
     {
       "epoch": 14.0,
-      "eval_cer": 0.2522847404040903,
-      "eval_loss": 431.7320861816406,
-      "eval_runtime": 25.8162,
-      "eval_samples_per_second": 34.474,
-      "eval_steps_per_second": 4.338,
-      "eval_wer": 0.6496212121212122,
+      "eval_cer": 0.16645260089907624,
+      "eval_loss": 304.7311706542969,
+      "eval_runtime": 26.1782,
+      "eval_samples_per_second": 33.998,
+      "eval_steps_per_second": 4.278,
+      "eval_wer": 0.4690656565656566,
       "step": 15120
     },
     {
       "epoch": 15.0,
-      "grad_norm": 5.936054706573486,
-      "learning_rate": 0.0004994135802469135,
-      "loss": 476.1046,
+      "grad_norm": 50.54726791381836,
+      "learning_rate": 0.0004995061728395062,
+      "loss": 243.4571,
       "step": 16200
     },
     {
       "epoch": 15.0,
-      "eval_cer": 0.2686113718322383,
-      "eval_loss": 456.8628234863281,
-      "eval_runtime": 27.4231,
-      "eval_samples_per_second": 32.454,
-      "eval_steps_per_second": 4.084,
-      "eval_wer": 0.6734532828282829,
+      "eval_cer": 0.17455416687249914,
+      "eval_loss": 320.8733825683594,
+      "eval_runtime": 27.8676,
+      "eval_samples_per_second": 31.937,
+      "eval_steps_per_second": 4.019,
+      "eval_wer": 0.4971590909090909,
       "step": 16200
     },
     {
       "epoch": 16.0,
-      "grad_norm": 66.7186279296875,
-      "learning_rate": 0.0004890843621399176,
-      "loss": 478.12,
+      "grad_norm": 42.936180114746094,
+      "learning_rate": 0.0004890534979423869,
+      "loss": 238.0247,
       "step": 17280
     },
     {
       "epoch": 16.0,
-      "eval_cer": 0.25418663241614387,
-      "eval_loss": 477.70703125,
-      "eval_runtime": 27.501,
-      "eval_samples_per_second": 32.362,
-      "eval_steps_per_second": 4.073,
-      "eval_wer": 0.6666666666666666,
+      "eval_cer": 0.1745294669762387,
+      "eval_loss": 322.06427001953125,
+      "eval_runtime": 27.4467,
+      "eval_samples_per_second": 32.427,
+      "eval_steps_per_second": 4.081,
+      "eval_wer": 0.5134154040404041,
       "step": 17280
     },
     {
       "epoch": 17.0,
-      "grad_norm": 5.836469650268555,
-      "learning_rate": 0.0004779938271604938,
-      "loss": 439.3815,
+      "grad_norm": 1.6827819347381592,
+      "learning_rate": 0.0004779423868312757,
+      "loss": 222.2451,
       "step": 18360
     },
     {
       "epoch": 17.0,
-      "eval_cer": 0.25999110803734626,
-      "eval_loss": 441.1052551269531,
-      "eval_runtime": 26.8497,
-      "eval_samples_per_second": 33.147,
-      "eval_steps_per_second": 4.171,
-      "eval_wer": 0.6576704545454546,
+      "eval_cer": 0.17255347527540385,
+      "eval_loss": 329.7898254394531,
+      "eval_runtime": 25.8325,
+      "eval_samples_per_second": 34.453,
+      "eval_steps_per_second": 4.336,
+      "eval_wer": 0.48390151515151514,
       "step": 18360
     },
     {
       "epoch": 18.0,
-      "grad_norm": 4.255939483642578,
-      "learning_rate": 0.0004668827160493827,
-      "loss": 406.8238,
+      "grad_norm": 0.07622645050287247,
+      "learning_rate": 0.0004668312757201646,
+      "loss": 207.317,
       "step": 19440
     },
     {
       "epoch": 18.0,
-      "eval_cer": 0.24961715160796324,
-      "eval_loss": 417.4708557128906,
-      "eval_runtime": 26.667,
-      "eval_samples_per_second": 33.375,
-      "eval_steps_per_second": 4.2,
-      "eval_wer": 0.6792929292929293,
+      "eval_cer": 0.17309687299313342,
+      "eval_loss": 318.4466247558594,
+      "eval_runtime": 25.9662,
+      "eval_samples_per_second": 34.275,
+      "eval_steps_per_second": 4.313,
+      "eval_wer": 0.4790088383838384,
       "step": 19440
     },
     {
       "epoch": 19.0,
-      "grad_norm": 0.6231774687767029,
-      "learning_rate": 0.0004557716049382716,
-      "loss": 381.221,
+      "grad_norm": 0.3686951994895935,
+      "learning_rate": 0.00045575102880658435,
+      "loss": 197.8997,
       "step": 20520
     },
     {
       "epoch": 19.0,
-      "eval_cer": 0.23459961468161833,
-      "eval_loss": 385.214599609375,
-      "eval_runtime": 27.8465,
-      "eval_samples_per_second": 31.961,
-      "eval_steps_per_second": 4.022,
-      "eval_wer": 0.6339962121212122,
+      "eval_cer": 0.17159017932124684,
+      "eval_loss": 351.1441650390625,
+      "eval_runtime": 25.816,
+      "eval_samples_per_second": 34.475,
+      "eval_steps_per_second": 4.338,
+      "eval_wer": 0.4840593434343434,
       "step": 20520
     },
     {
       "epoch": 20.0,
-      "grad_norm": 7.669593811035156,
-      "learning_rate": 0.0004446604938271605,
-      "loss": 349.7026,
+      "grad_norm": 30.71696662902832,
+      "learning_rate": 0.00044463991769547325,
+      "loss": 179.4079,
       "step": 21600
     },
     {
       "epoch": 20.0,
-      "eval_cer": 0.24514647038482437,
-      "eval_loss": 417.8040771484375,
-      "eval_runtime": 26.5289,
-      "eval_samples_per_second": 33.548,
-      "eval_steps_per_second": 4.222,
-      "eval_wer": 0.6319444444444444,
+      "eval_cer": 0.1729486736155708,
+      "eval_loss": 331.9079284667969,
+      "eval_runtime": 26.2966,
+      "eval_samples_per_second": 33.845,
+      "eval_steps_per_second": 4.259,
+      "eval_wer": 0.48374368686868685,
       "step": 21600
     },
     {
       "epoch": 21.0,
-      "grad_norm": 45.017032623291016,
-      "learning_rate": 0.000433559670781893,
-      "loss": 336.5142,
+      "grad_norm": 38.379737854003906,
+      "learning_rate": 0.00043352880658436215,
+      "loss": 172.7036,
       "step": 22680
     },
     {
       "epoch": 21.0,
-      "eval_cer": 0.2432198784765104,
-      "eval_loss": 446.5249328613281,
-      "eval_runtime": 27.1849,
-      "eval_samples_per_second": 32.739,
-      "eval_steps_per_second": 4.12,
-      "eval_wer": 0.6699810606060606,
+      "eval_cer": 0.17331917205947736,
+      "eval_loss": 329.3490905761719,
+      "eval_runtime": 26.7664,
+      "eval_samples_per_second": 33.251,
+      "eval_steps_per_second": 4.184,
+      "eval_wer": 0.4965277777777778,
       "step": 22680
     },
     {
       "epoch": 22.0,
-      "grad_norm": 43.4493522644043,
-      "learning_rate": 0.0004224485596707819,
-      "loss": 316.0003,
+      "grad_norm": 28.166221618652344,
+      "learning_rate": 0.00042241769547325106,
+      "loss": 164.6914,
       "step": 23760
     },
     {
       "epoch": 22.0,
-      "eval_cer": 0.22963493553327077,
-      "eval_loss": 414.8999938964844,
-      "eval_runtime": 26.6715,
-      "eval_samples_per_second": 33.369,
-      "eval_steps_per_second": 4.199,
-      "eval_wer": 0.657354797979798,
+      "eval_cer": 0.17089858222595464,
+      "eval_loss": 323.5361328125,
+      "eval_runtime": 26.5854,
+      "eval_samples_per_second": 33.477,
+      "eval_steps_per_second": 4.213,
+      "eval_wer": 0.4741161616161616,
       "step": 23760
     },
     {
       "epoch": 23.0,
-      "grad_norm": 74.57106018066406,
-      "learning_rate": 0.0004113374485596708,
-      "loss": 301.4025,
+      "grad_norm": 31.44266700744629,
+      "learning_rate": 0.0004113065843621399,
+      "loss": 162.8205,
       "step": 24840
     },
     {
       "epoch": 23.0,
-      "eval_cer": 0.22943733636318728,
-      "eval_loss": 435.7232360839844,
-      "eval_runtime": 27.1302,
-      "eval_samples_per_second": 32.805,
-      "eval_steps_per_second": 4.128,
-      "eval_wer": 0.6242108585858586,
+      "eval_cer": 0.16445190930198092,
+      "eval_loss": 330.853515625,
+      "eval_runtime": 26.1035,
+      "eval_samples_per_second": 34.095,
+      "eval_steps_per_second": 4.291,
+      "eval_wer": 0.4654356060606061,
       "step": 24840
     },
     {
       "epoch": 24.0,
-      "grad_norm": 99.78678894042969,
-      "learning_rate": 0.0004002263374485597,
-      "loss": 276.272,
+      "grad_norm": 124.80506896972656,
+      "learning_rate": 0.00040019547325102886,
+      "loss": 146.7201,
       "step": 25920
     },
     {
       "epoch": 24.0,
-      "eval_cer": 0.22733784518105024,
-      "eval_loss": 447.37799072265625,
-      "eval_runtime": 26.3157,
-      "eval_samples_per_second": 33.82,
-      "eval_steps_per_second": 4.256,
-      "eval_wer": 0.6107954545454546,
+      "eval_cer": 0.1692683890727659,
+      "eval_loss": 311.2599792480469,
+      "eval_runtime": 26.2596,
+      "eval_samples_per_second": 33.892,
+      "eval_steps_per_second": 4.265,
+      "eval_wer": 0.4704861111111111,
       "step": 25920
     },
     {
       "epoch": 25.0,
-      "grad_norm": 39.30922317504883,
-      "learning_rate": 0.0003891255144032922,
-      "loss": 267.6378,
+      "grad_norm": 165.0308837890625,
+      "learning_rate": 0.0003891049382716049,
+      "loss": 139.5994,
       "step": 27000
     },
     {
       "epoch": 25.0,
-      "eval_cer": 0.24299757941016648,
-      "eval_loss": 412.04449462890625,
-      "eval_runtime": 26.2307,
-      "eval_samples_per_second": 33.93,
-      "eval_steps_per_second": 4.27,
-      "eval_wer": 0.6068497474747475,
+      "eval_cer": 0.17075038284839203,
+      "eval_loss": 303.4479064941406,
+      "eval_runtime": 26.4699,
+      "eval_samples_per_second": 33.623,
+      "eval_steps_per_second": 4.231,
+      "eval_wer": 0.46985479797979796,
       "step": 27000
     },
     {
       "epoch": 26.0,
-      "grad_norm": 0.1310397982597351,
-      "learning_rate": 0.00037801440329218104,
-      "loss": 249.7945,
+      "grad_norm": 137.65919494628906,
+      "learning_rate": 0.0003779938271604939,
+      "loss": 132.5791,
       "step": 28080
     },
     {
       "epoch": 26.0,
-      "eval_cer": 0.23860099787580893,
-      "eval_loss": 411.8222351074219,
-      "eval_runtime": 26.2906,
-      "eval_samples_per_second": 33.852,
-      "eval_steps_per_second": 4.26,
-      "eval_wer": 0.6144255050505051,
+      "eval_cer": 0.17596206095934397,
+      "eval_loss": 315.4703674316406,
+      "eval_runtime": 26.306,
+      "eval_samples_per_second": 33.833,
+      "eval_steps_per_second": 4.258,
+      "eval_wer": 0.47143308080808083,
       "step": 28080
     },
     {
       "epoch": 27.0,
-      "grad_norm": 26.45827293395996,
-      "learning_rate": 0.0003669135802469136,
-      "loss": 238.2421,
+      "grad_norm": 9.270563125610352,
+      "learning_rate": 0.0003668827160493827,
+      "loss": 129.8284,
       "step": 29160
     },
     {
       "epoch": 27.0,
-      "eval_cer": 0.2365015066936719,
-      "eval_loss": 403.0203552246094,
-      "eval_runtime": 26.1924,
-      "eval_samples_per_second": 33.979,
-      "eval_steps_per_second": 4.276,
-      "eval_wer": 0.6283143939393939,
+      "eval_cer": 0.18480462382057997,
+      "eval_loss": 348.47021484375,
+      "eval_runtime": 26.077,
+      "eval_samples_per_second": 34.13,
+      "eval_steps_per_second": 4.295,
+      "eval_wer": 0.49873737373737376,
       "step": 29160
     },
     {
       "epoch": 28.0,
-      "grad_norm": 3.8607170581817627,
-      "learning_rate": 0.0003558024691358025,
-      "loss": 226.9799,
+      "grad_norm": 3.142731189727783,
+      "learning_rate": 0.0003557716049382716,
+      "loss": 121.1056,
       "step": 30240
     },
     {
       "epoch": 28.0,
-      "eval_cer": 0.2248925554512671,
-      "eval_loss": 399.3008728027344,
-      "eval_runtime": 28.3266,
-      "eval_samples_per_second": 31.419,
-      "eval_steps_per_second": 3.954,
-      "eval_wer": 0.6117424242424242,
+      "eval_cer": 0.16390851158425135,
+      "eval_loss": 356.3645935058594,
+      "eval_runtime": 26.2097,
+      "eval_samples_per_second": 33.957,
+      "eval_steps_per_second": 4.273,
+      "eval_wer": 0.47632575757575757,
       "step": 30240
     },
     {
       "epoch": 29.0,
-      "grad_norm": 0.44324901700019836,
-      "learning_rate": 0.0003446913580246914,
-      "loss": 209.7066,
+      "grad_norm": 1.4301165342330933,
+      "learning_rate": 0.00034466049382716053,
+      "loss": 111.3771,
       "step": 31320
     },
     {
       "epoch": 29.0,
-      "eval_cer": 0.22968433532579163,
-      "eval_loss": 418.8338623046875,
-      "eval_runtime": 26.3962,
-      "eval_samples_per_second": 33.717,
-      "eval_steps_per_second": 4.243,
-      "eval_wer": 0.5924873737373737,
+      "eval_cer": 0.16904609000642198,
+      "eval_loss": 367.5716552734375,
+      "eval_runtime": 26.5107,
+      "eval_samples_per_second": 33.571,
+      "eval_steps_per_second": 4.225,
+      "eval_wer": 0.4652777777777778,
       "step": 31320
     },
     {
       "epoch": 30.0,
-      "grad_norm": 52.79179000854492,
-      "learning_rate": 0.00033359053497942386,
-      "loss": 200.3202,
+      "grad_norm": 45.340206146240234,
+      "learning_rate": 0.000333559670781893,
+      "loss": 108.3256,
       "step": 32400
     },
     {
       "epoch": 30.0,
-      "eval_cer": 0.2221261670700983,
-      "eval_loss": 411.5075988769531,
-      "eval_runtime": 26.3529,
-      "eval_samples_per_second": 33.772,
-      "eval_steps_per_second": 4.25,
-      "eval_wer": 0.6036931818181818,
+      "eval_cer": 0.16506940670849182,
+      "eval_loss": 318.64056396484375,
+      "eval_runtime": 27.5742,
+      "eval_samples_per_second": 32.277,
+      "eval_steps_per_second": 4.062,
+      "eval_wer": 0.46085858585858586,
       "step": 32400
     },
     {
       "epoch": 31.0,
-      "grad_norm": 19.727815628051758,
-      "learning_rate": 0.0003224897119341564,
-      "loss": 193.5312,
+      "grad_norm": 33.360191345214844,
+      "learning_rate": 0.0003224485596707819,
+      "loss": 105.0885,
       "step": 33480
     },
     {
       "epoch": 31.0,
-      "eval_cer": 0.21987847651039866,
-      "eval_loss": 421.6947937011719,
-      "eval_runtime": 26.3853,
-      "eval_samples_per_second": 33.731,
-      "eval_steps_per_second": 4.245,
-      "eval_wer": 0.5924873737373737,
+      "eval_cer": 0.16798399446722323,
+      "eval_loss": 389.4572448730469,
+      "eval_runtime": 27.5484,
+      "eval_samples_per_second": 32.307,
+      "eval_steps_per_second": 4.066,
+      "eval_wer": 0.4619633838383838,
       "step": 33480
     },
     {
       "epoch": 32.0,
-      "grad_norm": 11.547778129577637,
-      "learning_rate": 0.00031137860082304525,
-      "loss": 186.5759,
+      "grad_norm": 25.099489212036133,
+      "learning_rate": 0.00031135802469135803,
+      "loss": 103.1692,
       "step": 34560
     },
     {
       "epoch": 32.0,
-      "eval_cer": 0.22017487526552387,
-      "eval_loss": 403.4676818847656,
-      "eval_runtime": 26.3332,
-      "eval_samples_per_second": 33.798,
-      "eval_steps_per_second": 4.253,
-      "eval_wer": 0.5838068181818182,
+      "eval_cer": 0.1633898137627822,
+      "eval_loss": 337.2361755371094,
+      "eval_runtime": 26.0544,
+      "eval_samples_per_second": 34.159,
+      "eval_steps_per_second": 4.299,
+      "eval_wer": 0.47001262626262624,
       "step": 34560
     },
     {
       "epoch": 33.0,
-      "grad_norm": 26.520191192626953,
-      "learning_rate": 0.0003002674897119342,
-      "loss": 176.4434,
+      "grad_norm": 74.40531158447266,
+      "learning_rate": 0.00030024691358024693,
+      "loss": 95.9316,
       "step": 35640
     },
     {
       "epoch": 33.0,
-      "eval_cer": 0.22563355233908017,
-      "eval_loss": 449.66937255859375,
-      "eval_runtime": 28.5656,
-      "eval_samples_per_second": 31.156,
-      "eval_steps_per_second": 3.921,
-      "eval_wer": 0.5984848484848485,
+      "eval_cer": 0.1638591117917305,
+      "eval_loss": 366.3456115722656,
+      "eval_runtime": 26.1478,
+      "eval_samples_per_second": 34.037,
+      "eval_steps_per_second": 4.283,
+      "eval_wer": 0.46291035353535354,
       "step": 35640
     },
     {
       "epoch": 34.0,
-      "grad_norm": 44.24443435668945,
-      "learning_rate": 0.00028915637860082305,
-      "loss": 164.8203,
+      "grad_norm": 0.38387709856033325,
+      "learning_rate": 0.0002891358024691358,
+      "loss": 91.8423,
       "step": 36720
     },
     {
       "epoch": 34.0,
-      "eval_cer": 0.2237810601195475,
-      "eval_loss": 373.5114440917969,
-      "eval_runtime": 26.1741,
-      "eval_samples_per_second": 34.003,
-      "eval_steps_per_second": 4.279,
-      "eval_wer": 0.5935921717171717,
+      "eval_cer": 0.16605740255890925,
+      "eval_loss": 363.75274658203125,
+      "eval_runtime": 26.4449,
+      "eval_samples_per_second": 33.655,
+      "eval_steps_per_second": 4.235,
+      "eval_wer": 0.46480429292929293,
       "step": 36720
     },
     {
       "epoch": 35.0,
-      "grad_norm": 263.10101318359375,
-      "learning_rate": 0.00027804526748971195,
-      "loss": 153.8149,
+      "grad_norm": 237.123046875,
+      "learning_rate": 0.00027802469135802474,
+      "loss": 86.3658,
       "step": 37800
     },
     {
       "epoch": 35.0,
-      "eval_cer": 0.21960677765153386,
-      "eval_loss": 387.5101013183594,
-      "eval_runtime": 25.9639,
-      "eval_samples_per_second": 34.278,
-      "eval_steps_per_second": 4.314,
-      "eval_wer": 0.5713383838383839,
+      "eval_cer": 0.1668477992392432,
+      "eval_loss": 356.63299560546875,
+      "eval_runtime": 26.8886,
+      "eval_samples_per_second": 33.1,
+      "eval_steps_per_second": 4.165,
+      "eval_wer": 0.4594381313131313,
       "step": 37800
     },
     {
       "epoch": 36.0,
-      "grad_norm": 24.568132400512695,
-      "learning_rate": 0.00026694444444444443,
-      "loss": 148.2955,
+      "grad_norm": 25.518943786621094,
+      "learning_rate": 0.0002669135802469136,
+      "loss": 82.4053,
       "step": 38880
     },
     {
       "epoch": 36.0,
-      "eval_cer": 0.22192856790001483,
-      "eval_loss": 422.5177307128906,
-      "eval_runtime": 25.9731,
-      "eval_samples_per_second": 34.266,
-      "eval_steps_per_second": 4.312,
-      "eval_wer": 0.58364898989899,
+      "eval_cer": 0.16729239737193105,
+      "eval_loss": 356.5137939453125,
+      "eval_runtime": 31.8004,
+      "eval_samples_per_second": 27.987,
+      "eval_steps_per_second": 3.522,
+      "eval_wer": 0.45233585858585856,
       "step": 38880
     },
     {
       "epoch": 37.0,
-      "grad_norm": 10.271224975585938,
-      "learning_rate": 0.00025583333333333334,
-      "loss": 139.206,
+      "grad_norm": 0.05784684047102928,
+      "learning_rate": 0.0002558127572016461,
+      "loss": 80.7273,
       "step": 39960
     },
     {
       "epoch": 37.0,
-      "eval_cer": 0.22331176209059922,
-      "eval_loss": 424.9959411621094,
-      "eval_runtime": 25.8486,
-      "eval_samples_per_second": 34.431,
-      "eval_steps_per_second": 4.333,
-      "eval_wer": 0.58364898989899,
+      "eval_cer": 0.15795583658548634,
+      "eval_loss": 363.0180358886719,
+      "eval_runtime": 26.4947,
+      "eval_samples_per_second": 33.592,
+      "eval_steps_per_second": 4.227,
+      "eval_wer": 0.448074494949495,
       "step": 39960
     },
     {
       "epoch": 38.0,
-      "grad_norm": 0.18810251355171204,
-      "learning_rate": 0.00024472222222222224,
-      "loss": 137.3851,
+      "grad_norm": 0.5493535399436951,
+      "learning_rate": 0.000244701646090535,
+      "loss": 77.05,
       "step": 41040
     },
     {
       "epoch": 38.0,
-      "eval_cer": 0.21884108086746035,
-      "eval_loss": 428.62481689453125,
-      "eval_runtime": 29.4353,
-      "eval_samples_per_second": 30.236,
-      "eval_steps_per_second": 3.805,
-      "eval_wer": 0.5863320707070707,
+      "eval_cer": 0.15891913253964332,
+      "eval_loss": 378.66741943359375,
+      "eval_runtime": 26.0466,
+      "eval_samples_per_second": 34.17,
+      "eval_steps_per_second": 4.3,
+      "eval_wer": 0.4513888888888889,
       "step": 41040
     },
     {
       "epoch": 39.0,
-      "grad_norm": 126.07501220703125,
-      "learning_rate": 0.00023362139917695475,
-      "loss": 128.706,
+      "grad_norm": 349.0174255371094,
+      "learning_rate": 0.0002336008230452675,
+      "loss": 73.1437,
       "step": 42120
     },
     {
       "epoch": 39.0,
-      "eval_cer": 0.21730968729931335,
-      "eval_loss": 420.81134033203125,
-      "eval_runtime": 27.0888,
-      "eval_samples_per_second": 32.855,
-      "eval_steps_per_second": 4.135,
-      "eval_wer": 0.5759154040404041,
+      "eval_cer": 0.1606234253816134,
+      "eval_loss": 391.6517639160156,
+      "eval_runtime": 26.0262,
+      "eval_samples_per_second": 34.196,
+      "eval_steps_per_second": 4.303,
+      "eval_wer": 0.4521780303030303,
       "step": 42120
     },
     {
       "epoch": 40.0,
-      "grad_norm": 3.7092888355255127,
-      "learning_rate": 0.00022251028806584362,
-      "loss": 122.1991,
+      "grad_norm": 30.239347457885742,
+      "learning_rate": 0.00022248971193415638,
+      "loss": 70.8281,
       "step": 43200
     },
     {
       "epoch": 40.0,
-      "eval_cer": 0.2217803685224522,
-      "eval_loss": 449.3416748046875,
-      "eval_runtime": 28.1156,
-      "eval_samples_per_second": 31.655,
-      "eval_steps_per_second": 3.984,
-      "eval_wer": 0.579229797979798,
+      "eval_cer": 0.166156202143951,
+      "eval_loss": 375.7931213378906,
+      "eval_runtime": 27.0432,
+      "eval_samples_per_second": 32.91,
+      "eval_steps_per_second": 4.142,
+      "eval_wer": 0.4569128787878788,
       "step": 43200
     },
     {
       "epoch": 41.0,
-      "grad_norm": 53.470237731933594,
-      "learning_rate": 0.00021140946502057616,
-      "loss": 114.7006,
+      "grad_norm": 22.0565185546875,
+      "learning_rate": 0.00021137860082304528,
+      "loss": 66.8237,
       "step": 44280
     },
     {
       "epoch": 41.0,
-      "eval_cer": 0.22318826260929703,
-      "eval_loss": 472.12640380859375,
-      "eval_runtime": 27.1058,
-      "eval_samples_per_second": 32.834,
-      "eval_steps_per_second": 4.132,
-      "eval_wer": 0.5743371212121212,
+      "eval_cer": 0.15859803388825766,
+      "eval_loss": 365.82177734375,
+      "eval_runtime": 26.6045,
+      "eval_samples_per_second": 33.453,
+      "eval_steps_per_second": 4.21,
+      "eval_wer": 0.4428661616161616,
       "step": 44280
     },
     {
       "epoch": 42.0,
-      "grad_norm": 6.227956295013428,
-      "learning_rate": 0.00020030864197530864,
-      "loss": 112.9208,
+      "grad_norm": 38.3385124206543,
+      "learning_rate": 0.00020028806584362143,
+      "loss": 64.3243,
       "step": 45360
     },
     {
       "epoch": 42.0,
-      "eval_cer": 0.21343180358642494,
-      "eval_loss": 484.6367492675781,
-      "eval_runtime": 27.3265,
-      "eval_samples_per_second": 32.569,
-      "eval_steps_per_second": 4.099,
-      "eval_wer": 0.5869633838383839,
+      "eval_cer": 0.16264881687496913,
+      "eval_loss": 400.6961364746094,
+      "eval_runtime": 27.0273,
+      "eval_samples_per_second": 32.93,
+      "eval_steps_per_second": 4.144,
+      "eval_wer": 0.4725378787878788,
       "step": 45360
     },
     {
       "epoch": 43.0,
-      "grad_norm": 45.22465896606445,
-      "learning_rate": 0.00018919753086419752,
-      "loss": 104.3684,
+      "grad_norm": 0.24667246639728546,
+      "learning_rate": 0.00018917695473251027,
+      "loss": 59.5604,
       "step": 46440
     },
     {
       "epoch": 43.0,
-      "eval_cer": 0.20782492713530604,
-      "eval_loss": 475.1777648925781,
-      "eval_runtime": 30.3439,
-      "eval_samples_per_second": 29.33,
-      "eval_steps_per_second": 3.691,
-      "eval_wer": 0.5617108585858586,
+      "eval_cer": 0.1658104035963049,
+      "eval_loss": 417.8609619140625,
+      "eval_runtime": 26.536,
+      "eval_samples_per_second": 33.539,
+      "eval_steps_per_second": 4.221,
+      "eval_wer": 0.45391414141414144,
       "step": 46440
     },
     {
       "epoch": 44.0,
-      "grad_norm": 4.431763648986816,
-      "learning_rate": 0.00017808641975308642,
-      "loss": 101.3417,
+      "grad_norm": 0.03135277330875397,
+      "learning_rate": 0.00017806584362139918,
+      "loss": 57.4405,
       "step": 47520
     },
     {
       "epoch": 44.0,
-      "eval_cer": 0.2081460257866917,
-      "eval_loss": 438.15020751953125,
-      "eval_runtime": 34.0126,
-      "eval_samples_per_second": 26.167,
-      "eval_steps_per_second": 3.293,
-      "eval_wer": 0.5639204545454546,
+      "eval_cer": 0.16067282517413428,
+      "eval_loss": 377.4517822265625,
+      "eval_runtime": 26.3646,
+      "eval_samples_per_second": 33.757,
+      "eval_steps_per_second": 4.248,
+      "eval_wer": 0.4474431818181818,
       "step": 47520
     },
     {
       "epoch": 45.0,
-      "grad_norm": 27.047523498535156,
-      "learning_rate": 0.0001669753086419753,
-      "loss": 93.9768,
+      "grad_norm": 3.01446533203125,
+      "learning_rate": 0.00016695473251028805,
+      "loss": 54.1485,
       "step": 48600
     },
     {
       "epoch": 45.0,
-      "eval_cer": 0.20668873190732598,
-      "eval_loss": 450.406494140625,
-      "eval_runtime": 26.339,
-      "eval_samples_per_second": 33.79,
-      "eval_steps_per_second": 4.252,
-      "eval_wer": 0.5582386363636364,
+      "eval_cer": 0.16040112631526948,
+      "eval_loss": 375.0706787109375,
+      "eval_runtime": 26.0727,
+      "eval_samples_per_second": 34.135,
+      "eval_steps_per_second": 4.296,
+      "eval_wer": 0.4464962121212121,
       "step": 48600
     },
     {
       "epoch": 46.0,
-      "grad_norm": 2.549964666366577,
-      "learning_rate": 0.00015587448559670783,
-      "loss": 95.7108,
+      "grad_norm": 0.4665185213088989,
+      "learning_rate": 0.00015585390946502056,
+      "loss": 54.5365,
       "step": 49680
     },
     {
       "epoch": 46.0,
-      "eval_cer": 0.2114064120930692,
-      "eval_loss": 471.1705627441406,
-      "eval_runtime": 26.695,
-      "eval_samples_per_second": 33.34,
-      "eval_steps_per_second": 4.196,
-      "eval_wer": 0.5645517676767676,
+      "eval_cer": 0.15983302870127947,
+      "eval_loss": 376.6304016113281,
+      "eval_runtime": 27.0787,
+      "eval_samples_per_second": 32.867,
+      "eval_steps_per_second": 4.136,
+      "eval_wer": 0.4378156565656566,
       "step": 49680
     },
     {
       "epoch": 47.0,
-      "grad_norm": 37.24876022338867,
-      "learning_rate": 0.0001447633744855967,
-      "loss": 85.0469,
+      "grad_norm": 139.39511108398438,
+      "learning_rate": 0.00014474279835390946,
+      "loss": 49.1435,
       "step": 50760
     },
     {
       "epoch": 47.0,
-      "eval_cer": 0.20777552734278515,
-      "eval_loss": 466.9778747558594,
-      "eval_runtime": 27.1859,
-      "eval_samples_per_second": 32.738,
-      "eval_steps_per_second": 4.12,
-      "eval_wer": 0.561395202020202,
+      "eval_cer": 0.15711604011263153,
+      "eval_loss": 412.8172302246094,
+      "eval_runtime": 26.891,
+      "eval_samples_per_second": 33.097,
+      "eval_steps_per_second": 4.165,
+      "eval_wer": 0.4378156565656566,
       "step": 50760
     },
     {
       "epoch": 48.0,
-      "grad_norm": 0.1757580041885376,
-      "learning_rate": 0.0001336625514403292,
-      "loss": 86.9859,
+      "grad_norm": 0.0027996094431728125,
+      "learning_rate": 0.00013364197530864197,
+      "loss": 49.8476,
       "step": 51840
     },
     {
       "epoch": 48.0,
-      "eval_cer": 0.20693573086993033,
-      "eval_loss": 457.32208251953125,
-      "eval_runtime": 26.898,
-      "eval_samples_per_second": 33.088,
-      "eval_steps_per_second": 4.164,
-      "eval_wer": 0.5639204545454546,
+      "eval_cer": 0.15741243886775674,
+      "eval_loss": 402.0115051269531,
+      "eval_runtime": 26.386,
+      "eval_samples_per_second": 33.73,
+      "eval_steps_per_second": 4.245,
+      "eval_wer": 0.44365530303030304,
       "step": 51840
     },
     {
       "epoch": 49.0,
-      "grad_norm": 33.84082794189453,
-      "learning_rate": 0.00012255144032921811,
-      "loss": 82.6203,
+      "grad_norm": 85.25370788574219,
+      "learning_rate": 0.00012253086419753087,
+      "loss": 48.5295,
       "step": 52920
     },
     {
       "epoch": 49.0,
-      "eval_cer": 0.2059971348120338,
-      "eval_loss": 467.1033630371094,
-      "eval_runtime": 26.7011,
-      "eval_samples_per_second": 33.332,
-      "eval_steps_per_second": 4.195,
-      "eval_wer": 0.5558712121212122,
+      "eval_cer": 0.15422615225016054,
+      "eval_loss": 400.1714172363281,
+      "eval_runtime": 27.8535,
+      "eval_samples_per_second": 31.953,
+      "eval_steps_per_second": 4.021,
+      "eval_wer": 0.43166035353535354,
       "step": 52920
     },
     {
       "epoch": 50.0,
-      "grad_norm": 0.03401001915335655,
-      "learning_rate": 0.000111440329218107,
-      "loss": 76.6465,
+      "grad_norm": 0.001213193405419588,
+      "learning_rate": 0.00011141975308641976,
+      "loss": 45.4287,
       "step": 54000
     },
     {
       "epoch": 50.0,
-      "eval_cer": 0.2074544286913995,
-      "eval_loss": 488.0355224609375,
-      "eval_runtime": 25.9988,
-      "eval_samples_per_second": 34.232,
-      "eval_steps_per_second": 4.308,
-      "eval_wer": 0.5606060606060606,
+      "eval_cer": 0.1509410660475226,
+      "eval_loss": 395.4508361816406,
+      "eval_runtime": 26.6376,
+      "eval_samples_per_second": 33.411,
+      "eval_steps_per_second": 4.205,
+      "eval_wer": 0.4307133838383838,
       "step": 54000
     },
     {
       "epoch": 51.0,
-      "grad_norm": 0.004368220455944538,
-      "learning_rate": 0.00010032921810699588,
-      "loss": 72.2897,
+      "grad_norm": 0.1676294207572937,
+      "learning_rate": 0.00010030864197530865,
+      "loss": 39.8247,
       "step": 55080
     },
     {
       "epoch": 51.0,
-      "eval_cer": 0.20172405275897842,
-      "eval_loss": 506.8006286621094,
-      "eval_runtime": 26.3849,
-      "eval_samples_per_second": 33.731,
-      "eval_steps_per_second": 4.245,
-      "eval_wer": 0.5528724747474747,
+      "eval_cer": 0.1501753692634491,
+      "eval_loss": 419.62457275390625,
+      "eval_runtime": 26.461,
+      "eval_samples_per_second": 33.634,
+      "eval_steps_per_second": 4.233,
+      "eval_wer": 0.4255050505050505,
       "step": 55080
     },
     {
       "epoch": 52.0,
-      "grad_norm": 0.492847204208374,
-      "learning_rate": 8.921810699588477e-05,
-      "loss": 66.3217,
+      "grad_norm": 0.34273698925971985,
+      "learning_rate": 8.920781893004115e-05,
+      "loss": 38.7,
       "step": 56160
     },
     {
       "epoch": 52.0,
-      "eval_cer": 0.20658993232228426,
-      "eval_loss": 513.1781616210938,
-      "eval_runtime": 26.4834,
-      "eval_samples_per_second": 33.606,
-      "eval_steps_per_second": 4.229,
-      "eval_wer": 0.5503472222222222,
+      "eval_cer": 0.1519043620016796,
+      "eval_loss": 429.1017150878906,
+      "eval_runtime": 26.7845,
+      "eval_samples_per_second": 33.228,
+      "eval_steps_per_second": 4.182,
+      "eval_wer": 0.429135101010101,
       "step": 56160
     },
     {
       "epoch": 53.0,
-      "grad_norm": 216.27200317382812,
+      "grad_norm": 138.7336883544922,
       "learning_rate": 7.811728395061729e-05,
-      "loss": 63.3613,
+      "loss": 37.6587,
       "step": 57240
     },
     {
       "epoch": 53.0,
-      "eval_cer": 0.20547843699056464,
-      "eval_loss": 509.8887023925781,
-      "eval_runtime": 26.8112,
-      "eval_samples_per_second": 33.195,
-      "eval_steps_per_second": 4.177,
-      "eval_wer": 0.5571338383838383,
+      "eval_cer": 0.15027416884849085,
+      "eval_loss": 404.33428955078125,
+      "eval_runtime": 26.9634,
+      "eval_samples_per_second": 33.008,
+      "eval_steps_per_second": 4.154,
+      "eval_wer": 0.429135101010101,
       "step": 57240
     },
     {
       "epoch": 54.0,
-      "grad_norm": 82.88473510742188,
+      "grad_norm": 19.26589584350586,
       "learning_rate": 6.700617283950618e-05,
-      "loss": 63.2432,
+      "loss": 37.8115,
       "step": 58320
     },
     {
       "epoch": 54.0,
-      "eval_cer": 0.20933162080719261,
-      "eval_loss": 532.0458374023438,
-      "eval_runtime": 27.9984,
-      "eval_samples_per_second": 31.788,
-      "eval_steps_per_second": 4.0,
-      "eval_wer": 0.5602904040404041,
+      "eval_cer": 0.1501012695746678,
+      "eval_loss": 420.6966857910156,
+      "eval_runtime": 27.5833,
+      "eval_samples_per_second": 32.266,
+      "eval_steps_per_second": 4.06,
+      "eval_wer": 0.42613636363636365,
       "step": 58320
     },
     {
       "epoch": 55.0,
-      "grad_norm": 167.6044464111328,
+      "grad_norm": 0.544121265411377,
       "learning_rate": 5.5895061728395066e-05,
-      "loss": 58.8891,
+      "loss": 33.9205,
       "step": 59400
     },
     {
       "epoch": 55.0,
-      "eval_cer": 0.20617003408585685,
-      "eval_loss": 529.6932983398438,
-      "eval_runtime": 26.5553,
-      "eval_samples_per_second": 33.515,
-      "eval_steps_per_second": 4.218,
-      "eval_wer": 0.5525568181818182,
+      "eval_cer": 0.1468655831645507,
+      "eval_loss": 433.5399475097656,
+      "eval_runtime": 26.3217,
+      "eval_samples_per_second": 33.812,
+      "eval_steps_per_second": 4.255,
+      "eval_wer": 0.4161931818181818,
       "step": 59400
     },
     {
       "epoch": 56.0,
-      "grad_norm": 0.0747949555516243,
-      "learning_rate": 4.4794238683127575e-05,
-      "loss": 64.8119,
+      "grad_norm": 2.6075133064296097e-05,
+      "learning_rate": 4.478395061728395e-05,
+      "loss": 35.9225,
       "step": 60480
     },
     {
       "epoch": 56.0,
-      "eval_cer": 0.2025391493355728,
-      "eval_loss": 530.6396484375,
-      "eval_runtime": 26.4411,
-      "eval_samples_per_second": 33.66,
-      "eval_steps_per_second": 4.236,
-      "eval_wer": 0.5465593434343434,
+      "eval_cer": 0.14973077113076125,
+      "eval_loss": 435.69970703125,
+      "eval_runtime": 26.2399,
+      "eval_samples_per_second": 33.918,
+      "eval_steps_per_second": 4.268,
+      "eval_wer": 0.4253472222222222,
       "step": 60480
     },
     {
       "epoch": 57.0,
-      "grad_norm": 111.74144744873047,
-      "learning_rate": 3.3683127572016464e-05,
-      "loss": 53.2317,
+      "grad_norm": 0.002031653420999646,
+      "learning_rate": 3.367283950617284e-05,
+      "loss": 30.8749,
       "step": 61560
     },
     {
       "epoch": 57.0,
-      "eval_cer": 0.19937756261423703,
-      "eval_loss": 548.7178955078125,
-      "eval_runtime": 26.8929,
-      "eval_samples_per_second": 33.094,
-      "eval_steps_per_second": 4.165,
-      "eval_wer": 0.5415088383838383,
+      "eval_cer": 0.14703848243837375,
+      "eval_loss": 450.0592346191406,
+      "eval_runtime": 26.104,
+      "eval_samples_per_second": 34.094,
+      "eval_steps_per_second": 4.291,
+      "eval_wer": 0.41761363636363635,
       "step": 61560
     },
     {
       "epoch": 58.0,
-      "grad_norm": 0.0003476064885035157,
+      "grad_norm": 0.0005908387829549611,
       "learning_rate": 2.257201646090535e-05,
-      "loss": 52.8465,
+      "loss": 31.0171,
       "step": 62640
     },
     {
       "epoch": 58.0,
-      "eval_cer": 0.20071135701230056,
-      "eval_loss": 547.544677734375,
-      "eval_runtime": 26.8853,
-      "eval_samples_per_second": 33.104,
-      "eval_steps_per_second": 4.166,
-      "eval_wer": 0.5399305555555556,
+      "eval_cer": 0.1477547794299264,
+      "eval_loss": 441.6795654296875,
+      "eval_runtime": 26.4317,
+      "eval_samples_per_second": 33.672,
+      "eval_steps_per_second": 4.237,
+      "eval_wer": 0.4171401515151515,
       "step": 62640
     },
     {
       "epoch": 59.0,
-      "grad_norm": 331.07489013671875,
-      "learning_rate": 1.147119341563786e-05,
-      "loss": 50.4224,
+      "grad_norm": 0.14264369010925293,
+      "learning_rate": 1.1460905349794239e-05,
+      "loss": 30.2995,
       "step": 63720
     },
     {
       "epoch": 59.0,
-      "eval_cer": 0.20066195721977967,
-      "eval_loss": 545.9059448242188,
-      "eval_runtime": 26.3111,
-      "eval_samples_per_second": 33.826,
-      "eval_steps_per_second": 4.257,
-      "eval_wer": 0.5426136363636364,
+      "eval_cer": 0.14787827891122857,
+      "eval_loss": 445.0783386230469,
+      "eval_runtime": 31.3254,
+      "eval_samples_per_second": 28.411,
+      "eval_steps_per_second": 3.575,
+      "eval_wer": 0.4236111111111111,
       "step": 63720
     },
     {
       "epoch": 60.0,
-      "grad_norm": 0.002585458569228649,
-      "learning_rate": 3.60082304526749e-07,
-      "loss": 50.2867,
+      "grad_norm": 0.141180619597435,
+      "learning_rate": 3.497942386831276e-07,
+      "loss": 29.8519,
       "step": 64800
     },
     {
       "epoch": 60.0,
-      "eval_cer": 0.19982216074692485,
-      "eval_loss": 551.5419921875,
-      "eval_runtime": 26.3733,
-      "eval_samples_per_second": 33.746,
-      "eval_steps_per_second": 4.247,
-      "eval_wer": 0.5422979797979798,
+      "eval_cer": 0.146371585239342,
+      "eval_loss": 447.4113464355469,
+      "eval_runtime": 27.8181,
+      "eval_samples_per_second": 31.994,
+      "eval_steps_per_second": 4.026,
+      "eval_wer": 0.41887626262626265,
       "step": 64800
     },
     {
       "epoch": 60.0,
       "step": 64800,
       "total_flos": 1.8440987587856836e+20,
-      "train_loss": 306.16538580246913,
-      "train_runtime": 47401.1605,
-      "train_samples_per_second": 21.856,
-      "train_steps_per_second": 1.367
+      "train_loss": 277.69142749927664,
+      "train_runtime": 46658.0373,
+      "train_samples_per_second": 22.205,
+      "train_steps_per_second": 1.389
     }
   ],
   "logging_steps": 500,