diff --git "a/trainer_state.json" "b/trainer_state.json"
--- "a/trainer_state.json"
+++ "b/trainer_state.json"
@@ -1,2327 +1,4627 @@
 {
   "best_metric": null,
   "best_model_checkpoint": null,
-  "epoch": 3.9294274824158117,
+  "epoch": 7.858854964831624,
   "eval_steps": 1000,
-  "global_step": 100000,
+  "global_step": 200000,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
   "log_history": [
     {
       "epoch": 0.01964713741207906,
-      "grad_norm": 2.5734992027282715,
+      "grad_norm": 3.3487088680267334,
       "learning_rate": 0.0002982,
-      "loss": 4.7324,
+      "loss": 4.7396,
       "step": 500
     },
     {
       "epoch": 0.03929427482415812,
-      "grad_norm": 3.079489231109619,
-      "learning_rate": 0.0002985015075376884,
-      "loss": 1.043,
+      "grad_norm": 4.3209547996521,
+      "learning_rate": 0.00029925263157894736,
+      "loss": 1.092,
       "step": 1000
     },
     {
       "epoch": 0.03929427482415812,
-      "eval_loss": 1.1073596477508545,
-      "eval_runtime": 144.3869,
-      "eval_samples_per_second": 39.173,
-      "eval_steps_per_second": 4.897,
-      "eval_wer": 0.7791240711912825,
+      "eval_loss": 1.1751947402954102,
+      "eval_runtime": 144.9265,
+      "eval_samples_per_second": 39.027,
+      "eval_steps_per_second": 4.878,
+      "eval_wer": 0.8087496589687214,
       "step": 1000
     },
     {
       "epoch": 0.05894141223623718,
-      "grad_norm": 2.484443187713623,
-      "learning_rate": 0.0002969939698492462,
-      "loss": 0.8123,
+      "grad_norm": 2.453352451324463,
+      "learning_rate": 0.0002985007518796992,
+      "loss": 0.8198,
       "step": 1500
     },
     {
       "epoch": 0.07858854964831624,
-      "grad_norm": 2.6904137134552,
-      "learning_rate": 0.00029548643216080397,
-      "loss": 0.716,
+      "grad_norm": 4.910336494445801,
+      "learning_rate": 0.0002977488721804511,
+      "loss": 0.7307,
       "step": 2000
     },
     {
       "epoch": 0.07858854964831624,
-      "eval_loss": 0.9458802342414856,
-      "eval_runtime": 143.0757,
-      "eval_samples_per_second": 39.532,
-      "eval_steps_per_second": 4.941,
-      "eval_wer": 0.7145287348943204,
+      "eval_loss": 0.9363270998001099,
+      "eval_runtime": 143.2813,
+      "eval_samples_per_second": 39.475,
+      "eval_steps_per_second": 4.934,
+      "eval_wer": 0.6985123011988252,
       "step": 2000
     },
     {
       "epoch": 0.0982356870603953,
-      "grad_norm": 2.3412747383117676,
-      "learning_rate": 0.00029397889447236176,
-      "loss": 0.6822,
+      "grad_norm": 2.127075672149658,
+      "learning_rate": 0.00029699699248120296,
+      "loss": 0.6921,
       "step": 2500
     },
     {
       "epoch": 0.11788282447247436,
-      "grad_norm": 2.9600541591644287,
-      "learning_rate": 0.00029247135678391956,
-      "loss": 0.6667,
+      "grad_norm": 2.57724928855896,
+      "learning_rate": 0.0002962451127819549,
+      "loss": 0.6633,
       "step": 3000
     },
     {
       "epoch": 0.11788282447247436,
-      "eval_loss": 0.8828235268592834,
-      "eval_runtime": 143.1736,
-      "eval_samples_per_second": 39.504,
-      "eval_steps_per_second": 4.938,
-      "eval_wer": 0.6616006804577041,
+      "eval_loss": 0.8888041377067566,
+      "eval_runtime": 144.2307,
+      "eval_samples_per_second": 39.215,
+      "eval_steps_per_second": 4.902,
+      "eval_wer": 0.6624352040570686,
       "step": 3000
     },
     {
       "epoch": 0.1375299618845534,
-      "grad_norm": 2.9464612007141113,
-      "learning_rate": 0.00029096381909547736,
-      "loss": 0.6284,
+      "grad_norm": 3.0028977394104004,
+      "learning_rate": 0.00029549323308270675,
+      "loss": 0.6395,
       "step": 3500
     },
     {
       "epoch": 0.15717709929663248,
-      "grad_norm": 1.9161436557769775,
-      "learning_rate": 0.00028945628140703515,
-      "loss": 0.6238,
+      "grad_norm": 1.864983320236206,
+      "learning_rate": 0.0002947413533834586,
+      "loss": 0.6273,
       "step": 4000
     },
     {
       "epoch": 0.15717709929663248,
-      "eval_loss": 0.8481988906860352,
-      "eval_runtime": 144.1673,
-      "eval_samples_per_second": 39.232,
-      "eval_steps_per_second": 4.904,
-      "eval_wer": 0.6633178732487041,
+      "eval_loss": 0.8302789330482483,
+      "eval_runtime": 143.3729,
+      "eval_samples_per_second": 39.45,
+      "eval_steps_per_second": 4.931,
+      "eval_wer": 0.640432668389209,
       "step": 4000
     },
     {
       "epoch": 0.17682423670871153,
-      "grad_norm": 2.5864381790161133,
-      "learning_rate": 0.0002879517587939698,
-      "loss": 0.6139,
+      "grad_norm": 1.865403175354004,
+      "learning_rate": 0.000293990977443609,
+      "loss": 0.6196,
       "step": 4500
     },
     {
       "epoch": 0.1964713741207906,
-      "grad_norm": 5.355952739715576,
-      "learning_rate": 0.0002864442211055276,
-      "loss": 0.591,
+      "grad_norm": 2.015836000442505,
+      "learning_rate": 0.0002932390977443609,
+      "loss": 0.6031,
       "step": 5000
     },
     {
       "epoch": 0.1964713741207906,
-      "eval_loss": 0.8120559453964233,
-      "eval_runtime": 144.0897,
-      "eval_samples_per_second": 39.253,
-      "eval_steps_per_second": 4.907,
-      "eval_wer": 0.6214954020959381,
+      "eval_loss": 0.7927883267402649,
+      "eval_runtime": 142.89,
+      "eval_samples_per_second": 39.583,
+      "eval_steps_per_second": 4.948,
+      "eval_wer": 0.6134229911251625,
       "step": 5000
     },
     {
       "epoch": 0.21611851153286965,
-      "grad_norm": 2.9285130500793457,
-      "learning_rate": 0.0002849366834170854,
-      "loss": 0.5983,
+      "grad_norm": 2.023449420928955,
+      "learning_rate": 0.00029248721804511275,
+      "loss": 0.6089,
       "step": 5500
     },
     {
       "epoch": 0.23576564894494872,
-      "grad_norm": 2.175675630569458,
-      "learning_rate": 0.00028342914572864316,
-      "loss": 0.588,
+      "grad_norm": 2.1012802124023438,
+      "learning_rate": 0.00029173533834586467,
+      "loss": 0.6007,
       "step": 6000
     },
     {
       "epoch": 0.23576564894494872,
-      "eval_loss": 0.7926327586174011,
-      "eval_runtime": 144.2057,
-      "eval_samples_per_second": 39.222,
-      "eval_steps_per_second": 4.903,
-      "eval_wer": 0.5952079087159571,
+      "eval_loss": 0.771953821182251,
+      "eval_runtime": 144.3238,
+      "eval_samples_per_second": 39.19,
+      "eval_steps_per_second": 4.899,
+      "eval_wer": 0.5832357047712282,
       "step": 6000
     },
     {
       "epoch": 0.2554127863570278,
-      "grad_norm": 1.7357851266860962,
-      "learning_rate": 0.0002819246231155779,
-      "loss": 0.5695,
+      "grad_norm": 1.6933408975601196,
+      "learning_rate": 0.0002909834586466165,
+      "loss": 0.5693,
       "step": 6500
     },
     {
       "epoch": 0.2750599237691068,
-      "grad_norm": 3.1059882640838623,
-      "learning_rate": 0.00028041708542713567,
-      "loss": 0.5732,
+      "grad_norm": 1.8497469425201416,
+      "learning_rate": 0.0002902315789473684,
+      "loss": 0.5739,
       "step": 7000
     },
     {
       "epoch": 0.2750599237691068,
-      "eval_loss": 0.7581684589385986,
-      "eval_runtime": 144.8221,
-      "eval_samples_per_second": 39.055,
-      "eval_steps_per_second": 4.882,
-      "eval_wer": 0.570685753719247,
+      "eval_loss": 0.7533236145973206,
+      "eval_runtime": 144.3945,
+      "eval_samples_per_second": 39.17,
+      "eval_steps_per_second": 4.896,
+      "eval_wer": 0.5685673476593218,
       "step": 7000
     },
     {
       "epoch": 0.2947070611811859,
-      "grad_norm": 2.637263059616089,
-      "learning_rate": 0.0002789095477386934,
-      "loss": 0.571,
+      "grad_norm": 2.9493396282196045,
+      "learning_rate": 0.0002894796992481203,
+      "loss": 0.57,
       "step": 7500
     },
     {
       "epoch": 0.31435419859326497,
-      "grad_norm": 10.5114107131958,
-      "learning_rate": 0.00027740201005025127,
-      "loss": 0.5557,
+      "grad_norm": 3.0881130695343018,
+      "learning_rate": 0.00028872781954887214,
+      "loss": 0.5655,
       "step": 8000
     },
     {
       "epoch": 0.31435419859326497,
-      "eval_loss": 0.759090781211853,
-      "eval_runtime": 144.7625,
-      "eval_samples_per_second": 39.071,
-      "eval_steps_per_second": 4.884,
-      "eval_wer": 0.5600295292965929,
+      "eval_loss": 0.7523130178451538,
+      "eval_runtime": 143.9033,
+      "eval_samples_per_second": 39.304,
+      "eval_steps_per_second": 4.913,
+      "eval_wer": 0.5594517821893406,
       "step": 8000
     },
     {
       "epoch": 0.33400133600534404,
-      "grad_norm": 3.085407257080078,
-      "learning_rate": 0.000275894472361809,
-      "loss": 0.5434,
+      "grad_norm": 2.8726112842559814,
+      "learning_rate": 0.000287975939849624,
+      "loss": 0.5541,
       "step": 8500
     },
     {
       "epoch": 0.35364847341742306,
-      "grad_norm": 2.6379830837249756,
-      "learning_rate": 0.0002743899497487437,
-      "loss": 0.5587,
+      "grad_norm": 2.6348774433135986,
+      "learning_rate": 0.0002872255639097744,
+      "loss": 0.5584,
       "step": 9000
     },
     {
       "epoch": 0.35364847341742306,
-      "eval_loss": 0.724540650844574,
-      "eval_runtime": 145.5811,
-      "eval_samples_per_second": 38.851,
-      "eval_steps_per_second": 4.856,
-      "eval_wer": 0.574120139301247,
+      "eval_loss": 0.7173847556114197,
+      "eval_runtime": 144.396,
+      "eval_samples_per_second": 39.17,
+      "eval_steps_per_second": 4.896,
+      "eval_wer": 0.5668020092760507,
       "step": 9000
     },
     {
       "epoch": 0.37329561082950213,
-      "grad_norm": 3.1712732315063477,
-      "learning_rate": 0.00027288241206030147,
-      "loss": 0.5465,
+      "grad_norm": 3.332054853439331,
+      "learning_rate": 0.00028647368421052627,
+      "loss": 0.5485,
       "step": 9500
     },
     {
       "epoch": 0.3929427482415812,
-      "grad_norm": 3.744135618209839,
-      "learning_rate": 0.00027137487437185927,
-      "loss": 0.531,
+      "grad_norm": 2.823880434036255,
+      "learning_rate": 0.0002857218045112782,
+      "loss": 0.5454,
       "step": 10000
     },
     {
       "epoch": 0.3929427482415812,
-      "eval_loss": 0.7107329964637756,
-      "eval_runtime": 145.1162,
-      "eval_samples_per_second": 38.976,
-      "eval_steps_per_second": 4.872,
-      "eval_wer": 0.5469499767296304,
+      "eval_loss": 0.7537470459938049,
+      "eval_runtime": 144.0918,
+      "eval_samples_per_second": 39.253,
+      "eval_steps_per_second": 4.907,
+      "eval_wer": 0.5798815618430133,
       "step": 10000
     },
     {
       "epoch": 0.4125898856536603,
-      "grad_norm": 2.3687188625335693,
-      "learning_rate": 0.00026986733668341706,
-      "loss": 0.5346,
+      "grad_norm": 2.8141980171203613,
+      "learning_rate": 0.00028496992481203006,
+      "loss": 0.5527,
       "step": 10500
     },
     {
       "epoch": 0.4322370230657393,
-      "grad_norm": 3.6938724517822266,
-      "learning_rate": 0.00026835979899497486,
-      "loss": 0.5275,
+      "grad_norm": 4.174062728881836,
+      "learning_rate": 0.0002842180451127819,
+      "loss": 0.5322,
       "step": 11000
     },
     {
       "epoch": 0.4322370230657393,
-      "eval_loss": 0.7101730704307556,
-      "eval_runtime": 144.5782,
-      "eval_samples_per_second": 39.121,
-      "eval_steps_per_second": 4.89,
-      "eval_wer": 0.5447513280159202,
+      "eval_loss": 0.7155322432518005,
+      "eval_runtime": 144.2272,
+      "eval_samples_per_second": 39.216,
+      "eval_steps_per_second": 4.902,
+      "eval_wer": 0.5613615573494246,
       "step": 11000
     },
     {
       "epoch": 0.4518841604778184,
-      "grad_norm": 2.88944935798645,
-      "learning_rate": 0.00026685226130653266,
-      "loss": 0.5258,
+      "grad_norm": 2.2155849933624268,
+      "learning_rate": 0.0002834676691729323,
+      "loss": 0.5373,
       "step": 11500
     },
     {
       "epoch": 0.47153129788989745,
-      "grad_norm": 3.0208890438079834,
-      "learning_rate": 0.00026534472361809046,
-      "loss": 0.5101,
+      "grad_norm": 2.1979432106018066,
+      "learning_rate": 0.0002827157894736842,
+      "loss": 0.5206,
       "step": 12000
     },
     {
       "epoch": 0.47153129788989745,
-      "eval_loss": 0.690454363822937,
-      "eval_runtime": 144.959,
-      "eval_samples_per_second": 39.018,
-      "eval_steps_per_second": 4.877,
-      "eval_wer": 0.5404342732422847,
+      "eval_loss": 0.7130174040794373,
+      "eval_runtime": 144.2801,
+      "eval_samples_per_second": 39.202,
+      "eval_steps_per_second": 4.9,
+      "eval_wer": 0.5746336922854712,
       "step": 12000
     },
     {
       "epoch": 0.4911784353019765,
-      "grad_norm": 2.797706127166748,
-      "learning_rate": 0.00026384020100502506,
-      "loss": 0.5052,
+      "grad_norm": 3.059553861618042,
+      "learning_rate": 0.00028196390977443605,
+      "loss": 0.5303,
       "step": 12500
     },
     {
       "epoch": 0.5108255727140556,
-      "grad_norm": 2.1760404109954834,
-      "learning_rate": 0.0002623326633165829,
-      "loss": 0.5215,
+      "grad_norm": 2.1422946453094482,
+      "learning_rate": 0.0002812120300751879,
+      "loss": 0.5304,
       "step": 13000
     },
     {
       "epoch": 0.5108255727140556,
-      "eval_loss": 0.682437002658844,
-      "eval_runtime": 144.6862,
-      "eval_samples_per_second": 39.091,
-      "eval_steps_per_second": 4.886,
-      "eval_wer": 0.5254770425767521,
+      "eval_loss": 0.6817054748535156,
+      "eval_runtime": 144.6048,
+      "eval_samples_per_second": 39.114,
+      "eval_steps_per_second": 4.889,
+      "eval_wer": 0.5390220025356679,
       "step": 13000
     },
     {
       "epoch": 0.5304727101261346,
-      "grad_norm": 2.243823289871216,
-      "learning_rate": 0.00026082512562814066,
-      "loss": 0.5048,
+      "grad_norm": 3.0140066146850586,
+      "learning_rate": 0.00028046015037593984,
+      "loss": 0.5156,
       "step": 13500
     },
     {
       "epoch": 0.5501198475382136,
-      "grad_norm": 2.9280102252960205,
-      "learning_rate": 0.00025931758793969846,
-      "loss": 0.5293,
+      "grad_norm": 2.334425449371338,
+      "learning_rate": 0.0002797082706766917,
+      "loss": 0.55,
       "step": 14000
     },
     {
       "epoch": 0.5501198475382136,
-      "eval_loss": 0.6682488322257996,
-      "eval_runtime": 144.9037,
-      "eval_samples_per_second": 39.033,
-      "eval_steps_per_second": 4.879,
-      "eval_wer": 0.5163133315144999,
+      "eval_loss": 0.6902604699134827,
+      "eval_runtime": 144.3591,
+      "eval_samples_per_second": 39.18,
+      "eval_steps_per_second": 4.898,
+      "eval_wer": 0.534014860939481,
       "step": 14000
     },
     {
       "epoch": 0.5697669849502928,
-      "grad_norm": 2.742039918899536,
-      "learning_rate": 0.00025781005025125625,
-      "loss": 0.5211,
+      "grad_norm": 1.8595026731491089,
+      "learning_rate": 0.0002789563909774436,
+      "loss": 0.5296,
       "step": 14500
     },
     {
       "epoch": 0.5894141223623718,
-      "grad_norm": 6.8413472175598145,
-      "learning_rate": 0.00025630251256281405,
-      "loss": 0.4981,
+      "grad_norm": 1.83585524559021,
+      "learning_rate": 0.00027820451127819545,
+      "loss": 0.5115,
       "step": 15000
     },
     {
       "epoch": 0.5894141223623718,
-      "eval_loss": 0.6615394949913025,
-      "eval_runtime": 144.3823,
-      "eval_samples_per_second": 39.174,
-      "eval_steps_per_second": 4.897,
-      "eval_wer": 0.5139702460239765,
+      "eval_loss": 0.6973890662193298,
+      "eval_runtime": 143.7301,
+      "eval_samples_per_second": 39.352,
+      "eval_steps_per_second": 4.919,
+      "eval_wer": 0.5437081735167145,
       "step": 15000
     },
     {
       "epoch": 0.6090612597744509,
-      "grad_norm": 19.166942596435547,
-      "learning_rate": 0.00025479497487437185,
-      "loss": 0.4884,
+      "grad_norm": 3.633268117904663,
+      "learning_rate": 0.00027745413533834584,
+      "loss": 0.5083,
       "step": 15500
     },
     {
       "epoch": 0.6287083971865299,
-      "grad_norm": 2.290081024169922,
-      "learning_rate": 0.00025328743718592965,
-      "loss": 0.4891,
+      "grad_norm": 1.6785953044891357,
+      "learning_rate": 0.0002767022556390977,
+      "loss": 0.5097,
       "step": 16000
     },
     {
       "epoch": 0.6287083971865299,
-      "eval_loss": 0.6634296178817749,
-      "eval_runtime": 145.228,
-      "eval_samples_per_second": 38.946,
-      "eval_steps_per_second": 4.868,
-      "eval_wer": 0.5249474410617708,
+      "eval_loss": 0.6785907745361328,
+      "eval_runtime": 144.7589,
+      "eval_samples_per_second": 39.072,
+      "eval_steps_per_second": 4.884,
+      "eval_wer": 0.5198119112195279,
       "step": 16000
     },
     {
       "epoch": 0.648355534598609,
-      "grad_norm": 2.9746391773223877,
-      "learning_rate": 0.0002517829145728643,
-      "loss": 0.4809,
+      "grad_norm": 3.022169589996338,
+      "learning_rate": 0.00027595037593984963,
+      "loss": 0.5025,
       "step": 16500
     },
     {
       "epoch": 0.6680026720106881,
-      "grad_norm": 3.1741456985473633,
-      "learning_rate": 0.0002502753768844221,
-      "loss": 0.4813,
+      "grad_norm": 3.9664957523345947,
+      "learning_rate": 0.00027519999999999997,
+      "loss": 0.504,
       "step": 17000
     },
     {
       "epoch": 0.6680026720106881,
-      "eval_loss": 0.6469489336013794,
-      "eval_runtime": 146.1886,
-      "eval_samples_per_second": 38.69,
-      "eval_steps_per_second": 4.836,
-      "eval_wer": 0.5105198119112195,
+      "eval_loss": 0.6679931282997131,
+      "eval_runtime": 144.4325,
+      "eval_samples_per_second": 39.16,
+      "eval_steps_per_second": 4.895,
+      "eval_wer": 0.5067163101218084,
       "step": 17000
     },
     {
       "epoch": 0.6876498094227671,
-      "grad_norm": 2.78464937210083,
-      "learning_rate": 0.0002487678391959799,
-      "loss": 0.4788,
+      "grad_norm": 1.8392106294631958,
+      "learning_rate": 0.00027444812030075184,
+      "loss": 0.4996,
       "step": 17500
     },
     {
       "epoch": 0.7072969468348461,
-      "grad_norm": 2.7327282428741455,
-      "learning_rate": 0.00024726331658291457,
-      "loss": 0.4799,
+      "grad_norm": 22.285568237304688,
+      "learning_rate": 0.0002736977443609022,
+      "loss": 0.4951,
       "step": 18000
     },
     {
       "epoch": 0.7072969468348461,
-      "eval_loss": 0.6420897245407104,
-      "eval_runtime": 145.2793,
-      "eval_samples_per_second": 38.932,
-      "eval_steps_per_second": 4.866,
-      "eval_wer": 0.5014202949719954,
+      "eval_loss": 0.6599805951118469,
+      "eval_runtime": 144.0873,
+      "eval_samples_per_second": 39.254,
+      "eval_steps_per_second": 4.907,
+      "eval_wer": 0.5222191908330792,
       "step": 18000
     },
     {
       "epoch": 0.7269440842469252,
-      "grad_norm": 1.9746254682540894,
-      "learning_rate": 0.00024575577889447236,
-      "loss": 0.4698,
+      "grad_norm": 1.8236407041549683,
+      "learning_rate": 0.0002729458646616541,
+      "loss": 0.5008,
       "step": 18500
     },
     {
       "epoch": 0.7465912216590043,
-      "grad_norm": 1.833760142326355,
-      "learning_rate": 0.0002442482412060301,
-      "loss": 0.4799,
+      "grad_norm": 3.337568998336792,
+      "learning_rate": 0.000272193984962406,
+      "loss": 0.4982,
       "step": 19000
     },
     {
       "epoch": 0.7465912216590043,
-      "eval_loss": 0.614359438419342,
-      "eval_runtime": 144.72,
-      "eval_samples_per_second": 39.082,
-      "eval_steps_per_second": 4.885,
-      "eval_wer": 0.48193737863298614,
+      "eval_loss": 0.6371914744377136,
+      "eval_runtime": 144.7615,
+      "eval_samples_per_second": 39.071,
+      "eval_steps_per_second": 4.884,
+      "eval_wer": 0.5010993243568551,
       "step": 19000
     },
     {
       "epoch": 0.7662383590710834,
-      "grad_norm": 2.3315391540527344,
-      "learning_rate": 0.0002427407035175879,
-      "loss": 0.4773,
+      "grad_norm": 5.579843997955322,
+      "learning_rate": 0.0002714421052631579,
+      "loss": 0.4966,
       "step": 19500
     },
     {
       "epoch": 0.7858854964831624,
-      "grad_norm": 1.5900750160217285,
-      "learning_rate": 0.0002412331658291457,
-      "loss": 0.471,
+      "grad_norm": 1.800836443901062,
+      "learning_rate": 0.00027069022556390975,
+      "loss": 0.493,
       "step": 20000
     },
     {
       "epoch": 0.7858854964831624,
-      "eval_loss": 0.6183646321296692,
-      "eval_runtime": 144.8701,
-      "eval_samples_per_second": 39.042,
-      "eval_steps_per_second": 4.88,
-      "eval_wer": 0.4914220603103786,
+      "eval_loss": 0.6563202738761902,
+      "eval_runtime": 145.0143,
+      "eval_samples_per_second": 39.003,
+      "eval_steps_per_second": 4.875,
+      "eval_wer": 0.523583315947425,
       "step": 20000
     },
     {
       "epoch": 0.8055326338952414,
-      "grad_norm": 9.128410339355469,
-      "learning_rate": 0.00023972864321608037,
-      "loss": 0.4721,
+      "grad_norm": 2.785470724105835,
+      "learning_rate": 0.0002699383458646616,
+      "loss": 0.5046,
       "step": 20500
     },
     {
       "epoch": 0.8251797713073206,
-      "grad_norm": 2.2207324504852295,
-      "learning_rate": 0.00023822110552763816,
-      "loss": 0.4644,
+      "grad_norm": 2.336695909500122,
+      "learning_rate": 0.00026918646616541354,
+      "loss": 0.4928,
       "step": 21000
     },
     {
       "epoch": 0.8251797713073206,
-      "eval_loss": 0.619006872177124,
-      "eval_runtime": 146.3215,
-      "eval_samples_per_second": 38.655,
-      "eval_steps_per_second": 4.832,
-      "eval_wer": 0.4983068800051355,
+      "eval_loss": 0.6477507948875427,
+      "eval_runtime": 146.5574,
+      "eval_samples_per_second": 38.592,
+      "eval_steps_per_second": 4.824,
+      "eval_wer": 0.5030090995169392,
       "step": 21000
     },
     {
       "epoch": 0.8448269087193996,
-      "grad_norm": 2.999115467071533,
-      "learning_rate": 0.00023671356783919596,
-      "loss": 0.4649,
+      "grad_norm": 1.3052055835723877,
+      "learning_rate": 0.0002684360902255639,
+      "loss": 0.4878,
       "step": 21500
     },
     {
       "epoch": 0.8644740461314786,
-      "grad_norm": 1.8293167352676392,
-      "learning_rate": 0.00023520904522613062,
-      "loss": 0.4645,
+      "grad_norm": 2.2889890670776367,
+      "learning_rate": 0.00026768571428571427,
+      "loss": 0.4964,
       "step": 22000
     },
     {
       "epoch": 0.8644740461314786,
-      "eval_loss": 0.6084673404693604,
-      "eval_runtime": 146.1674,
-      "eval_samples_per_second": 38.695,
-      "eval_steps_per_second": 4.837,
-      "eval_wer": 0.47840670186644413,
+      "eval_loss": 0.6431675553321838,
+      "eval_runtime": 146.4682,
+      "eval_samples_per_second": 38.616,
+      "eval_steps_per_second": 4.827,
+      "eval_wer": 0.5103272295421354,
       "step": 22000
     },
     {
       "epoch": 0.8841211835435577,
-      "grad_norm": 6.625216484069824,
-      "learning_rate": 0.00023370150753768845,
-      "loss": 0.4539,
+      "grad_norm": 3.50435733795166,
+      "learning_rate": 0.00026693383458646614,
+      "loss": 0.4796,
       "step": 22500
     },
     {
       "epoch": 0.9037683209556368,
-      "grad_norm": 2.523075819015503,
-      "learning_rate": 0.00023219396984924622,
-      "loss": 0.4506,
+      "grad_norm": 1.364182472229004,
+      "learning_rate": 0.000266181954887218,
+      "loss": 0.4818,
       "step": 23000
     },
     {
       "epoch": 0.9037683209556368,
-      "eval_loss": 0.6067140698432922,
-      "eval_runtime": 150.1662,
-      "eval_samples_per_second": 37.665,
-      "eval_steps_per_second": 4.708,
-      "eval_wer": 0.4699972717497713,
+      "eval_loss": 0.6235994100570679,
+      "eval_runtime": 146.6113,
+      "eval_samples_per_second": 38.578,
+      "eval_steps_per_second": 4.822,
+      "eval_wer": 0.48960857633483656,
       "step": 23000
     },
     {
       "epoch": 0.9234154583677158,
-      "grad_norm": 2.416762590408325,
-      "learning_rate": 0.000230686432160804,
-      "loss": 0.4405,
+      "grad_norm": 2.1214494705200195,
+      "learning_rate": 0.00026543007518796993,
+      "loss": 0.4688,
       "step": 23500
     },
     {
       "epoch": 0.9430625957797949,
-      "grad_norm": 2.332602024078369,
-      "learning_rate": 0.00022917889447236179,
-      "loss": 0.4439,
+      "grad_norm": 2.6261446475982666,
+      "learning_rate": 0.00026467819548872174,
+      "loss": 0.4752,
       "step": 24000
     },
     {
       "epoch": 0.9430625957797949,
-      "eval_loss": 0.5994084477424622,
-      "eval_runtime": 145.8786,
-      "eval_samples_per_second": 38.772,
+      "eval_loss": 0.6326233744621277,
+      "eval_runtime": 145.9082,
+      "eval_samples_per_second": 38.764,
       "eval_steps_per_second": 4.846,
-      "eval_wer": 0.47725120765193946,
+      "eval_wer": 0.500810450803229,
       "step": 24000
     },
     {
       "epoch": 0.9627097331918739,
-      "grad_norm": 1.971414566040039,
-      "learning_rate": 0.00022767135678391956,
-      "loss": 0.4407,
+      "grad_norm": 4.142037391662598,
+      "learning_rate": 0.00026392631578947367,
+      "loss": 0.4698,
       "step": 24500
     },
     {
       "epoch": 0.982356870603953,
-      "grad_norm": 2.5181193351745605,
-      "learning_rate": 0.00022616381909547738,
-      "loss": 0.4476,
+      "grad_norm": 2.097465991973877,
+      "learning_rate": 0.00026317443609022553,
+      "loss": 0.4736,
       "step": 25000
     },
     {
       "epoch": 0.982356870603953,
-      "eval_loss": 0.5945668816566467,
-      "eval_runtime": 146.0094,
-      "eval_samples_per_second": 38.737,
-      "eval_steps_per_second": 4.842,
-      "eval_wer": 0.4709601835951919,
+      "eval_loss": 0.6309667229652405,
+      "eval_runtime": 145.598,
+      "eval_samples_per_second": 38.847,
+      "eval_steps_per_second": 4.856,
+      "eval_wer": 0.5081446293591821,
       "step": 25000
     },
     {
       "epoch": 1.002004008016032,
-      "grad_norm": 0.8755355477333069,
-      "learning_rate": 0.00022465628140703518,
-      "loss": 0.4328,
+      "grad_norm": 1.872559666633606,
+      "learning_rate": 0.0002624225563909774,
+      "loss": 0.461,
       "step": 25500
     },
     {
       "epoch": 1.0216511454281112,
-      "grad_norm": 1.2238690853118896,
-      "learning_rate": 0.00022314874371859295,
-      "loss": 0.3905,
+      "grad_norm": 0.9968547821044922,
+      "learning_rate": 0.00026167067669172927,
+      "loss": 0.4241,
       "step": 26000
     },
     {
       "epoch": 1.0216511454281112,
-      "eval_loss": 0.5904036164283752,
-      "eval_runtime": 146.0538,
-      "eval_samples_per_second": 38.725,
-      "eval_steps_per_second": 4.841,
-      "eval_wer": 0.4539326924620051,
+      "eval_loss": 0.6126999258995056,
+      "eval_runtime": 145.6037,
+      "eval_samples_per_second": 38.845,
+      "eval_steps_per_second": 4.856,
+      "eval_wer": 0.47134534833336006,
       "step": 26000
     },
     {
       "epoch": 1.0412982828401902,
-      "grad_norm": 0.8446155786514282,
-      "learning_rate": 0.00022164120603015072,
-      "loss": 0.3785,
+      "grad_norm": 0.840844988822937,
+      "learning_rate": 0.00026092030075187966,
+      "loss": 0.4146,
       "step": 26500
     },
     {
       "epoch": 1.0609454202522692,
-      "grad_norm": 1.0839017629623413,
-      "learning_rate": 0.00022013366834170852,
-      "loss": 0.3807,
+      "grad_norm": 0.8603857159614563,
+      "learning_rate": 0.00026016842105263153,
+      "loss": 0.4196,
       "step": 27000
     },
     {
       "epoch": 1.0609454202522692,
-      "eval_loss": 0.5839076042175293,
-      "eval_runtime": 146.3357,
-      "eval_samples_per_second": 38.651,
-      "eval_steps_per_second": 4.831,
-      "eval_wer": 0.45728683539022,
+      "eval_loss": 0.6066301465034485,
+      "eval_runtime": 144.6102,
+      "eval_samples_per_second": 39.112,
+      "eval_steps_per_second": 4.889,
+      "eval_wer": 0.46828007895877133,
       "step": 27000
     },
     {
       "epoch": 1.0805925576643483,
-      "grad_norm": 1.0781913995742798,
-      "learning_rate": 0.0002186291457286432,
-      "loss": 0.3873,
+      "grad_norm": 0.8338613510131836,
+      "learning_rate": 0.00025941654135338345,
+      "loss": 0.4213,
       "step": 27500
     },
     {
       "epoch": 1.1002396950764273,
-      "grad_norm": 1.6534680128097534,
-      "learning_rate": 0.00021712160804020098,
-      "loss": 0.3782,
+      "grad_norm": 1.0736676454544067,
+      "learning_rate": 0.0002586646616541353,
+      "loss": 0.4177,
       "step": 28000
     },
     {
       "epoch": 1.1002396950764273,
-      "eval_loss": 0.5693557262420654,
-      "eval_runtime": 145.969,
-      "eval_samples_per_second": 38.748,
-      "eval_steps_per_second": 4.843,
-      "eval_wer": 0.44768981399752855,
+      "eval_loss": 0.5958611965179443,
+      "eval_runtime": 145.1774,
+      "eval_samples_per_second": 38.959,
+      "eval_steps_per_second": 4.87,
+      "eval_wer": 0.47737959589799556,
       "step": 28000
     },
     {
       "epoch": 1.1198868324885065,
-      "grad_norm": 0.9230701923370361,
-      "learning_rate": 0.00021561407035175877,
-      "loss": 0.3879,
+      "grad_norm": 0.7483401894569397,
+      "learning_rate": 0.0002579127819548872,
+      "loss": 0.4204,
       "step": 28500
     },
     {
       "epoch": 1.1395339699005855,
-      "grad_norm": 0.8729119896888733,
-      "learning_rate": 0.00021410954773869344,
-      "loss": 0.3777,
+      "grad_norm": 1.1266822814941406,
+      "learning_rate": 0.0002571624060150376,
+      "loss": 0.4204,
       "step": 29000
     },
     {
       "epoch": 1.1395339699005855,
-      "eval_loss": 0.5712409615516663,
-      "eval_runtime": 145.9912,
-      "eval_samples_per_second": 38.742,
-      "eval_steps_per_second": 4.843,
-      "eval_wer": 0.4580892619280705,
+      "eval_loss": 0.607071042060852,
+      "eval_runtime": 145.6021,
+      "eval_samples_per_second": 38.846,
+      "eval_steps_per_second": 4.856,
+      "eval_wer": 0.4893517998427244,
       "step": 29000
     },
     {
       "epoch": 1.1591811073126645,
-      "grad_norm": 4.197005271911621,
-      "learning_rate": 0.00021260201005025123,
-      "loss": 0.3763,
+      "grad_norm": 2.852926731109619,
+      "learning_rate": 0.00025641052631578945,
+      "loss": 0.4255,
       "step": 29500
     },
     {
       "epoch": 1.1788282447247436,
-      "grad_norm": 1.8312333822250366,
-      "learning_rate": 0.0002110974874371859,
-      "loss": 0.3879,
+      "grad_norm": 2.134554862976074,
+      "learning_rate": 0.00025566015037593984,
+      "loss": 0.4238,
       "step": 30000
     },
     {
       "epoch": 1.1788282447247436,
-      "eval_loss": 0.5694305300712585,
-      "eval_runtime": 145.4398,
-      "eval_samples_per_second": 38.889,
-      "eval_steps_per_second": 4.861,
-      "eval_wer": 0.45385244980822004,
+      "eval_loss": 0.600638747215271,
+      "eval_runtime": 144.9068,
+      "eval_samples_per_second": 39.032,
+      "eval_steps_per_second": 4.879,
+      "eval_wer": 0.476432732583332,
       "step": 30000
     },
     {
       "epoch": 1.1984753821368226,
-      "grad_norm": 1.415966510772705,
-      "learning_rate": 0.0002095899497487437,
-      "loss": 0.3746,
+      "grad_norm": 1.6866382360458374,
+      "learning_rate": 0.0002549082706766917,
+      "loss": 0.4156,
       "step": 30500
     },
     {
       "epoch": 1.2181225195489018,
-      "grad_norm": 0.7083045244216919,
-      "learning_rate": 0.00020808241206030147,
-      "loss": 0.3817,
+      "grad_norm": 0.7769395709037781,
+      "learning_rate": 0.0002541563909774436,
+      "loss": 0.4253,
       "step": 31000
     },
     {
       "epoch": 1.2181225195489018,
-      "eval_loss": 0.555791974067688,
-      "eval_runtime": 146.2218,
-      "eval_samples_per_second": 38.681,
-      "eval_steps_per_second": 4.835,
-      "eval_wer": 0.4413987899407809,
+      "eval_loss": 0.5803025960922241,
+      "eval_runtime": 144.9773,
+      "eval_samples_per_second": 39.013,
+      "eval_steps_per_second": 4.877,
+      "eval_wer": 0.4623421225786779,
       "step": 31000
     },
     {
       "epoch": 1.2377696569609808,
-      "grad_norm": 3.3128950595855713,
-      "learning_rate": 0.0002065748743718593,
-      "loss": 0.379,
+      "grad_norm": 1.305306077003479,
+      "learning_rate": 0.00025340451127819544,
+      "loss": 0.4257,
       "step": 31500
     },
     {
       "epoch": 1.2574167943730599,
-      "grad_norm": 1.0298165082931519,
-      "learning_rate": 0.00020506733668341706,
-      "loss": 0.3755,
+      "grad_norm": 0.9590096473693848,
+      "learning_rate": 0.00025265263157894736,
+      "loss": 0.4156,
       "step": 32000
     },
     {
       "epoch": 1.2574167943730599,
-      "eval_loss": 0.5634162425994873,
-      "eval_runtime": 144.9179,
-      "eval_samples_per_second": 39.029,
-      "eval_steps_per_second": 4.879,
-      "eval_wer": 0.43432138787693986,
+      "eval_loss": 0.5940248966217041,
+      "eval_runtime": 145.3704,
+      "eval_samples_per_second": 38.908,
+      "eval_steps_per_second": 4.863,
+      "eval_wer": 0.45728683539022,
       "step": 32000
     },
     {
       "epoch": 1.2770639317851389,
-      "grad_norm": 1.2316133975982666,
-      "learning_rate": 0.00020355979899497486,
-      "loss": 0.3656,
+      "grad_norm": 1.598183274269104,
+      "learning_rate": 0.00025190075187969923,
+      "loss": 0.4103,
       "step": 32500
     },
     {
       "epoch": 1.296711069197218,
-      "grad_norm": 1.7103346586227417,
-      "learning_rate": 0.00020205226130653263,
-      "loss": 0.3629,
+      "grad_norm": 1.9396251440048218,
+      "learning_rate": 0.0002511488721804511,
+      "loss": 0.4058,
       "step": 33000
     },
     {
       "epoch": 1.296711069197218,
-      "eval_loss": 0.5455352663993835,
-      "eval_runtime": 144.6962,
-      "eval_samples_per_second": 39.089,
-      "eval_steps_per_second": 4.886,
-      "eval_wer": 0.4341929996308838,
+      "eval_loss": 0.5802159905433655,
+      "eval_runtime": 146.1677,
+      "eval_samples_per_second": 38.695,
+      "eval_steps_per_second": 4.837,
+      "eval_wer": 0.4614594533870424,
       "step": 33000
     },
     {
       "epoch": 1.316358206609297,
-      "grad_norm": 1.2906711101531982,
-      "learning_rate": 0.00020054472361809043,
-      "loss": 0.3786,
+      "grad_norm": 1.7467131614685059,
+      "learning_rate": 0.00025039699248120297,
+      "loss": 0.417,
       "step": 33500
     },
     {
       "epoch": 1.3360053440213762,
-      "grad_norm": 1.3450685739517212,
-      "learning_rate": 0.00019903718592964822,
-      "loss": 0.3636,
+      "grad_norm": 0.9871892333030701,
+      "learning_rate": 0.0002496451127819549,
+      "loss": 0.404,
       "step": 34000
     },
     {
       "epoch": 1.3360053440213762,
-      "eval_loss": 0.5471683740615845,
-      "eval_runtime": 145.1506,
-      "eval_samples_per_second": 38.966,
-      "eval_steps_per_second": 4.871,
-      "eval_wer": 0.434562115838295,
+      "eval_loss": 0.5882492065429688,
+      "eval_runtime": 145.2339,
+      "eval_samples_per_second": 38.944,
+      "eval_steps_per_second": 4.868,
+      "eval_wer": 0.46015952239572466,
       "step": 34000
     },
     {
       "epoch": 1.3556524814334552,
-      "grad_norm": 1.1486555337905884,
-      "learning_rate": 0.00019752964824120602,
-      "loss": 0.3553,
+      "grad_norm": 1.0927810668945312,
+      "learning_rate": 0.00024889323308270676,
+      "loss": 0.3955,
       "step": 34500
     },
     {
       "epoch": 1.3752996188455342,
-      "grad_norm": 1.4870734214782715,
-      "learning_rate": 0.0001960221105527638,
-      "loss": 0.3566,
+      "grad_norm": 1.8038376569747925,
+      "learning_rate": 0.0002481413533834586,
+      "loss": 0.3995,
       "step": 35000
     },
     {
       "epoch": 1.3752996188455342,
-      "eval_loss": 0.5466644167900085,
-      "eval_runtime": 145.1639,
-      "eval_samples_per_second": 38.963,
-      "eval_steps_per_second": 4.87,
-      "eval_wer": 0.4321548362247436,
+      "eval_loss": 0.5840802788734436,
+      "eval_runtime": 144.2215,
+      "eval_samples_per_second": 39.217,
+      "eval_steps_per_second": 4.902,
+      "eval_wer": 0.46152364751007047,
       "step": 35000
     },
     {
       "epoch": 1.3949467562576132,
-      "grad_norm": 2.2068591117858887,
-      "learning_rate": 0.0001945145728643216,
-      "loss": 0.3592,
+      "grad_norm": 1.3720190525054932,
+      "learning_rate": 0.0002473894736842105,
+      "loss": 0.4077,
       "step": 35500
     },
     {
       "epoch": 1.4145938936696925,
-      "grad_norm": 1.370450735092163,
-      "learning_rate": 0.00019300703517587936,
-      "loss": 0.3683,
+      "grad_norm": 1.4488073587417603,
+      "learning_rate": 0.0002466375939849624,
+      "loss": 0.4049,
       "step": 36000
     },
     {
       "epoch": 1.4145938936696925,
-      "eval_loss": 0.5440675020217896,
-      "eval_runtime": 144.9148,
-      "eval_samples_per_second": 39.03,
-      "eval_steps_per_second": 4.879,
-      "eval_wer": 0.4324597583091268,
+      "eval_loss": 0.5853234529495239,
+      "eval_runtime": 144.437,
+      "eval_samples_per_second": 39.159,
+      "eval_steps_per_second": 4.895,
+      "eval_wer": 0.4635618109162106,
       "step": 36000
     },
     {
       "epoch": 1.4342410310817715,
-      "grad_norm": 1.2991498708724976,
-      "learning_rate": 0.00019149949748743718,
-      "loss": 0.372,
+      "grad_norm": 2.5115835666656494,
+      "learning_rate": 0.00024588571428571423,
+      "loss": 0.4202,
       "step": 36500
     },
     {
       "epoch": 1.4538881684938505,
-      "grad_norm": 0.8429681062698364,
-      "learning_rate": 0.00018999195979899495,
-      "loss": 0.3581,
+      "grad_norm": 0.7861095070838928,
+      "learning_rate": 0.0002451353383458647,
+      "loss": 0.4018,
       "step": 37000
     },
     {
       "epoch": 1.4538881684938505,
-      "eval_loss": 0.5278629660606384,
-      "eval_runtime": 144.4965,
-      "eval_samples_per_second": 39.143,
-      "eval_steps_per_second": 4.893,
-      "eval_wer": 0.4191876233730802,
+      "eval_loss": 0.5737255215644836,
+      "eval_runtime": 143.969,
+      "eval_samples_per_second": 39.286,
+      "eval_steps_per_second": 4.911,
+      "eval_wer": 0.4532747027009677,
       "step": 37000
     },
     {
       "epoch": 1.4735353059059295,
-      "grad_norm": 1.4340102672576904,
-      "learning_rate": 0.00018848743718592964,
-      "loss": 0.3522,
+      "grad_norm": 1.29477858543396,
+      "learning_rate": 0.0002443834586466165,
+      "loss": 0.4028,
       "step": 37500
     },
     {
       "epoch": 1.4931824433180085,
-      "grad_norm": 3.3566503524780273,
-      "learning_rate": 0.00018697989949748744,
-      "loss": 0.3448,
+      "grad_norm": 1.5367540121078491,
+      "learning_rate": 0.0002436315789473684,
+      "loss": 0.3906,
       "step": 38000
     },
     {
       "epoch": 1.4931824433180085,
-      "eval_loss": 0.5341060161590576,
-      "eval_runtime": 144.4725,
-      "eval_samples_per_second": 39.149,
-      "eval_steps_per_second": 4.894,
-      "eval_wer": 0.41947649692670635,
+      "eval_loss": 0.5848459005355835,
+      "eval_runtime": 144.7268,
+      "eval_samples_per_second": 39.081,
+      "eval_steps_per_second": 4.885,
+      "eval_wer": 0.4637222962237807,
       "step": 38000
     },
     {
       "epoch": 1.5128295807300876,
-      "grad_norm": 5.104796886444092,
-      "learning_rate": 0.0001854723618090452,
-      "loss": 0.3658,
+      "grad_norm": 1.7903566360473633,
+      "learning_rate": 0.00024288120300751878,
+      "loss": 0.4147,
       "step": 38500
     },
     {
       "epoch": 1.5324767181421666,
-      "grad_norm": 0.7495909929275513,
-      "learning_rate": 0.000183964824120603,
-      "loss": 0.3558,
+      "grad_norm": 0.7342734336853027,
+      "learning_rate": 0.00024212932330827064,
+      "loss": 0.3932,
       "step": 39000
     },
     {
       "epoch": 1.5324767181421666,
-      "eval_loss": 0.5193748474121094,
-      "eval_runtime": 144.086,
-      "eval_samples_per_second": 39.254,
-      "eval_steps_per_second": 4.907,
-      "eval_wer": 0.4212418353099774,
+      "eval_loss": 0.551567792892456,
+      "eval_runtime": 144.7104,
+      "eval_samples_per_second": 39.085,
+      "eval_steps_per_second": 4.886,
+      "eval_wer": 0.44000256776492114,
       "step": 39000
     },
     {
       "epoch": 1.5521238555542456,
-      "grad_norm": 1.2974355220794678,
-      "learning_rate": 0.00018246030150753767,
-      "loss": 0.3487,
+      "grad_norm": 1.3719693422317505,
+      "learning_rate": 0.00024137894736842104,
+      "loss": 0.3984,
       "step": 39500
     },
     {
       "epoch": 1.5717709929663248,
-      "grad_norm": 1.052008867263794,
-      "learning_rate": 0.00018095276381909547,
-      "loss": 0.3492,
+      "grad_norm": 3.398484706878662,
+      "learning_rate": 0.0002406270676691729,
+      "loss": 0.4026,
       "step": 40000
     },
     {
       "epoch": 1.5717709929663248,
-      "eval_loss": 0.5243468284606934,
-      "eval_runtime": 144.7366,
-      "eval_samples_per_second": 39.078,
-      "eval_steps_per_second": 4.885,
-      "eval_wer": 0.4139076567540242,
+      "eval_loss": 0.5641522407531738,
+      "eval_runtime": 145.5525,
+      "eval_samples_per_second": 38.859,
+      "eval_steps_per_second": 4.857,
+      "eval_wer": 0.44844409494310794,
       "step": 40000
     },
     {
       "epoch": 1.5914181303784039,
-      "grad_norm": 1.5572227239608765,
-      "learning_rate": 0.00017944522613065324,
-      "loss": 0.3581,
+      "grad_norm": 1.1821295022964478,
+      "learning_rate": 0.0002398751879699248,
+      "loss": 0.4086,
       "step": 40500
     },
     {
       "epoch": 1.611065267790483,
-      "grad_norm": 1.3091472387313843,
-      "learning_rate": 0.00017793768844221104,
-      "loss": 0.3461,
+      "grad_norm": 1.3344157934188843,
+      "learning_rate": 0.00023912330827067667,
+      "loss": 0.396,
       "step": 41000
     },
     {
       "epoch": 1.611065267790483,
-      "eval_loss": 0.5143682360649109,
-      "eval_runtime": 145.6844,
-      "eval_samples_per_second": 38.824,
-      "eval_steps_per_second": 4.853,
-      "eval_wer": 0.40458346038420184,
+      "eval_loss": 0.5584043264389038,
+      "eval_runtime": 145.5026,
+      "eval_samples_per_second": 38.872,
+      "eval_steps_per_second": 4.859,
+      "eval_wer": 0.4512044422333135,
       "step": 41000
     },
     {
       "epoch": 1.630712405202562,
-      "grad_norm": 0.52028888463974,
-      "learning_rate": 0.00017643015075376883,
-      "loss": 0.34,
+      "grad_norm": 0.6132605671882629,
+      "learning_rate": 0.00023837142857142856,
+      "loss": 0.3863,
       "step": 41500
     },
     {
       "epoch": 1.6503595426146411,
-      "grad_norm": 1.1602140665054321,
-      "learning_rate": 0.00017492261306532663,
-      "loss": 0.3412,
+      "grad_norm": 2.896801710128784,
+      "learning_rate": 0.00023761954887218043,
+      "loss": 0.3976,
       "step": 42000
     },
     {
       "epoch": 1.6503595426146411,
-      "eval_loss": 0.5344811081886292,
-      "eval_runtime": 146.2652,
-      "eval_samples_per_second": 38.669,
-      "eval_steps_per_second": 4.834,
-      "eval_wer": 0.42369726051579976,
+      "eval_loss": 0.5537524819374084,
+      "eval_runtime": 145.3562,
+      "eval_samples_per_second": 38.911,
+      "eval_steps_per_second": 4.864,
+      "eval_wer": 0.4436455842467622,
       "step": 42000
     },
     {
       "epoch": 1.6700066800267201,
-      "grad_norm": 0.4986182749271393,
-      "learning_rate": 0.0001734150753768844,
-      "loss": 0.3501,
+      "grad_norm": 0.4839102029800415,
+      "learning_rate": 0.00023686766917293232,
+      "loss": 0.3977,
       "step": 42500
     },
     {
       "epoch": 1.6896538174387992,
-      "grad_norm": 1.2773915529251099,
-      "learning_rate": 0.0001719075376884422,
-      "loss": 0.3424,
+      "grad_norm": 0.7648475170135498,
+      "learning_rate": 0.0002361157894736842,
+      "loss": 0.3936,
       "step": 43000
     },
     {
       "epoch": 1.6896538174387992,
-      "eval_loss": 0.5192400813102722,
-      "eval_runtime": 144.477,
-      "eval_samples_per_second": 39.148,
-      "eval_steps_per_second": 4.894,
-      "eval_wer": 0.40917334018070645,
+      "eval_loss": 0.551811158657074,
+      "eval_runtime": 144.7074,
+      "eval_samples_per_second": 39.086,
+      "eval_steps_per_second": 4.886,
+      "eval_wer": 0.4412222561024538,
       "step": 43000
     },
     {
       "epoch": 1.7093009548508782,
-      "grad_norm": 1.184127926826477,
-      "learning_rate": 0.00017040301507537686,
-      "loss": 0.3353,
+      "grad_norm": 1.953736662864685,
+      "learning_rate": 0.00023536541353383458,
+      "loss": 0.3865,
       "step": 43500
     },
     {
       "epoch": 1.7289480922629572,
-      "grad_norm": 1.3090980052947998,
-      "learning_rate": 0.00016889547738693466,
-      "loss": 0.341,
+      "grad_norm": 1.4531214237213135,
+      "learning_rate": 0.00023461353383458645,
+      "loss": 0.3879,
       "step": 44000
     },
     {
       "epoch": 1.7289480922629572,
-      "eval_loss": 0.5130703449249268,
-      "eval_runtime": 144.4345,
-      "eval_samples_per_second": 39.16,
-      "eval_steps_per_second": 4.895,
-      "eval_wer": 0.40559451782189343,
+      "eval_loss": 0.5469211935997009,
+      "eval_runtime": 145.1619,
+      "eval_samples_per_second": 38.963,
+      "eval_steps_per_second": 4.87,
+      "eval_wer": 0.42974755661119224,
       "step": 44000
     },
     {
       "epoch": 1.7485952296750362,
-      "grad_norm": 1.4495439529418945,
-      "learning_rate": 0.00016738793969849243,
-      "loss": 0.3448,
+      "grad_norm": 1.0637991428375244,
+      "learning_rate": 0.00023386165413533835,
+      "loss": 0.3942,
       "step": 44500
     },
     {
       "epoch": 1.7682423670871152,
-      "grad_norm": 2.0205602645874023,
-      "learning_rate": 0.00016588040201005025,
-      "loss": 0.3428,
+      "grad_norm": 1.0606558322906494,
+      "learning_rate": 0.00023310977443609021,
+      "loss": 0.3939,
       "step": 45000
     },
     {
       "epoch": 1.7682423670871152,
-      "eval_loss": 0.5109943151473999,
-      "eval_runtime": 145.377,
-      "eval_samples_per_second": 38.906,
-      "eval_steps_per_second": 4.863,
-      "eval_wer": 0.4030107043700149,
+      "eval_loss": 0.5502393245697021,
+      "eval_runtime": 144.8654,
+      "eval_samples_per_second": 39.043,
+      "eval_steps_per_second": 4.88,
+      "eval_wer": 0.44024329572627624,
       "step": 45000
     },
     {
       "epoch": 1.7878895044991945,
-      "grad_norm": 0.9400632381439209,
-      "learning_rate": 0.00016437286432160802,
-      "loss": 0.3419,
+      "grad_norm": 0.7499143481254578,
+      "learning_rate": 0.0002323578947368421,
+      "loss": 0.3926,
       "step": 45500
     },
     {
       "epoch": 1.8075366419112735,
-      "grad_norm": 1.3378790616989136,
-      "learning_rate": 0.00016286834170854271,
-      "loss": 0.3337,
+      "grad_norm": 1.3015657663345337,
+      "learning_rate": 0.00023160601503759395,
+      "loss": 0.386,
       "step": 46000
     },
     {
       "epoch": 1.8075366419112735,
-      "eval_loss": 0.5063343644142151,
-      "eval_runtime": 144.676,
-      "eval_samples_per_second": 39.094,
-      "eval_steps_per_second": 4.887,
-      "eval_wer": 0.4051130618991831,
+      "eval_loss": 0.5626779198646545,
+      "eval_runtime": 145.6466,
+      "eval_samples_per_second": 38.834,
+      "eval_steps_per_second": 4.854,
+      "eval_wer": 0.4409012854873136,
       "step": 46000
     },
     {
       "epoch": 1.8271837793233527,
-      "grad_norm": 1.5063832998275757,
-      "learning_rate": 0.0001613608040201005,
-      "loss": 0.3298,
+      "grad_norm": 1.1235235929489136,
+      "learning_rate": 0.00023085413533834585,
+      "loss": 0.3833,
       "step": 46500
     },
     {
       "epoch": 1.8468309167354318,
-      "grad_norm": 0.6801648736000061,
-      "learning_rate": 0.00015985326633165828,
-      "loss": 0.3286,
+      "grad_norm": 0.8004291653633118,
+      "learning_rate": 0.0002301022556390977,
+      "loss": 0.3823,
       "step": 47000
     },
     {
       "epoch": 1.8468309167354318,
-      "eval_loss": 0.5045257210731506,
-      "eval_runtime": 144.5535,
-      "eval_samples_per_second": 39.127,
-      "eval_steps_per_second": 4.891,
-      "eval_wer": 0.39761839803565985,
+      "eval_loss": 0.5602549910545349,
+      "eval_runtime": 145.4825,
+      "eval_samples_per_second": 38.878,
+      "eval_steps_per_second": 4.86,
+      "eval_wer": 0.43724222047471556,
       "step": 47000
     },
     {
       "epoch": 1.8664780541475108,
-      "grad_norm": 1.5206536054611206,
-      "learning_rate": 0.00015834572864321605,
-      "loss": 0.3303,
+      "grad_norm": 2.6733715534210205,
+      "learning_rate": 0.0002293503759398496,
+      "loss": 0.3868,
       "step": 47500
     },
     {
       "epoch": 1.8861251915595898,
-      "grad_norm": 2.1600515842437744,
-      "learning_rate": 0.00015683819095477385,
-      "loss": 0.3422,
+      "grad_norm": 1.501878261566162,
+      "learning_rate": 0.00022859849624060148,
+      "loss": 0.3955,
       "step": 48000
     },
     {
       "epoch": 1.8861251915595898,
-      "eval_loss": 0.4938177168369293,
-      "eval_runtime": 145.0534,
-      "eval_samples_per_second": 38.993,
-      "eval_steps_per_second": 4.874,
-      "eval_wer": 0.40270578228563175,
+      "eval_loss": 0.534982442855835,
+      "eval_runtime": 145.013,
+      "eval_samples_per_second": 39.003,
+      "eval_steps_per_second": 4.875,
+      "eval_wer": 0.4308228081719119,
       "step": 48000
     },
     {
       "epoch": 1.9057723289716688,
-      "grad_norm": 0.6919649243354797,
-      "learning_rate": 0.00015533065326633162,
-      "loss": 0.3262,
+      "grad_norm": 1.785569667816162,
+      "learning_rate": 0.00022784661654135337,
+      "loss": 0.3766,
       "step": 48500
     },
     {
       "epoch": 1.9254194663837478,
-      "grad_norm": 1.5237656831741333,
-      "learning_rate": 0.00015382311557788944,
-      "loss": 0.3271,
+      "grad_norm": 0.956910252571106,
+      "learning_rate": 0.00022709473684210524,
+      "loss": 0.3808,
       "step": 49000
     },
     {
       "epoch": 1.9254194663837478,
-      "eval_loss": 0.4978640377521515,
-      "eval_runtime": 145.4834,
-      "eval_samples_per_second": 38.877,
+      "eval_loss": 0.550835132598877,
+      "eval_runtime": 145.4747,
+      "eval_samples_per_second": 38.88,
       "eval_steps_per_second": 4.86,
-      "eval_wer": 0.39097430630225805,
+      "eval_wer": 0.44476898139975285,
       "step": 49000
     },
     {
       "epoch": 1.9450666037958269,
-      "grad_norm": 0.9466130137443542,
-      "learning_rate": 0.0001523155778894472,
-      "loss": 0.3243,
+      "grad_norm": 1.0630252361297607,
+      "learning_rate": 0.00022634285714285713,
+      "loss": 0.3794,
       "step": 49500
     },
     {
       "epoch": 1.9647137412079059,
-      "grad_norm": 0.8876349925994873,
-      "learning_rate": 0.0001508110552763819,
-      "loss": 0.3313,
+      "grad_norm": 0.9397912621498108,
+      "learning_rate": 0.000225593984962406,
+      "loss": 0.3871,
       "step": 50000
     },
     {
       "epoch": 1.9647137412079059,
-      "eval_loss": 0.490749329328537,
-      "eval_runtime": 144.969,
-      "eval_samples_per_second": 39.015,
-      "eval_steps_per_second": 4.877,
-      "eval_wer": 0.3973776700743047,
+      "eval_loss": 0.5386993885040283,
+      "eval_runtime": 144.7105,
+      "eval_samples_per_second": 39.085,
+      "eval_steps_per_second": 4.886,
+      "eval_wer": 0.43197830238641655,
       "step": 50000
     },
     {
       "epoch": 1.9843608786199851,
-      "grad_norm": 1.058655023574829,
-      "learning_rate": 0.00014930653266331657,
-      "loss": 0.3138,
+      "grad_norm": 1.227219581604004,
+      "learning_rate": 0.00022484360902255636,
+      "loss": 0.371,
       "step": 50500
     },
     {
       "epoch": 2.004008016032064,
-      "grad_norm": 0.8039568066596985,
-      "learning_rate": 0.00014779899497487437,
-      "loss": 0.3069,
+      "grad_norm": 1.0407652854919434,
+      "learning_rate": 0.00022409172932330825,
+      "loss": 0.3668,
       "step": 51000
     },
     {
       "epoch": 2.004008016032064,
-      "eval_loss": 0.4898728132247925,
-      "eval_runtime": 145.2853,
-      "eval_samples_per_second": 38.93,
-      "eval_steps_per_second": 4.866,
-      "eval_wer": 0.38524498082200576,
+      "eval_loss": 0.5476531982421875,
+      "eval_runtime": 144.735,
+      "eval_samples_per_second": 39.078,
+      "eval_steps_per_second": 4.885,
+      "eval_wer": 0.4207443308565101,
       "step": 51000
     },
     {
       "epoch": 2.0236551534441434,
-      "grad_norm": 1.5901250839233398,
-      "learning_rate": 0.00014629145728643214,
-      "loss": 0.2734,
+      "grad_norm": 3.713465929031372,
+      "learning_rate": 0.00022333984962406012,
+      "loss": 0.3303,
       "step": 51500
     },
     {
       "epoch": 2.0433022908562224,
-      "grad_norm": 0.6160286664962769,
-      "learning_rate": 0.00014478391959798993,
-      "loss": 0.2771,
+      "grad_norm": 1.074621319770813,
+      "learning_rate": 0.00022258796992481202,
+      "loss": 0.3324,
       "step": 52000
     },
     {
       "epoch": 2.0433022908562224,
-      "eval_loss": 0.4836235046386719,
-      "eval_runtime": 145.1471,
-      "eval_samples_per_second": 38.967,
-      "eval_steps_per_second": 4.871,
-      "eval_wer": 0.3845067484071833,
+      "eval_loss": 0.5283042788505554,
+      "eval_runtime": 144.7457,
+      "eval_samples_per_second": 39.075,
+      "eval_steps_per_second": 4.884,
+      "eval_wer": 0.4227985427934073,
       "step": 52000
     },
     {
       "epoch": 2.0629494282683014,
-      "grad_norm": 1.2271616458892822,
-      "learning_rate": 0.00014327638190954773,
-      "loss": 0.2705,
+      "grad_norm": 1.2761338949203491,
+      "learning_rate": 0.00022183759398496238,
+      "loss": 0.3299,
       "step": 52500
     },
     {
       "epoch": 2.0825965656803804,
-      "grad_norm": 1.0109779834747314,
-      "learning_rate": 0.00014176884422110553,
-      "loss": 0.2705,
+      "grad_norm": 0.9065299034118652,
+      "learning_rate": 0.00022108571428571425,
+      "loss": 0.3327,
       "step": 53000
     },
     {
       "epoch": 2.0825965656803804,
-      "eval_loss": 0.4929197132587433,
-      "eval_runtime": 150.0346,
-      "eval_samples_per_second": 37.698,
-      "eval_steps_per_second": 4.712,
-      "eval_wer": 0.38250068206255716,
+      "eval_loss": 0.5217949151992798,
+      "eval_runtime": 145.0885,
+      "eval_samples_per_second": 38.983,
+      "eval_steps_per_second": 4.873,
+      "eval_wer": 0.41557670395275315,
       "step": 53000
     },
     {
       "epoch": 2.1022437030924594,
-      "grad_norm": 0.9157425165176392,
-      "learning_rate": 0.0001402613065326633,
-      "loss": 0.2748,
+      "grad_norm": 1.1222054958343506,
+      "learning_rate": 0.00022033383458646615,
+      "loss": 0.3347,
       "step": 53500
     },
     {
       "epoch": 2.1218908405045385,
-      "grad_norm": 1.036487102508545,
-      "learning_rate": 0.0001387537688442211,
-      "loss": 0.2654,
+      "grad_norm": 3.639472484588623,
+      "learning_rate": 0.000219581954887218,
+      "loss": 0.3251,
       "step": 54000
     },
     {
       "epoch": 2.1218908405045385,
-      "eval_loss": 0.4842771887779236,
-      "eval_runtime": 149.1221,
-      "eval_samples_per_second": 37.929,
-      "eval_steps_per_second": 4.741,
-      "eval_wer": 0.3813451878480525,
+      "eval_loss": 0.5330758094787598,
+      "eval_runtime": 144.669,
+      "eval_samples_per_second": 39.096,
+      "eval_steps_per_second": 4.887,
+      "eval_wer": 0.41357063760812696,
       "step": 54000
     },
     {
       "epoch": 2.1415379779166175,
-      "grad_norm": 0.9291256070137024,
-      "learning_rate": 0.00013724924623115576,
-      "loss": 0.2719,
+      "grad_norm": 2.337876558303833,
+      "learning_rate": 0.0002188300751879699,
+      "loss": 0.3368,
       "step": 54500
     },
     {
       "epoch": 2.1611851153286965,
-      "grad_norm": 1.0071407556533813,
-      "learning_rate": 0.00013574170854271356,
-      "loss": 0.2794,
+      "grad_norm": 0.8467469811439514,
+      "learning_rate": 0.00021807819548872178,
+      "loss": 0.3466,
       "step": 55000
     },
     {
       "epoch": 2.1611851153286965,
-      "eval_loss": 0.4819933772087097,
-      "eval_runtime": 150.0521,
-      "eval_samples_per_second": 37.694,
-      "eval_steps_per_second": 4.712,
-      "eval_wer": 0.3780712875736226,
+      "eval_loss": 0.5276508927345276,
+      "eval_runtime": 145.3583,
+      "eval_samples_per_second": 38.911,
+      "eval_steps_per_second": 4.864,
+      "eval_wer": 0.4141002391231083,
       "step": 55000
     },
     {
       "epoch": 2.1808322527407755,
-      "grad_norm": 0.558372974395752,
-      "learning_rate": 0.00013423417085427135,
-      "loss": 0.2699,
+      "grad_norm": 0.7859643697738647,
+      "learning_rate": 0.00021732781954887217,
+      "loss": 0.3337,
       "step": 55500
     },
     {
       "epoch": 2.2004793901528545,
-      "grad_norm": 0.5932472944259644,
-      "learning_rate": 0.00013272663316582912,
-      "loss": 0.2644,
+      "grad_norm": 0.8069686889648438,
+      "learning_rate": 0.00021657593984962404,
+      "loss": 0.3259,
       "step": 56000
     },
     {
       "epoch": 2.2004793901528545,
-      "eval_loss": 0.47418466210365295,
-      "eval_runtime": 145.6714,
-      "eval_samples_per_second": 38.827,
-      "eval_steps_per_second": 4.853,
-      "eval_wer": 0.3755195711832582,
+      "eval_loss": 0.522844672203064,
+      "eval_runtime": 145.284,
+      "eval_samples_per_second": 38.931,
+      "eval_steps_per_second": 4.866,
+      "eval_wer": 0.40875607838102423,
       "step": 56000
     },
     {
       "epoch": 2.2201265275649336,
-      "grad_norm": 1.2624306678771973,
-      "learning_rate": 0.00013121909547738692,
-      "loss": 0.2767,
+      "grad_norm": 0.5765830278396606,
+      "learning_rate": 0.00021582406015037593,
+      "loss": 0.337,
       "step": 56500
     },
     {
       "epoch": 2.239773664977013,
-      "grad_norm": 1.0963220596313477,
-      "learning_rate": 0.00012971758793969848,
-      "loss": 0.2624,
+      "grad_norm": 0.9564582109451294,
+      "learning_rate": 0.0002150736842105263,
+      "loss": 0.3292,
       "step": 57000
     },
     {
       "epoch": 2.239773664977013,
-      "eval_loss": 0.46851301193237305,
-      "eval_runtime": 145.312,
-      "eval_samples_per_second": 38.923,
-      "eval_steps_per_second": 4.865,
-      "eval_wer": 0.3680570043812489,
+      "eval_loss": 0.5119462013244629,
+      "eval_runtime": 145.6008,
+      "eval_samples_per_second": 38.846,
+      "eval_steps_per_second": 4.856,
+      "eval_wer": 0.4132657155237438,
       "step": 57000
     },
     {
       "epoch": 2.259420802389092,
-      "grad_norm": 0.6824894547462463,
-      "learning_rate": 0.00012821608040201004,
-      "loss": 0.263,
+      "grad_norm": 0.7495951056480408,
+      "learning_rate": 0.00021432330827067666,
+      "loss": 0.3259,
       "step": 57500
     },
     {
       "epoch": 2.279067939801171,
-      "grad_norm": 0.6939342021942139,
-      "learning_rate": 0.00012670854271356783,
-      "loss": 0.2689,
+      "grad_norm": 0.825587272644043,
+      "learning_rate": 0.00021357142857142855,
+      "loss": 0.3323,
       "step": 58000
     },
     {
       "epoch": 2.279067939801171,
-      "eval_loss": 0.4649806618690491,
-      "eval_runtime": 145.0546,
-      "eval_samples_per_second": 38.992,
-      "eval_steps_per_second": 4.874,
-      "eval_wer": 0.366532393959333,
+      "eval_loss": 0.5191282033920288,
+      "eval_runtime": 145.5654,
+      "eval_samples_per_second": 38.855,
+      "eval_steps_per_second": 4.857,
+      "eval_wer": 0.40739195326667843,
       "step": 58000
     },
     {
       "epoch": 2.29871507721325,
-      "grad_norm": 3.4403460025787354,
-      "learning_rate": 0.0001252010050251256,
-      "loss": 0.2634,
+      "grad_norm": 0.9213058948516846,
+      "learning_rate": 0.00021281954887218042,
+      "loss": 0.3292,
       "step": 58500
     },
     {
       "epoch": 2.318362214625329,
-      "grad_norm": 0.7877224683761597,
-      "learning_rate": 0.0001236934673366834,
-      "loss": 0.2584,
+      "grad_norm": 6.7399773597717285,
+      "learning_rate": 0.00021206766917293232,
+      "loss": 0.3228,
       "step": 59000
     },
     {
       "epoch": 2.318362214625329,
-      "eval_loss": 0.4691295921802521,
-      "eval_runtime": 145.6624,
-      "eval_samples_per_second": 38.83,
-      "eval_steps_per_second": 4.854,
-      "eval_wer": 0.3657620644829966,
+      "eval_loss": 0.5073339939117432,
+      "eval_runtime": 145.4974,
+      "eval_samples_per_second": 38.874,
+      "eval_steps_per_second": 4.859,
+      "eval_wer": 0.3955802346295197,
       "step": 59000
     },
     {
       "epoch": 2.338009352037408,
-      "grad_norm": 0.8857819437980652,
-      "learning_rate": 0.0001221859296482412,
-      "loss": 0.2579,
+      "grad_norm": 0.9246654510498047,
+      "learning_rate": 0.00021131578947368419,
+      "loss": 0.3311,
       "step": 59500
     },
     {
       "epoch": 2.357656489449487,
-      "grad_norm": 2.5980610847473145,
-      "learning_rate": 0.000120678391959799,
-      "loss": 0.2535,
+      "grad_norm": 0.8129465579986572,
+      "learning_rate": 0.00021056390977443608,
+      "loss": 0.3172,
       "step": 60000
     },
     {
       "epoch": 2.357656489449487,
-      "eval_loss": 0.46268430352211,
-      "eval_runtime": 145.2629,
-      "eval_samples_per_second": 38.936,
+      "eval_loss": 0.5084324479103088,
+      "eval_runtime": 145.2717,
+      "eval_samples_per_second": 38.934,
       "eval_steps_per_second": 4.867,
-      "eval_wer": 0.37131485612492177,
+      "eval_wer": 0.4045353147919308,
       "step": 60000
     },
     {
       "epoch": 2.377303626861566,
-      "grad_norm": 0.38764873147010803,
-      "learning_rate": 0.00011917085427135678,
-      "loss": 0.2536,
+      "grad_norm": 0.6280909776687622,
+      "learning_rate": 0.00020981203007518795,
+      "loss": 0.3235,
       "step": 60500
     },
     {
       "epoch": 2.396950764273645,
-      "grad_norm": 0.6207642555236816,
-      "learning_rate": 0.00011766331658291456,
-      "loss": 0.2623,
+      "grad_norm": 3.0157957077026367,
+      "learning_rate": 0.00020906015037593984,
+      "loss": 0.332,
       "step": 61000
     },
     {
       "epoch": 2.396950764273645,
-      "eval_loss": 0.4667174816131592,
-      "eval_runtime": 145.1487,
-      "eval_samples_per_second": 38.967,
-      "eval_steps_per_second": 4.871,
-      "eval_wer": 0.3669817528205293,
+      "eval_loss": 0.512955367565155,
+      "eval_runtime": 145.4819,
+      "eval_samples_per_second": 38.878,
+      "eval_steps_per_second": 4.86,
+      "eval_wer": 0.40151819100961306,
       "step": 61000
     },
     {
       "epoch": 2.4165979016857246,
-      "grad_norm": 0.9032436609268188,
-      "learning_rate": 0.00011615577889447236,
-      "loss": 0.2631,
+      "grad_norm": 1.2728731632232666,
+      "learning_rate": 0.0002083082706766917,
+      "loss": 0.3298,
       "step": 61500
     },
     {
       "epoch": 2.4362450390978037,
-      "grad_norm": 1.0269511938095093,
-      "learning_rate": 0.00011464824120603014,
-      "loss": 0.2502,
+      "grad_norm": 6.008030891418457,
+      "learning_rate": 0.0002075563909774436,
+      "loss": 0.3218,
       "step": 62000
     },
     {
       "epoch": 2.4362450390978037,
-      "eval_loss": 0.4592076241970062,
-      "eval_runtime": 144.6818,
-      "eval_samples_per_second": 39.093,
-      "eval_steps_per_second": 4.887,
-      "eval_wer": 0.3680570043812489,
+      "eval_loss": 0.5102687478065491,
+      "eval_runtime": 145.1668,
+      "eval_samples_per_second": 38.962,
+      "eval_steps_per_second": 4.87,
+      "eval_wer": 0.39972075556482806,
       "step": 62000
     },
     {
       "epoch": 2.4558921765098827,
-      "grad_norm": 0.4968740940093994,
-      "learning_rate": 0.00011314070351758793,
-      "loss": 0.2514,
+      "grad_norm": 1.1027765274047852,
+      "learning_rate": 0.00020680451127819547,
+      "loss": 0.3207,
       "step": 62500
     },
     {
       "epoch": 2.4755393139219617,
-      "grad_norm": 0.9083556532859802,
-      "learning_rate": 0.00011163316582914572,
-      "loss": 0.2593,
+      "grad_norm": 0.9439337849617004,
+      "learning_rate": 0.00020605263157894737,
+      "loss": 0.3317,
       "step": 63000
     },
     {
       "epoch": 2.4755393139219617,
-      "eval_loss": 0.4568769037723541,
-      "eval_runtime": 144.6454,
-      "eval_samples_per_second": 39.103,
-      "eval_steps_per_second": 4.888,
-      "eval_wer": 0.3675755484585386,
+      "eval_loss": 0.5019811391830444,
+      "eval_runtime": 145.2988,
+      "eval_samples_per_second": 38.927,
+      "eval_steps_per_second": 4.866,
+      "eval_wer": 0.40500072218388405,
       "step": 63000
     },
     {
       "epoch": 2.4951864513340407,
-      "grad_norm": 0.7074434757232666,
-      "learning_rate": 0.00011012562814070351,
-      "loss": 0.2551,
+      "grad_norm": 1.4857794046401978,
+      "learning_rate": 0.0002053007518796992,
+      "loss": 0.3272,
       "step": 63500
     },
     {
       "epoch": 2.5148335887461197,
-      "grad_norm": 1.97989821434021,
-      "learning_rate": 0.0001086180904522613,
-      "loss": 0.2521,
+      "grad_norm": 0.9404523968696594,
+      "learning_rate": 0.0002045488721804511,
+      "loss": 0.3222,
       "step": 64000
     },
     {
       "epoch": 2.5148335887461197,
-      "eval_loss": 0.4576202929019928,
-      "eval_runtime": 143.8418,
-      "eval_samples_per_second": 39.321,
-      "eval_steps_per_second": 4.915,
-      "eval_wer": 0.35895748744202466,
+      "eval_loss": 0.5072047114372253,
+      "eval_runtime": 145.1717,
+      "eval_samples_per_second": 38.961,
+      "eval_steps_per_second": 4.87,
+      "eval_wer": 0.39964051291104297,
       "step": 64000
     },
     {
       "epoch": 2.5344807261581987,
-      "grad_norm": 0.7055521607398987,
-      "learning_rate": 0.00010711055276381909,
-      "loss": 0.2545,
+      "grad_norm": 1.1841472387313843,
+      "learning_rate": 0.00020379699248120297,
+      "loss": 0.3261,
       "step": 64500
     },
     {
       "epoch": 2.5541278635702778,
-      "grad_norm": 0.7131598591804504,
-      "learning_rate": 0.00010560603015075377,
-      "loss": 0.2415,
+      "grad_norm": 0.7141321301460266,
+      "learning_rate": 0.0002030466165413534,
+      "loss": 0.3138,
       "step": 65000
     },
     {
       "epoch": 2.5541278635702778,
-      "eval_loss": 0.4510234296321869,
-      "eval_runtime": 144.8155,
-      "eval_samples_per_second": 39.057,
-      "eval_steps_per_second": 4.882,
-      "eval_wer": 0.3541589767456789,
+      "eval_loss": 0.5098404884338379,
+      "eval_runtime": 144.9074,
+      "eval_samples_per_second": 39.032,
+      "eval_steps_per_second": 4.879,
+      "eval_wer": 0.40357240294651026,
       "step": 65000
     },
     {
       "epoch": 2.573775000982357,
-      "grad_norm": 2.750293254852295,
-      "learning_rate": 0.00010409849246231155,
-      "loss": 0.251,
+      "grad_norm": 0.8695092797279358,
+      "learning_rate": 0.00020229473684210523,
+      "loss": 0.3198,
       "step": 65500
     },
     {
       "epoch": 2.593422138394436,
-      "grad_norm": 2.111311435699463,
-      "learning_rate": 0.00010259095477386933,
-      "loss": 0.2349,
+      "grad_norm": 1.1630802154541016,
+      "learning_rate": 0.00020154285714285713,
+      "loss": 0.3074,
       "step": 66000
     },
     {
       "epoch": 2.593422138394436,
-      "eval_loss": 0.445352166891098,
-      "eval_runtime": 144.573,
-      "eval_samples_per_second": 39.122,
-      "eval_steps_per_second": 4.89,
-      "eval_wer": 0.35335655020782847,
+      "eval_loss": 0.5026105046272278,
+      "eval_runtime": 145.0532,
+      "eval_samples_per_second": 38.993,
+      "eval_steps_per_second": 4.874,
+      "eval_wer": 0.3981159024891271,
       "step": 66000
     },
     {
       "epoch": 2.613069275806515,
-      "grad_norm": 2.4952309131622314,
-      "learning_rate": 0.00010108643216080401,
-      "loss": 0.2467,
+      "grad_norm": 1.2439523935317993,
+      "learning_rate": 0.000200790977443609,
+      "loss": 0.3234,
       "step": 66500
     },
     {
       "epoch": 2.632716413218594,
-      "grad_norm": 0.967322587966919,
-      "learning_rate": 9.95788944723618e-05,
-      "loss": 0.2482,
+      "grad_norm": 0.7216903567314148,
+      "learning_rate": 0.0002000390977443609,
+      "loss": 0.3261,
       "step": 67000
     },
     {
       "epoch": 2.632716413218594,
-      "eval_loss": 0.4530712068080902,
-      "eval_runtime": 144.4196,
-      "eval_samples_per_second": 39.164,
-      "eval_steps_per_second": 4.895,
-      "eval_wer": 0.35855627417309943,
+      "eval_loss": 0.5030384063720703,
+      "eval_runtime": 145.8235,
+      "eval_samples_per_second": 38.787,
+      "eval_steps_per_second": 4.848,
+      "eval_wer": 0.39349392563110847,
       "step": 67000
     },
     {
       "epoch": 2.652363550630673,
-      "grad_norm": 2.702293634414673,
-      "learning_rate": 9.807135678391959e-05,
-      "loss": 0.2382,
+      "grad_norm": 1.0616713762283325,
+      "learning_rate": 0.00019928721804511276,
+      "loss": 0.3147,
       "step": 67500
     },
     {
       "epoch": 2.6720106880427523,
-      "grad_norm": 7.968295097351074,
-      "learning_rate": 9.656381909547737e-05,
-      "loss": 0.2527,
+      "grad_norm": 10.39274787902832,
+      "learning_rate": 0.00019853533834586465,
+      "loss": 0.3257,
       "step": 68000
     },
     {
       "epoch": 2.6720106880427523,
-      "eval_loss": 0.4417664110660553,
-      "eval_runtime": 144.8639,
-      "eval_samples_per_second": 39.044,
-      "eval_steps_per_second": 4.88,
-      "eval_wer": 0.3521529104010528,
+      "eval_loss": 0.500296413898468,
+      "eval_runtime": 144.7526,
+      "eval_samples_per_second": 39.074,
+      "eval_steps_per_second": 4.884,
+      "eval_wer": 0.39028421947970665,
       "step": 68000
     },
     {
       "epoch": 2.6916578254548313,
-      "grad_norm": 0.7025226950645447,
-      "learning_rate": 9.505628140703516e-05,
-      "loss": 0.2471,
+      "grad_norm": 0.7911710739135742,
+      "learning_rate": 0.00019778345864661652,
+      "loss": 0.3274,
       "step": 68500
     },
     {
       "epoch": 2.7113049628669104,
-      "grad_norm": 0.968565821647644,
-      "learning_rate": 9.354874371859296e-05,
-      "loss": 0.2473,
+      "grad_norm": 0.6936825513839722,
+      "learning_rate": 0.00019703157894736842,
+      "loss": 0.3179,
       "step": 69000
     },
     {
       "epoch": 2.7113049628669104,
-      "eval_loss": 0.44372445344924927,
-      "eval_runtime": 144.5528,
-      "eval_samples_per_second": 39.128,
-      "eval_steps_per_second": 4.891,
-      "eval_wer": 0.35826740061947326,
+      "eval_loss": 0.5139185786247253,
+      "eval_runtime": 145.2074,
+      "eval_samples_per_second": 38.951,
+      "eval_steps_per_second": 4.869,
+      "eval_wer": 0.4003947938566224,
       "step": 69000
     },
     {
       "epoch": 2.7309521002789894,
-      "grad_norm": 0.7599900364875793,
-      "learning_rate": 9.204422110552763e-05,
-      "loss": 0.2348,
+      "grad_norm": 0.5706244111061096,
+      "learning_rate": 0.00019627969924812028,
+      "loss": 0.3147,
       "step": 69500
     },
     {
       "epoch": 2.7505992376910684,
-      "grad_norm": 2.434638261795044,
-      "learning_rate": 9.053668341708543e-05,
-      "loss": 0.2334,
+      "grad_norm": 0.9982422590255737,
+      "learning_rate": 0.00019552932330827065,
+      "loss": 0.3154,
       "step": 70000
     },
     {
       "epoch": 2.7505992376910684,
-      "eval_loss": 0.43379753828048706,
-      "eval_runtime": 145.1622,
-      "eval_samples_per_second": 38.963,
-      "eval_steps_per_second": 4.87,
-      "eval_wer": 0.34566930397522105,
+      "eval_loss": 0.5041365027427673,
+      "eval_runtime": 144.6141,
+      "eval_samples_per_second": 39.111,
+      "eval_steps_per_second": 4.889,
+      "eval_wer": 0.39455312866107106,
       "step": 70000
     },
     {
       "epoch": 2.7702463751031474,
-      "grad_norm": 0.5358386635780334,
-      "learning_rate": 8.902914572864321e-05,
-      "loss": 0.2355,
+      "grad_norm": 0.6709697842597961,
+      "learning_rate": 0.00019477894736842104,
+      "loss": 0.3116,
       "step": 70500
     },
     {
       "epoch": 2.7898935125152264,
-      "grad_norm": 1.4655603170394897,
-      "learning_rate": 8.7521608040201e-05,
-      "loss": 0.2314,
+      "grad_norm": 1.2155810594558716,
+      "learning_rate": 0.0001940270676691729,
+      "loss": 0.3119,
       "step": 71000
     },
     {
       "epoch": 2.7898935125152264,
-      "eval_loss": 0.4286041557788849,
-      "eval_runtime": 144.4601,
-      "eval_samples_per_second": 39.153,
-      "eval_steps_per_second": 4.894,
-      "eval_wer": 0.34557301279067903,
+      "eval_loss": 0.49135103821754456,
+      "eval_runtime": 144.8705,
+      "eval_samples_per_second": 39.042,
+      "eval_steps_per_second": 4.88,
+      "eval_wer": 0.3940877212691178,
       "step": 71000
     },
     {
       "epoch": 2.8095406499273055,
-      "grad_norm": 1.992447018623352,
-      "learning_rate": 8.60140703517588e-05,
-      "loss": 0.2391,
+      "grad_norm": 9.424253463745117,
+      "learning_rate": 0.0001932766917293233,
+      "loss": 0.3177,
       "step": 71500
     },
     {
       "epoch": 2.829187787339385,
-      "grad_norm": 1.4909594058990479,
-      "learning_rate": 8.450653266331658e-05,
-      "loss": 0.2318,
+      "grad_norm": 0.6820365786552429,
+      "learning_rate": 0.00019252481203007517,
+      "loss": 0.3128,
       "step": 72000
     },
     {
       "epoch": 2.829187787339385,
-      "eval_loss": 0.42746907472610474,
-      "eval_runtime": 144.6523,
-      "eval_samples_per_second": 39.101,
-      "eval_steps_per_second": 4.888,
-      "eval_wer": 0.3371154370817352,
+      "eval_loss": 0.4867289066314697,
+      "eval_runtime": 144.797,
+      "eval_samples_per_second": 39.062,
+      "eval_steps_per_second": 4.883,
+      "eval_wer": 0.38309447770056654,
       "step": 72000
     },
     {
       "epoch": 2.848834924751464,
-      "grad_norm": 0.7300348281860352,
-      "learning_rate": 8.300201005025126e-05,
-      "loss": 0.2328,
+      "grad_norm": 0.6361156702041626,
+      "learning_rate": 0.00019177443609022553,
+      "loss": 0.3127,
       "step": 72500
     },
     {
       "epoch": 2.868482062163543,
-      "grad_norm": 0.5728158354759216,
-      "learning_rate": 8.149447236180904e-05,
-      "loss": 0.2347,
+      "grad_norm": 1.142830491065979,
+      "learning_rate": 0.00019102255639097743,
+      "loss": 0.3105,
       "step": 73000
     },
     {
       "epoch": 2.868482062163543,
-      "eval_loss": 0.42658358812332153,
-      "eval_runtime": 144.7899,
-      "eval_samples_per_second": 39.064,
-      "eval_steps_per_second": 4.883,
-      "eval_wer": 0.3408386962173613,
+      "eval_loss": 0.4870510995388031,
+      "eval_runtime": 145.1289,
+      "eval_samples_per_second": 38.972,
+      "eval_steps_per_second": 4.872,
+      "eval_wer": 0.3817945467092488,
       "step": 73000
     },
     {
       "epoch": 2.888129199575622,
-      "grad_norm": 0.7156108021736145,
-      "learning_rate": 7.998693467336682e-05,
-      "loss": 0.2421,
+      "grad_norm": 0.8262931704521179,
+      "learning_rate": 0.0001902706766917293,
+      "loss": 0.3234,
       "step": 73500
     },
     {
       "epoch": 2.907776336987701,
-      "grad_norm": 1.1177480220794678,
-      "learning_rate": 7.847939698492462e-05,
-      "loss": 0.2313,
+      "grad_norm": 1.1251935958862305,
+      "learning_rate": 0.0001895187969924812,
+      "loss": 0.309,
       "step": 74000
     },
     {
       "epoch": 2.907776336987701,
-      "eval_loss": 0.42381560802459717,
-      "eval_runtime": 144.7411,
-      "eval_samples_per_second": 39.077,
-      "eval_steps_per_second": 4.885,
-      "eval_wer": 0.33544638988300624,
+      "eval_loss": 0.48873621225357056,
+      "eval_runtime": 144.4047,
+      "eval_samples_per_second": 39.168,
+      "eval_steps_per_second": 4.896,
+      "eval_wer": 0.39885413490394955,
       "step": 74000
     },
     {
       "epoch": 2.92742347439978,
-      "grad_norm": 0.5893663763999939,
-      "learning_rate": 7.69718592964824e-05,
-      "loss": 0.2393,
+      "grad_norm": 0.825520396232605,
+      "learning_rate": 0.00018876691729323306,
+      "loss": 0.3204,
       "step": 74500
     },
     {
       "epoch": 2.947070611811859,
-      "grad_norm": 0.8663122057914734,
-      "learning_rate": 7.546733668341708e-05,
-      "loss": 0.2253,
+      "grad_norm": 0.5379465818405151,
+      "learning_rate": 0.00018801654135338345,
+      "loss": 0.3073,
       "step": 75000
     },
     {
       "epoch": 2.947070611811859,
-      "eval_loss": 0.41986486315727234,
-      "eval_runtime": 144.4042,
-      "eval_samples_per_second": 39.168,
-      "eval_steps_per_second": 4.896,
-      "eval_wer": 0.33164288809359505,
+      "eval_loss": 0.48394420742988586,
+      "eval_runtime": 145.2237,
+      "eval_samples_per_second": 38.947,
+      "eval_steps_per_second": 4.868,
+      "eval_wer": 0.3903644621334917,
       "step": 75000
     },
     {
       "epoch": 2.966717749223938,
-      "grad_norm": 0.48322752118110657,
-      "learning_rate": 7.395979899497487e-05,
-      "loss": 0.2222,
+      "grad_norm": 0.4377336800098419,
+      "learning_rate": 0.00018726466165413532,
+      "loss": 0.3044,
       "step": 75500
     },
     {
       "epoch": 2.986364886636017,
-      "grad_norm": 0.6252803206443787,
-      "learning_rate": 7.245226130653266e-05,
-      "loss": 0.217,
+      "grad_norm": 0.8088381290435791,
+      "learning_rate": 0.0001865127819548872,
+      "loss": 0.3023,
       "step": 76000
     },
     {
       "epoch": 2.986364886636017,
-      "eval_loss": 0.4222155809402466,
-      "eval_runtime": 144.6071,
-      "eval_samples_per_second": 39.113,
-      "eval_steps_per_second": 4.889,
-      "eval_wer": 0.33327983823081,
+      "eval_loss": 0.48391589522361755,
+      "eval_runtime": 145.3642,
+      "eval_samples_per_second": 38.909,
+      "eval_steps_per_second": 4.864,
+      "eval_wer": 0.38049461571793103,
       "step": 76000
     },
     {
       "epoch": 3.006012024048096,
-      "grad_norm": 0.9049218893051147,
-      "learning_rate": 7.094472361809045e-05,
-      "loss": 0.2258,
+      "grad_norm": 0.6517421007156372,
+      "learning_rate": 0.00018576090225563908,
+      "loss": 0.3049,
       "step": 76500
     },
     {
       "epoch": 3.025659161460175,
-      "grad_norm": 0.6731761693954468,
-      "learning_rate": 6.943718592964823e-05,
-      "loss": 0.194,
+      "grad_norm": 1.4399667978286743,
+      "learning_rate": 0.00018500902255639098,
+      "loss": 0.2715,
       "step": 77000
     },
     {
       "epoch": 3.025659161460175,
-      "eval_loss": 0.4252253770828247,
-      "eval_runtime": 145.6762,
-      "eval_samples_per_second": 38.826,
-      "eval_steps_per_second": 4.853,
-      "eval_wer": 0.3342267015454735,
+      "eval_loss": 0.48158180713653564,
+      "eval_runtime": 147.38,
+      "eval_samples_per_second": 38.377,
+      "eval_steps_per_second": 4.797,
+      "eval_wer": 0.38766830896631416,
       "step": 77000
     },
     {
       "epoch": 3.045306298872254,
-      "grad_norm": 3.186687707901001,
-      "learning_rate": 6.792964824120603e-05,
-      "loss": 0.1942,
+      "grad_norm": 1.0456621646881104,
+      "learning_rate": 0.00018425714285714284,
+      "loss": 0.2762,
       "step": 77500
     },
     {
       "epoch": 3.064953436284333,
-      "grad_norm": 1.0603581666946411,
-      "learning_rate": 6.642512562814069e-05,
-      "loss": 0.181,
+      "grad_norm": 0.6409999132156372,
+      "learning_rate": 0.00018350526315789474,
+      "loss": 0.2565,
       "step": 78000
     },
     {
       "epoch": 3.064953436284333,
-      "eval_loss": 0.42276450991630554,
-      "eval_runtime": 144.274,
-      "eval_samples_per_second": 39.203,
-      "eval_steps_per_second": 4.9,
-      "eval_wer": 0.33640930172842676,
+      "eval_loss": 0.49935096502304077,
+      "eval_runtime": 144.0075,
+      "eval_samples_per_second": 39.276,
+      "eval_steps_per_second": 4.909,
+      "eval_wer": 0.3811044598866974,
       "step": 78000
     },
     {
       "epoch": 3.0846005736964126,
-      "grad_norm": 0.5516805052757263,
-      "learning_rate": 6.492060301507537e-05,
-      "loss": 0.1871,
+      "grad_norm": 0.8721100687980652,
+      "learning_rate": 0.0001827548872180451,
+      "loss": 0.2681,
       "step": 78500
     },
     {
       "epoch": 3.1042477111084916,
-      "grad_norm": 0.4451453685760498,
-      "learning_rate": 6.341306532663317e-05,
-      "loss": 0.187,
+      "grad_norm": 1.0572487115859985,
+      "learning_rate": 0.00018200300751879697,
+      "loss": 0.2697,
       "step": 79000
     },
     {
       "epoch": 3.1042477111084916,
-      "eval_loss": 0.4196587800979614,
-      "eval_runtime": 144.9483,
-      "eval_samples_per_second": 39.021,
-      "eval_steps_per_second": 4.878,
-      "eval_wer": 0.3355908266598193,
+      "eval_loss": 0.48027363419532776,
+      "eval_runtime": 143.6893,
+      "eval_samples_per_second": 39.363,
+      "eval_steps_per_second": 4.92,
+      "eval_wer": 0.3813291393172955,
       "step": 79000
     },
     {
       "epoch": 3.1238948485205706,
-      "grad_norm": 0.48886096477508545,
-      "learning_rate": 6.190552763819095e-05,
-      "loss": 0.1894,
+      "grad_norm": 0.5640320777893066,
+      "learning_rate": 0.00018125112781954887,
+      "loss": 0.274,
       "step": 79500
     },
     {
       "epoch": 3.1435419859326497,
-      "grad_norm": 0.874115526676178,
-      "learning_rate": 6.039798994974873e-05,
-      "loss": 0.1855,
+      "grad_norm": 0.740835964679718,
+      "learning_rate": 0.00018049924812030073,
+      "loss": 0.2717,
       "step": 80000
     },
     {
       "epoch": 3.1435419859326497,
-      "eval_loss": 0.42146381735801697,
-      "eval_runtime": 144.3542,
-      "eval_samples_per_second": 39.181,
-      "eval_steps_per_second": 4.898,
-      "eval_wer": 0.3409831329941744,
+      "eval_loss": 0.48425012826919556,
+      "eval_runtime": 144.7996,
+      "eval_samples_per_second": 39.061,
+      "eval_steps_per_second": 4.883,
+      "eval_wer": 0.37988477154916467,
       "step": 80000
     },
     {
       "epoch": 3.1631891233447287,
-      "grad_norm": 0.5143038630485535,
-      "learning_rate": 5.8890452261306523e-05,
-      "loss": 0.1841,
+      "grad_norm": 0.4206051528453827,
+      "learning_rate": 0.00017974736842105263,
+      "loss": 0.2751,
       "step": 80500
     },
     {
       "epoch": 3.1828362607568077,
-      "grad_norm": 0.7534670829772949,
-      "learning_rate": 5.7382914572864314e-05,
-      "loss": 0.1886,
+      "grad_norm": 1.7560110092163086,
+      "learning_rate": 0.00017899548872180447,
+      "loss": 0.2738,
       "step": 81000
     },
     {
       "epoch": 3.1828362607568077,
-      "eval_loss": 0.4176616668701172,
-      "eval_runtime": 145.008,
-      "eval_samples_per_second": 39.005,
-      "eval_steps_per_second": 4.876,
-      "eval_wer": 0.33186756752419316,
+      "eval_loss": 0.4904831647872925,
+      "eval_runtime": 145.8617,
+      "eval_samples_per_second": 38.776,
+      "eval_steps_per_second": 4.847,
+      "eval_wer": 0.37967614064932353,
       "step": 81000
     },
     {
       "epoch": 3.2024833981688867,
-      "grad_norm": 0.7065553665161133,
-      "learning_rate": 5.5875376884422104e-05,
-      "loss": 0.1806,
+      "grad_norm": 0.5435498952865601,
+      "learning_rate": 0.00017824360902255637,
+      "loss": 0.2671,
       "step": 81500
     },
     {
       "epoch": 3.2221305355809657,
-      "grad_norm": 0.4848778247833252,
-      "learning_rate": 5.437085427135678e-05,
-      "loss": 0.1821,
+      "grad_norm": 0.8769587278366089,
+      "learning_rate": 0.00017749323308270673,
+      "loss": 0.2617,
       "step": 82000
     },
     {
       "epoch": 3.2221305355809657,
-      "eval_loss": 0.4128184914588928,
-      "eval_runtime": 145.0877,
-      "eval_samples_per_second": 38.983,
-      "eval_steps_per_second": 4.873,
-      "eval_wer": 0.32925165701080067,
+      "eval_loss": 0.4753645956516266,
+      "eval_runtime": 144.9825,
+      "eval_samples_per_second": 39.012,
+      "eval_steps_per_second": 4.876,
+      "eval_wer": 0.37282341801608065,
       "step": 82000
     },
     {
       "epoch": 3.2417776729930448,
-      "grad_norm": 0.9558472037315369,
-      "learning_rate": 5.286331658291457e-05,
-      "loss": 0.1802,
+      "grad_norm": 0.5920813083648682,
+      "learning_rate": 0.00017674135338345865,
+      "loss": 0.27,
       "step": 82500
     },
     {
       "epoch": 3.261424810405124,
-      "grad_norm": 0.5446112751960754,
-      "learning_rate": 5.1355778894472356e-05,
-      "loss": 0.1786,
+      "grad_norm": 0.4276420474052429,
+      "learning_rate": 0.0001759894736842105,
+      "loss": 0.2634,
       "step": 83000
     },
     {
       "epoch": 3.261424810405124,
-      "eval_loss": 0.41023358702659607,
-      "eval_runtime": 144.3156,
-      "eval_samples_per_second": 39.192,
-      "eval_steps_per_second": 4.899,
-      "eval_wer": 0.32260756527739887,
+      "eval_loss": 0.4729759693145752,
+      "eval_runtime": 144.8389,
+      "eval_samples_per_second": 39.05,
+      "eval_steps_per_second": 4.881,
+      "eval_wer": 0.3668052189822022,
       "step": 83000
     },
     {
       "epoch": 3.2810719478172032,
-      "grad_norm": 0.3519670367240906,
-      "learning_rate": 4.9848241206030146e-05,
-      "loss": 0.1877,
+      "grad_norm": 2.3023736476898193,
+      "learning_rate": 0.0001752375939849624,
+      "loss": 0.2771,
       "step": 83500
     },
     {
       "epoch": 3.3007190852292823,
-      "grad_norm": 0.7854357957839966,
-      "learning_rate": 4.834070351758794e-05,
-      "loss": 0.1758,
+      "grad_norm": 1.1624869108200073,
+      "learning_rate": 0.00017448571428571426,
+      "loss": 0.2648,
       "step": 84000
     },
     {
       "epoch": 3.3007190852292823,
-      "eval_loss": 0.4147268831729889,
-      "eval_runtime": 144.3652,
-      "eval_samples_per_second": 39.178,
-      "eval_steps_per_second": 4.897,
-      "eval_wer": 0.32637897000529603,
+      "eval_loss": 0.4768010377883911,
+      "eval_runtime": 144.7404,
+      "eval_samples_per_second": 39.077,
+      "eval_steps_per_second": 4.885,
+      "eval_wer": 0.3690520132881835,
       "step": 84000
     },
     {
       "epoch": 3.3203662226413613,
-      "grad_norm": 0.5795627236366272,
-      "learning_rate": 4.683316582914573e-05,
-      "loss": 0.185,
+      "grad_norm": 0.64561527967453,
+      "learning_rate": 0.00017373383458646615,
+      "loss": 0.2745,
       "step": 84500
     },
     {
       "epoch": 3.3400133600534403,
-      "grad_norm": 1.2406518459320068,
-      "learning_rate": 4.532562814070351e-05,
-      "loss": 0.171,
+      "grad_norm": 0.5857324600219727,
+      "learning_rate": 0.00017298195488721802,
+      "loss": 0.2567,
       "step": 85000
     },
     {
       "epoch": 3.3400133600534403,
-      "eval_loss": 0.41314879059791565,
-      "eval_runtime": 145.1379,
-      "eval_samples_per_second": 38.97,
-      "eval_steps_per_second": 4.871,
-      "eval_wer": 0.31997560623324933,
+      "eval_loss": 0.4812460243701935,
+      "eval_runtime": 145.4739,
+      "eval_samples_per_second": 38.88,
+      "eval_steps_per_second": 4.86,
+      "eval_wer": 0.37410730047664137,
       "step": 85000
     },
     {
       "epoch": 3.3596604974655193,
-      "grad_norm": 9.228507995605469,
-      "learning_rate": 4.38180904522613e-05,
-      "loss": 0.1754,
+      "grad_norm": 0.97500079870224,
+      "learning_rate": 0.00017223007518796991,
+      "loss": 0.2686,
       "step": 85500
     },
     {
       "epoch": 3.3793076348775983,
-      "grad_norm": 0.7298714518547058,
-      "learning_rate": 4.231356783919597e-05,
-      "loss": 0.1767,
+      "grad_norm": 0.6413397789001465,
+      "learning_rate": 0.00017147969924812028,
+      "loss": 0.2687,
       "step": 86000
     },
     {
       "epoch": 3.3793076348775983,
-      "eval_loss": 0.4097689688205719,
-      "eval_runtime": 145.5087,
-      "eval_samples_per_second": 38.871,
-      "eval_steps_per_second": 4.859,
-      "eval_wer": 0.31719921041228677,
+      "eval_loss": 0.46830272674560547,
+      "eval_runtime": 144.9613,
+      "eval_samples_per_second": 39.017,
+      "eval_steps_per_second": 4.877,
+      "eval_wer": 0.37160372967854793,
       "step": 86000
     },
     {
       "epoch": 3.3989547722896774,
-      "grad_norm": 1.8685766458511353,
-      "learning_rate": 4.080603015075376e-05,
-      "loss": 0.1738,
+      "grad_norm": 1.0018800497055054,
+      "learning_rate": 0.00017072781954887217,
+      "loss": 0.2595,
       "step": 86500
     },
     {
       "epoch": 3.4186019097017564,
-      "grad_norm": 0.5031083822250366,
-      "learning_rate": 3.929849246231156e-05,
-      "loss": 0.1804,
+      "grad_norm": 1.0536398887634277,
+      "learning_rate": 0.00016997593984962404,
+      "loss": 0.2757,
       "step": 87000
     },
     {
       "epoch": 3.4186019097017564,
-      "eval_loss": 0.40913575887680054,
-      "eval_runtime": 145.0065,
-      "eval_samples_per_second": 39.005,
-      "eval_steps_per_second": 4.876,
-      "eval_wer": 0.3209064210171559,
+      "eval_loss": 0.46901389956474304,
+      "eval_runtime": 144.4366,
+      "eval_samples_per_second": 39.159,
+      "eval_steps_per_second": 4.895,
+      "eval_wer": 0.37320858275424884,
       "step": 87000
     },
     {
       "epoch": 3.4382490471138354,
-      "grad_norm": 0.30902644991874695,
-      "learning_rate": 3.779095477386935e-05,
-      "loss": 0.178,
+      "grad_norm": 0.4486633837223053,
+      "learning_rate": 0.00016922406015037594,
+      "loss": 0.2655,
       "step": 87500
     },
     {
       "epoch": 3.4578961845259144,
-      "grad_norm": 0.7432768940925598,
-      "learning_rate": 3.6283417085427134e-05,
-      "loss": 0.1699,
+      "grad_norm": 0.6999643445014954,
+      "learning_rate": 0.0001684721804511278,
+      "loss": 0.2596,
       "step": 88000
     },
     {
       "epoch": 3.4578961845259144,
-      "eval_loss": 0.40437594056129456,
-      "eval_runtime": 144.7637,
-      "eval_samples_per_second": 39.071,
-      "eval_steps_per_second": 4.884,
-      "eval_wer": 0.3179374428271092,
+      "eval_loss": 0.47534072399139404,
+      "eval_runtime": 145.0031,
+      "eval_samples_per_second": 39.006,
+      "eval_steps_per_second": 4.876,
+      "eval_wer": 0.37824782141194974,
       "step": 88000
     },
     {
       "epoch": 3.4775433219379934,
-      "grad_norm": 0.6833881139755249,
-      "learning_rate": 3.477587939698492e-05,
-      "loss": 0.174,
+      "grad_norm": 0.9858837723731995,
+      "learning_rate": 0.0001677203007518797,
+      "loss": 0.2614,
       "step": 88500
     },
     {
       "epoch": 3.497190459350073,
-      "grad_norm": 0.38148313760757446,
-      "learning_rate": 3.3268341708542715e-05,
-      "loss": 0.1645,
+      "grad_norm": 0.5992431640625,
+      "learning_rate": 0.00016696842105263157,
+      "loss": 0.2589,
       "step": 89000
     },
     {
       "epoch": 3.497190459350073,
-      "eval_loss": 0.40410056710243225,
-      "eval_runtime": 145.5748,
-      "eval_samples_per_second": 38.853,
-      "eval_steps_per_second": 4.857,
-      "eval_wer": 0.3167498515510905,
+      "eval_loss": 0.4645041823387146,
+      "eval_runtime": 146.1245,
+      "eval_samples_per_second": 38.707,
+      "eval_steps_per_second": 4.838,
+      "eval_wer": 0.3691483044727255,
       "step": 89000
     },
     {
       "epoch": 3.516837596762152,
-      "grad_norm": 0.7338330149650574,
-      "learning_rate": 3.1763819095477385e-05,
-      "loss": 0.1765,
+      "grad_norm": 2.5524730682373047,
+      "learning_rate": 0.00016621804511278193,
+      "loss": 0.2724,
       "step": 89500
     },
     {
       "epoch": 3.536484734174231,
-      "grad_norm": 0.8941044211387634,
-      "learning_rate": 3.0256281407035173e-05,
-      "loss": 0.1707,
+      "grad_norm": 0.42577388882637024,
+      "learning_rate": 0.00016546616541353383,
+      "loss": 0.2627,
       "step": 90000
     },
     {
       "epoch": 3.536484734174231,
-      "eval_loss": 0.40083417296409607,
-      "eval_runtime": 145.469,
-      "eval_samples_per_second": 38.881,
-      "eval_steps_per_second": 4.86,
-      "eval_wer": 0.3202484312561185,
+      "eval_loss": 0.4689880907535553,
+      "eval_runtime": 146.114,
+      "eval_samples_per_second": 38.71,
+      "eval_steps_per_second": 4.839,
+      "eval_wer": 0.3675274028662676,
       "step": 90000
     },
     {
       "epoch": 3.55613187158631,
-      "grad_norm": 0.5399070382118225,
-      "learning_rate": 2.8748743718592963e-05,
-      "loss": 0.1767,
+      "grad_norm": 0.530576765537262,
+      "learning_rate": 0.0001647142857142857,
+      "loss": 0.2692,
       "step": 90500
     },
     {
       "epoch": 3.575779008998389,
-      "grad_norm": 0.8689693212509155,
-      "learning_rate": 2.724120603015075e-05,
-      "loss": 0.1838,
+      "grad_norm": 1.5638034343719482,
+      "learning_rate": 0.0001639624060150376,
+      "loss": 0.2804,
       "step": 91000
     },
     {
       "epoch": 3.575779008998389,
-      "eval_loss": 0.3981110751628876,
-      "eval_runtime": 144.9174,
-      "eval_samples_per_second": 39.029,
-      "eval_steps_per_second": 4.879,
-      "eval_wer": 0.3165412206512494,
+      "eval_loss": 0.46749356389045715,
+      "eval_runtime": 146.7363,
+      "eval_samples_per_second": 38.545,
+      "eval_steps_per_second": 4.818,
+      "eval_wer": 0.37420359166118344,
       "step": 91000
     },
     {
       "epoch": 3.595426146410468,
-      "grad_norm": 0.4749615788459778,
-      "learning_rate": 2.573366834170854e-05,
-      "loss": 0.1697,
+      "grad_norm": 0.7981226444244385,
+      "learning_rate": 0.00016321052631578946,
+      "loss": 0.2658,
       "step": 91500
     },
     {
       "epoch": 3.615073283822547,
-      "grad_norm": 0.7744316458702087,
-      "learning_rate": 2.4226130653266328e-05,
-      "loss": 0.1653,
+      "grad_norm": 0.4092627167701721,
+      "learning_rate": 0.00016245864661654135,
+      "loss": 0.2587,
       "step": 92000
     },
     {
       "epoch": 3.615073283822547,
-      "eval_loss": 0.3987283408641815,
-      "eval_runtime": 145.3191,
-      "eval_samples_per_second": 38.921,
-      "eval_steps_per_second": 4.865,
-      "eval_wer": 0.3132352233153055,
+      "eval_loss": 0.46739462018013,
+      "eval_runtime": 145.0739,
+      "eval_samples_per_second": 38.987,
+      "eval_steps_per_second": 4.873,
+      "eval_wer": 0.3593747492417069,
       "step": 92000
     },
     {
       "epoch": 3.634720421234626,
-      "grad_norm": 0.5036156177520752,
-      "learning_rate": 2.2718592964824118e-05,
-      "loss": 0.1708,
+      "grad_norm": 0.4350492060184479,
+      "learning_rate": 0.0001617067669172932,
+      "loss": 0.2664,
       "step": 92500
     },
     {
       "epoch": 3.654367558646705,
-      "grad_norm": 0.6102810502052307,
-      "learning_rate": 2.121105527638191e-05,
-      "loss": 0.1679,
+      "grad_norm": 0.7081959247589111,
+      "learning_rate": 0.00016095488721804512,
+      "loss": 0.2615,
       "step": 93000
     },
     {
       "epoch": 3.654367558646705,
-      "eval_loss": 0.39818885922431946,
-      "eval_runtime": 145.5258,
-      "eval_samples_per_second": 38.866,
-      "eval_steps_per_second": 4.858,
-      "eval_wer": 0.3110044775400812,
+      "eval_loss": 0.46574193239212036,
+      "eval_runtime": 145.5779,
+      "eval_samples_per_second": 38.852,
+      "eval_steps_per_second": 4.857,
+      "eval_wer": 0.36321034809263214,
       "step": 93000
     },
     {
       "epoch": 3.6740146960587845,
-      "grad_norm": 0.39582985639572144,
-      "learning_rate": 1.9703517587939696e-05,
-      "loss": 0.1726,
+      "grad_norm": 1.1376652717590332,
+      "learning_rate": 0.00016020300751879696,
+      "loss": 0.2664,
       "step": 93500
     },
     {
       "epoch": 3.6936618334708635,
-      "grad_norm": 0.4703328609466553,
-      "learning_rate": 1.8195979899497486e-05,
-      "loss": 0.1631,
+      "grad_norm": 0.8354430794715881,
+      "learning_rate": 0.00015945263157894738,
+      "loss": 0.2531,
       "step": 94000
     },
     {
       "epoch": 3.6936618334708635,
-      "eval_loss": 0.39044293761253357,
-      "eval_runtime": 145.0128,
-      "eval_samples_per_second": 39.003,
-      "eval_steps_per_second": 4.875,
-      "eval_wer": 0.3073775095889971,
+      "eval_loss": 0.45889467000961304,
+      "eval_runtime": 145.207,
+      "eval_samples_per_second": 38.951,
+      "eval_steps_per_second": 4.869,
+      "eval_wer": 0.3668373160437162,
       "step": 94000
     },
     {
       "epoch": 3.7133089708829425,
-      "grad_norm": 0.5020835399627686,
-      "learning_rate": 1.669145728643216e-05,
-      "loss": 0.1687,
+      "grad_norm": 0.7989226579666138,
+      "learning_rate": 0.00015870075187969922,
+      "loss": 0.2621,
       "step": 94500
     },
     {
       "epoch": 3.7329561082950216,
-      "grad_norm": 1.21161687374115,
-      "learning_rate": 1.5186934673366832e-05,
-      "loss": 0.1561,
+      "grad_norm": 1.2648522853851318,
+      "learning_rate": 0.00015794887218045114,
+      "loss": 0.2466,
       "step": 95000
     },
     {
       "epoch": 3.7329561082950216,
-      "eval_loss": 0.39342302083969116,
-      "eval_runtime": 148.1879,
-      "eval_samples_per_second": 38.168,
-      "eval_steps_per_second": 4.771,
-      "eval_wer": 0.30911075091075413,
+      "eval_loss": 0.46178776025772095,
+      "eval_runtime": 145.1044,
+      "eval_samples_per_second": 38.979,
+      "eval_steps_per_second": 4.872,
+      "eval_wer": 0.3691001588804545,
       "step": 95000
     },
     {
       "epoch": 3.7526032457071006,
-      "grad_norm": 0.49361735582351685,
-      "learning_rate": 1.3679396984924621e-05,
-      "loss": 0.1757,
+      "grad_norm": 0.6409050226211548,
+      "learning_rate": 0.00015719699248120298,
+      "loss": 0.2732,
       "step": 95500
     },
     {
       "epoch": 3.7722503831191796,
-      "grad_norm": 0.9406358003616333,
-      "learning_rate": 1.217185929648241e-05,
-      "loss": 0.1699,
+      "grad_norm": 0.8056377172470093,
+      "learning_rate": 0.00015644511278195487,
+      "loss": 0.2653,
       "step": 96000
     },
     {
       "epoch": 3.7722503831191796,
-      "eval_loss": 0.3916533589363098,
-      "eval_runtime": 147.0052,
-      "eval_samples_per_second": 38.475,
-      "eval_steps_per_second": 4.809,
-      "eval_wer": 0.3068158110125018,
+      "eval_loss": 0.46144935488700867,
+      "eval_runtime": 145.0964,
+      "eval_samples_per_second": 38.981,
+      "eval_steps_per_second": 4.873,
+      "eval_wer": 0.3774774919356133,
       "step": 96000
     },
     {
       "epoch": 3.7918975205312586,
-      "grad_norm": 0.6055164337158203,
-      "learning_rate": 1.0667336683417085e-05,
-      "loss": 0.1668,
+      "grad_norm": 0.6420221924781799,
+      "learning_rate": 0.00015569473684210524,
+      "loss": 0.267,
       "step": 96500
     },
     {
       "epoch": 3.8115446579433376,
-      "grad_norm": 0.946466863155365,
-      "learning_rate": 9.159798994974874e-06,
-      "loss": 0.1591,
+      "grad_norm": 1.7600951194763184,
+      "learning_rate": 0.0001549428571428571,
+      "loss": 0.2542,
       "step": 97000
     },
     {
       "epoch": 3.8115446579433376,
-      "eval_loss": 0.3917677402496338,
-      "eval_runtime": 142.9501,
-      "eval_samples_per_second": 39.566,
-      "eval_steps_per_second": 4.946,
-      "eval_wer": 0.30567636532875414,
+      "eval_loss": 0.4600285291671753,
+      "eval_runtime": 145.9311,
+      "eval_samples_per_second": 38.758,
+      "eval_steps_per_second": 4.845,
+      "eval_wer": 0.3726308356469965,
       "step": 97000
     },
     {
       "epoch": 3.8311917953554167,
-      "grad_norm": 9.30636215209961,
-      "learning_rate": 7.652261306532663e-06,
-      "loss": 0.1649,
+      "grad_norm": 7.725172519683838,
+      "learning_rate": 0.000154190977443609,
+      "loss": 0.2648,
       "step": 97500
     },
     {
       "epoch": 3.8508389327674957,
-      "grad_norm": 1.0511385202407837,
-      "learning_rate": 6.144723618090452e-06,
-      "loss": 0.1609,
+      "grad_norm": 0.9012848734855652,
+      "learning_rate": 0.00015343909774436087,
+      "loss": 0.2616,
       "step": 98000
     },
     {
       "epoch": 3.8508389327674957,
-      "eval_loss": 0.39084428548812866,
-      "eval_runtime": 143.2016,
-      "eval_samples_per_second": 39.497,
-      "eval_steps_per_second": 4.937,
-      "eval_wer": 0.3049541814446887,
+      "eval_loss": 0.4511352777481079,
+      "eval_runtime": 146.3391,
+      "eval_samples_per_second": 38.65,
+      "eval_steps_per_second": 4.831,
+      "eval_wer": 0.3660348895058657,
       "step": 98000
     },
     {
       "epoch": 3.8704860701795747,
-      "grad_norm": 0.48396241664886475,
-      "learning_rate": 4.637185929648241e-06,
-      "loss": 0.1486,
+      "grad_norm": 0.4818692207336426,
+      "learning_rate": 0.00015268721804511276,
+      "loss": 0.2429,
       "step": 98500
     },
     {
       "epoch": 3.8901332075916537,
-      "grad_norm": 0.4520857632160187,
-      "learning_rate": 3.12964824120603e-06,
-      "loss": 0.1675,
+      "grad_norm": 1.495732307434082,
+      "learning_rate": 0.00015193533834586463,
+      "loss": 0.2625,
       "step": 99000
     },
     {
       "epoch": 3.8901332075916537,
-      "eval_loss": 0.3901500105857849,
-      "eval_runtime": 143.3337,
-      "eval_samples_per_second": 39.46,
-      "eval_steps_per_second": 4.933,
-      "eval_wer": 0.3058849962285953,
+      "eval_loss": 0.4607318639755249,
+      "eval_runtime": 145.4992,
+      "eval_samples_per_second": 38.873,
+      "eval_steps_per_second": 4.859,
+      "eval_wer": 0.36436584230713676,
       "step": 99000
     },
     {
       "epoch": 3.9097803450037327,
-      "grad_norm": 0.444196879863739,
-      "learning_rate": 1.622110552763819e-06,
-      "loss": 0.1709,
+      "grad_norm": 0.5369844436645508,
+      "learning_rate": 0.00015118345864661653,
+      "loss": 0.2693,
       "step": 99500
     },
     {
       "epoch": 3.9294274824158117,
-      "grad_norm": 0.38674989342689514,
-      "learning_rate": 1.1457286432160803e-07,
-      "loss": 0.1666,
+      "grad_norm": 2.374652862548828,
+      "learning_rate": 0.0001504315789473684,
+      "loss": 0.2627,
       "step": 100000
     },
     {
       "epoch": 3.9294274824158117,
-      "eval_loss": 0.3890285789966583,
-      "eval_runtime": 143.9516,
-      "eval_samples_per_second": 39.291,
-      "eval_steps_per_second": 4.911,
-      "eval_wer": 0.3055961226749691,
+      "eval_loss": 0.4455793499946594,
+      "eval_runtime": 146.0128,
+      "eval_samples_per_second": 38.736,
+      "eval_steps_per_second": 4.842,
+      "eval_wer": 0.36256840686235176,
       "step": 100000
     },
     {
-      "epoch": 3.9294274824158117,
-      "step": 100000,
-      "total_flos": 1.24400740487767e+20,
-      "train_loss": 0.35405309791564943,
-      "train_runtime": 51737.6152,
-      "train_samples_per_second": 15.463,
-      "train_steps_per_second": 1.933
+      "epoch": 3.949074619827891,
+      "grad_norm": 1.0198256969451904,
+      "learning_rate": 0.0001496812030075188,
+      "loss": 0.2569,
+      "step": 100500
+    },
+    {
+      "epoch": 3.9687217572399702,
+      "grad_norm": 0.7093910574913025,
+      "learning_rate": 0.00014892932330827068,
+      "loss": 0.252,
+      "step": 101000
+    },
+    {
+      "epoch": 3.9687217572399702,
+      "eval_loss": 0.4579247534275055,
+      "eval_runtime": 145.1833,
+      "eval_samples_per_second": 38.958,
+      "eval_steps_per_second": 4.87,
+      "eval_wer": 0.36651634542857603,
+      "step": 101000
+    },
+    {
+      "epoch": 3.9883688946520492,
+      "grad_norm": 0.7897918820381165,
+      "learning_rate": 0.00014817744360902255,
+      "loss": 0.2528,
+      "step": 101500
+    },
+    {
+      "epoch": 4.008016032064128,
+      "grad_norm": 0.773681640625,
+      "learning_rate": 0.00014742556390977442,
+      "loss": 0.2489,
+      "step": 102000
+    },
+    {
+      "epoch": 4.008016032064128,
+      "eval_loss": 0.45104366540908813,
+      "eval_runtime": 145.9991,
+      "eval_samples_per_second": 38.74,
+      "eval_steps_per_second": 4.842,
+      "eval_wer": 0.36203880534737043,
+      "step": 102000
+    },
+    {
+      "epoch": 4.027663169476208,
+      "grad_norm": 0.6559247970581055,
+      "learning_rate": 0.0001466736842105263,
+      "loss": 0.222,
+      "step": 102500
+    },
+    {
+      "epoch": 4.047310306888287,
+      "grad_norm": 1.860120415687561,
+      "learning_rate": 0.00014592481203007517,
+      "loss": 0.2218,
+      "step": 103000
+    },
+    {
+      "epoch": 4.047310306888287,
+      "eval_loss": 0.4418700039386749,
+      "eval_runtime": 149.8176,
+      "eval_samples_per_second": 37.753,
+      "eval_steps_per_second": 4.719,
+      "eval_wer": 0.35350098698464155,
+      "step": 103000
+    },
+    {
+      "epoch": 4.066957444300366,
+      "grad_norm": 0.8769797682762146,
+      "learning_rate": 0.00014517293233082707,
+      "loss": 0.2218,
+      "step": 103500
+    },
+    {
+      "epoch": 4.086604581712445,
+      "grad_norm": 1.1328709125518799,
+      "learning_rate": 0.00014442105263157894,
+      "loss": 0.2211,
+      "step": 104000
+    },
+    {
+      "epoch": 4.086604581712445,
+      "eval_loss": 0.449856162071228,
+      "eval_runtime": 144.4159,
+      "eval_samples_per_second": 39.165,
+      "eval_steps_per_second": 4.896,
+      "eval_wer": 0.3771404727897161,
+      "step": 104000
+    },
+    {
+      "epoch": 4.106251719124524,
+      "grad_norm": 0.8746039271354675,
+      "learning_rate": 0.0001436706766917293,
+      "loss": 0.2188,
+      "step": 104500
+    },
+    {
+      "epoch": 4.125898856536603,
+      "grad_norm": 0.55832839012146,
+      "learning_rate": 0.0001429203007518797,
+      "loss": 0.2186,
+      "step": 105000
+    },
+    {
+      "epoch": 4.125898856536603,
+      "eval_loss": 0.4546278417110443,
+      "eval_runtime": 144.5427,
+      "eval_samples_per_second": 39.13,
+      "eval_steps_per_second": 4.891,
+      "eval_wer": 0.36601884097510873,
+      "step": 105000
+    },
+    {
+      "epoch": 4.145545993948682,
+      "grad_norm": 0.7782666087150574,
+      "learning_rate": 0.00014216842105263156,
+      "loss": 0.2184,
+      "step": 105500
+    },
+    {
+      "epoch": 4.165193131360761,
+      "grad_norm": 0.768484890460968,
+      "learning_rate": 0.00014141654135338346,
+      "loss": 0.2199,
+      "step": 106000
+    },
+    {
+      "epoch": 4.165193131360761,
+      "eval_loss": 0.4395730495452881,
+      "eval_runtime": 144.834,
+      "eval_samples_per_second": 39.052,
+      "eval_steps_per_second": 4.881,
+      "eval_wer": 0.35423921939946396,
+      "step": 106000
+    },
+    {
+      "epoch": 4.18484026877284,
+      "grad_norm": 0.5472589135169983,
+      "learning_rate": 0.00014066466165413532,
+      "loss": 0.2206,
+      "step": 106500
+    },
+    {
+      "epoch": 4.204487406184919,
+      "grad_norm": 0.4021967947483063,
+      "learning_rate": 0.0001399127819548872,
+      "loss": 0.2227,
+      "step": 107000
+    },
+    {
+      "epoch": 4.204487406184919,
+      "eval_loss": 0.4468631446361542,
+      "eval_runtime": 144.5271,
+      "eval_samples_per_second": 39.135,
+      "eval_steps_per_second": 4.892,
+      "eval_wer": 0.35748102261237985,
+      "step": 107000
+    },
+    {
+      "epoch": 4.224134543596998,
+      "grad_norm": 1.0301532745361328,
+      "learning_rate": 0.0001391609022556391,
+      "loss": 0.2292,
+      "step": 107500
+    },
+    {
+      "epoch": 4.243781681009077,
+      "grad_norm": 0.6561172008514404,
+      "learning_rate": 0.00013840902255639095,
+      "loss": 0.2212,
+      "step": 108000
+    },
+    {
+      "epoch": 4.243781681009077,
+      "eval_loss": 0.44032466411590576,
+      "eval_runtime": 144.5017,
+      "eval_samples_per_second": 39.141,
+      "eval_steps_per_second": 4.893,
+      "eval_wer": 0.3500826499333986,
+      "step": 108000
+    },
+    {
+      "epoch": 4.263428818421156,
+      "grad_norm": 0.7782973647117615,
+      "learning_rate": 0.00013765714285714285,
+      "loss": 0.218,
+      "step": 108500
+    },
+    {
+      "epoch": 4.283075955833235,
+      "grad_norm": 0.5677826404571533,
+      "learning_rate": 0.00013690526315789472,
+      "loss": 0.2182,
+      "step": 109000
+    },
+    {
+      "epoch": 4.283075955833235,
+      "eval_loss": 0.4507006108760834,
+      "eval_runtime": 144.3123,
+      "eval_samples_per_second": 39.193,
+      "eval_steps_per_second": 4.899,
+      "eval_wer": 0.3599364478182022,
+      "step": 109000
+    },
+    {
+      "epoch": 4.302723093245314,
+      "grad_norm": 0.48135581612586975,
+      "learning_rate": 0.0001361533834586466,
+      "loss": 0.2191,
+      "step": 109500
+    },
+    {
+      "epoch": 4.322370230657393,
+      "grad_norm": 0.686140775680542,
+      "learning_rate": 0.00013540150375939848,
+      "loss": 0.2212,
+      "step": 110000
+    },
+    {
+      "epoch": 4.322370230657393,
+      "eval_loss": 0.4435155391693115,
+      "eval_runtime": 144.5051,
+      "eval_samples_per_second": 39.14,
+      "eval_steps_per_second": 4.893,
+      "eval_wer": 0.3575612652661649,
+      "step": 110000
+    },
+    {
+      "epoch": 4.342017368069472,
+      "grad_norm": 2.3186769485473633,
+      "learning_rate": 0.00013464962406015038,
+      "loss": 0.2213,
+      "step": 110500
+    },
+    {
+      "epoch": 4.361664505481551,
+      "grad_norm": 2.254951238632202,
+      "learning_rate": 0.00013389774436090224,
+      "loss": 0.2211,
+      "step": 111000
+    },
+    {
+      "epoch": 4.361664505481551,
+      "eval_loss": 0.45138731598854065,
+      "eval_runtime": 144.5221,
+      "eval_samples_per_second": 39.136,
+      "eval_steps_per_second": 4.892,
+      "eval_wer": 0.36893967357288443,
+      "step": 111000
+    },
+    {
+      "epoch": 4.38131164289363,
+      "grad_norm": 0.5208560228347778,
+      "learning_rate": 0.0001331458646616541,
+      "loss": 0.2042,
+      "step": 111500
+    },
+    {
+      "epoch": 4.400958780305709,
+      "grad_norm": 0.7651325464248657,
+      "learning_rate": 0.0001323954887218045,
+      "loss": 0.2116,
+      "step": 112000
+    },
+    {
+      "epoch": 4.400958780305709,
+      "eval_loss": 0.44426095485687256,
+      "eval_runtime": 144.753,
+      "eval_samples_per_second": 39.073,
+      "eval_steps_per_second": 4.884,
+      "eval_wer": 0.35077273675595,
+      "step": 112000
+    },
+    {
+      "epoch": 4.420605917717788,
+      "grad_norm": 0.7976289987564087,
+      "learning_rate": 0.00013164360902255637,
+      "loss": 0.2213,
+      "step": 112500
+    },
+    {
+      "epoch": 4.440253055129867,
+      "grad_norm": 0.7153854966163635,
+      "learning_rate": 0.00013089172932330827,
+      "loss": 0.2218,
+      "step": 113000
+    },
+    {
+      "epoch": 4.440253055129867,
+      "eval_loss": 0.44099488854408264,
+      "eval_runtime": 145.5781,
+      "eval_samples_per_second": 38.852,
+      "eval_steps_per_second": 4.856,
+      "eval_wer": 0.3471618173356229,
+      "step": 113000
+    },
+    {
+      "epoch": 4.459900192541947,
+      "grad_norm": 0.8848706483840942,
+      "learning_rate": 0.00013014135338345863,
+      "loss": 0.2213,
+      "step": 113500
+    },
+    {
+      "epoch": 4.479547329954026,
+      "grad_norm": 0.590100109577179,
+      "learning_rate": 0.0001293894736842105,
+      "loss": 0.2152,
+      "step": 114000
+    },
+    {
+      "epoch": 4.479547329954026,
+      "eval_loss": 0.446841299533844,
+      "eval_runtime": 146.3889,
+      "eval_samples_per_second": 38.637,
+      "eval_steps_per_second": 4.83,
+      "eval_wer": 0.35348493845388457,
+      "step": 114000
+    },
+    {
+      "epoch": 4.499194467366105,
+      "grad_norm": 2.4068264961242676,
+      "learning_rate": 0.0001286375939849624,
+      "loss": 0.2149,
+      "step": 114500
+    },
+    {
+      "epoch": 4.518841604778184,
+      "grad_norm": 0.6972984671592712,
+      "learning_rate": 0.00012788571428571426,
+      "loss": 0.2174,
+      "step": 115000
+    },
+    {
+      "epoch": 4.518841604778184,
+      "eval_loss": 0.4498594105243683,
+      "eval_runtime": 145.1426,
+      "eval_samples_per_second": 38.969,
+      "eval_steps_per_second": 4.871,
+      "eval_wer": 0.3469692349665388,
+      "step": 115000
+    },
+    {
+      "epoch": 4.538488742190263,
+      "grad_norm": 0.6739790439605713,
+      "learning_rate": 0.00012713383458646616,
+      "loss": 0.2148,
+      "step": 115500
+    },
+    {
+      "epoch": 4.558135879602342,
+      "grad_norm": 4.946841716766357,
+      "learning_rate": 0.00012638195488721802,
+      "loss": 0.212,
+      "step": 116000
+    },
+    {
+      "epoch": 4.558135879602342,
+      "eval_loss": 0.4453933835029602,
+      "eval_runtime": 145.1072,
+      "eval_samples_per_second": 38.978,
+      "eval_steps_per_second": 4.872,
+      "eval_wer": 0.34401630530724914,
+      "step": 116000
+    },
+    {
+      "epoch": 4.577783017014421,
+      "grad_norm": 0.5079777240753174,
+      "learning_rate": 0.00012563007518796992,
+      "loss": 0.2097,
+      "step": 116500
+    },
+    {
+      "epoch": 4.5974301544265,
+      "grad_norm": 1.189431071281433,
+      "learning_rate": 0.0001248781954887218,
+      "loss": 0.2039,
+      "step": 117000
+    },
+    {
+      "epoch": 4.5974301544265,
+      "eval_loss": 0.4423506259918213,
+      "eval_runtime": 144.2129,
+      "eval_samples_per_second": 39.22,
+      "eval_steps_per_second": 4.902,
+      "eval_wer": 0.34892715571889393,
+      "step": 117000
+    },
+    {
+      "epoch": 4.617077291838579,
+      "grad_norm": 0.5739697813987732,
+      "learning_rate": 0.00012412781954887218,
+      "loss": 0.2137,
+      "step": 117500
+    },
+    {
+      "epoch": 4.636724429250658,
+      "grad_norm": 1.8628792762756348,
+      "learning_rate": 0.00012337593984962405,
+      "loss": 0.2073,
+      "step": 118000
+    },
+    {
+      "epoch": 4.636724429250658,
+      "eval_loss": 0.44371461868286133,
+      "eval_runtime": 144.8897,
+      "eval_samples_per_second": 39.037,
+      "eval_steps_per_second": 4.88,
+      "eval_wer": 0.3466161672898846,
+      "step": 118000
+    },
+    {
+      "epoch": 4.656371566662737,
+      "grad_norm": 0.6919093728065491,
+      "learning_rate": 0.00012262406015037594,
+      "loss": 0.2111,
+      "step": 118500
+    },
+    {
+      "epoch": 4.676018704074816,
+      "grad_norm": 0.6628223061561584,
+      "learning_rate": 0.00012187218045112781,
+      "loss": 0.2177,
+      "step": 119000
+    },
+    {
+      "epoch": 4.676018704074816,
+      "eval_loss": 0.43920648097991943,
+      "eval_runtime": 144.6466,
+      "eval_samples_per_second": 39.102,
+      "eval_steps_per_second": 4.888,
+      "eval_wer": 0.34218677280095006,
+      "step": 119000
+    },
+    {
+      "epoch": 4.695665841486895,
+      "grad_norm": 0.7294492721557617,
+      "learning_rate": 0.00012112030075187969,
+      "loss": 0.2154,
+      "step": 119500
+    },
+    {
+      "epoch": 4.715312978898974,
+      "grad_norm": 1.2088764905929565,
+      "learning_rate": 0.00012036842105263157,
+      "loss": 0.2121,
+      "step": 120000
+    },
+    {
+      "epoch": 4.715312978898974,
+      "eval_loss": 0.44427990913391113,
+      "eval_runtime": 144.8984,
+      "eval_samples_per_second": 39.034,
+      "eval_steps_per_second": 4.879,
+      "eval_wer": 0.34372743175362297,
+      "step": 120000
+    },
+    {
+      "epoch": 4.734960116311053,
+      "grad_norm": 0.3588174283504486,
+      "learning_rate": 0.00011961654135338345,
+      "loss": 0.2103,
+      "step": 120500
+    },
+    {
+      "epoch": 4.754607253723132,
+      "grad_norm": 0.5091924667358398,
+      "learning_rate": 0.00011886466165413532,
+      "loss": 0.2072,
+      "step": 121000
+    },
+    {
+      "epoch": 4.754607253723132,
+      "eval_loss": 0.42684319615364075,
+      "eval_runtime": 143.2476,
+      "eval_samples_per_second": 39.484,
+      "eval_steps_per_second": 4.936,
+      "eval_wer": 0.34615075989793137,
+      "step": 121000
+    },
+    {
+      "epoch": 4.774254391135211,
+      "grad_norm": 0.49059540033340454,
+      "learning_rate": 0.0001181127819548872,
+      "loss": 0.204,
+      "step": 121500
+    },
+    {
+      "epoch": 4.79390152854729,
+      "grad_norm": 0.3562159836292267,
+      "learning_rate": 0.00011736090225563909,
+      "loss": 0.2138,
+      "step": 122000
+    },
+    {
+      "epoch": 4.79390152854729,
+      "eval_loss": 0.4271770417690277,
+      "eval_runtime": 142.8263,
+      "eval_samples_per_second": 39.601,
+      "eval_steps_per_second": 4.95,
+      "eval_wer": 0.34318178170788466,
+      "step": 122000
+    },
+    {
+      "epoch": 4.813548665959369,
+      "grad_norm": 1.027219295501709,
+      "learning_rate": 0.00011660902255639097,
+      "loss": 0.1947,
+      "step": 122500
+    },
+    {
+      "epoch": 4.833195803371449,
+      "grad_norm": 0.5677986145019531,
+      "learning_rate": 0.00011585714285714285,
+      "loss": 0.2145,
+      "step": 123000
+    },
+    {
+      "epoch": 4.833195803371449,
+      "eval_loss": 0.43315112590789795,
+      "eval_runtime": 143.3445,
+      "eval_samples_per_second": 39.457,
+      "eval_steps_per_second": 4.932,
+      "eval_wer": 0.3453964789523519,
+      "step": 123000
+    },
+    {
+      "epoch": 4.852842940783528,
+      "grad_norm": 0.7301272749900818,
+      "learning_rate": 0.00011510676691729323,
+      "loss": 0.2019,
+      "step": 123500
+    },
+    {
+      "epoch": 4.872490078195607,
+      "grad_norm": 16.804716110229492,
+      "learning_rate": 0.0001143578947368421,
+      "loss": 0.2217,
+      "step": 124000
+    },
+    {
+      "epoch": 4.872490078195607,
+      "eval_loss": 0.42095693945884705,
+      "eval_runtime": 143.7819,
+      "eval_samples_per_second": 39.337,
+      "eval_steps_per_second": 4.917,
+      "eval_wer": 0.3391215034263613,
+      "step": 124000
+    },
+    {
+      "epoch": 4.892137215607686,
+      "grad_norm": 0.4827280640602112,
+      "learning_rate": 0.00011360601503759398,
+      "loss": 0.1994,
+      "step": 124500
+    },
+    {
+      "epoch": 4.911784353019765,
+      "grad_norm": 0.6648825407028198,
+      "learning_rate": 0.00011285413533834586,
+      "loss": 0.2069,
+      "step": 125000
+    },
+    {
+      "epoch": 4.911784353019765,
+      "eval_loss": 0.427772581577301,
+      "eval_runtime": 144.7524,
+      "eval_samples_per_second": 39.074,
+      "eval_steps_per_second": 4.884,
+      "eval_wer": 0.3376289900659595,
+      "step": 125000
+    },
+    {
+      "epoch": 4.931431490431844,
+      "grad_norm": 0.3194764256477356,
+      "learning_rate": 0.00011210225563909773,
+      "loss": 0.1946,
+      "step": 125500
+    },
+    {
+      "epoch": 4.951078627843923,
+      "grad_norm": 0.9185254573822021,
+      "learning_rate": 0.00011135187969924811,
+      "loss": 0.2068,
+      "step": 126000
+    },
+    {
+      "epoch": 4.951078627843923,
+      "eval_loss": 0.4216279685497284,
+      "eval_runtime": 143.9237,
+      "eval_samples_per_second": 39.299,
+      "eval_steps_per_second": 4.912,
+      "eval_wer": 0.33867214456516503,
+      "step": 126000
+    },
+    {
+      "epoch": 4.970725765256002,
+      "grad_norm": 0.4608317017555237,
+      "learning_rate": 0.00011059999999999998,
+      "loss": 0.2098,
+      "step": 126500
+    },
+    {
+      "epoch": 4.990372902668081,
+      "grad_norm": 0.7766122221946716,
+      "learning_rate": 0.00010984812030075186,
+      "loss": 0.2129,
+      "step": 127000
+    },
+    {
+      "epoch": 4.990372902668081,
+      "eval_loss": 0.42103302478790283,
+      "eval_runtime": 144.3261,
+      "eval_samples_per_second": 39.189,
+      "eval_steps_per_second": 4.899,
+      "eval_wer": 0.3361525252363146,
+      "step": 127000
+    },
+    {
+      "epoch": 5.01002004008016,
+      "grad_norm": 0.7110891342163086,
+      "learning_rate": 0.00010909624060150374,
+      "loss": 0.1932,
+      "step": 127500
+    },
+    {
+      "epoch": 5.0296671774922395,
+      "grad_norm": 0.5839011073112488,
+      "learning_rate": 0.00010834436090225562,
+      "loss": 0.1774,
+      "step": 128000
+    },
+    {
+      "epoch": 5.0296671774922395,
+      "eval_loss": 0.4340197741985321,
+      "eval_runtime": 144.3949,
+      "eval_samples_per_second": 39.17,
+      "eval_steps_per_second": 4.896,
+      "eval_wer": 0.3303590056330343,
+      "step": 128000
+    },
+    {
+      "epoch": 5.0493143149043185,
+      "grad_norm": 0.4871758222579956,
+      "learning_rate": 0.000107593984962406,
+      "loss": 0.1764,
+      "step": 128500
+    },
+    {
+      "epoch": 5.0689614523163975,
+      "grad_norm": 1.1092002391815186,
+      "learning_rate": 0.00010684210526315788,
+      "loss": 0.1705,
+      "step": 129000
+    },
+    {
+      "epoch": 5.0689614523163975,
+      "eval_loss": 0.44219356775283813,
+      "eval_runtime": 144.6388,
+      "eval_samples_per_second": 39.104,
+      "eval_steps_per_second": 4.888,
+      "eval_wer": 0.329877549710324,
+      "step": 129000
+    },
+    {
+      "epoch": 5.0886085897284765,
+      "grad_norm": 1.4170928001403809,
+      "learning_rate": 0.00010609022556390976,
+      "loss": 0.1799,
+      "step": 129500
+    },
+    {
+      "epoch": 5.1082557271405555,
+      "grad_norm": 0.5609749555587769,
+      "learning_rate": 0.00010533834586466164,
+      "loss": 0.1746,
+      "step": 130000
+    },
+    {
+      "epoch": 5.1082557271405555,
+      "eval_loss": 0.43062400817871094,
+      "eval_runtime": 144.8052,
+      "eval_samples_per_second": 39.059,
+      "eval_steps_per_second": 4.882,
+      "eval_wer": 0.3363451076053987,
+      "step": 130000
+    },
+    {
+      "epoch": 5.1279028645526346,
+      "grad_norm": 0.7241942882537842,
+      "learning_rate": 0.00010458646616541353,
+      "loss": 0.1719,
+      "step": 130500
+    },
+    {
+      "epoch": 5.147550001964714,
+      "grad_norm": 7.793860912322998,
+      "learning_rate": 0.00010383458646616541,
+      "loss": 0.1813,
+      "step": 131000
+    },
+    {
+      "epoch": 5.147550001964714,
+      "eval_loss": 0.41806095838546753,
+      "eval_runtime": 144.8895,
+      "eval_samples_per_second": 39.037,
+      "eval_steps_per_second": 4.88,
+      "eval_wer": 0.33342427500762306,
+      "step": 131000
+    },
+    {
+      "epoch": 5.167197139376793,
+      "grad_norm": 0.5914771556854248,
+      "learning_rate": 0.00010308270676691729,
+      "loss": 0.1799,
+      "step": 131500
+    },
+    {
+      "epoch": 5.186844276788872,
+      "grad_norm": 1.6738320589065552,
+      "learning_rate": 0.00010233082706766916,
+      "loss": 0.1729,
+      "step": 132000
+    },
+    {
+      "epoch": 5.186844276788872,
+      "eval_loss": 0.4319230020046234,
+      "eval_runtime": 144.7317,
+      "eval_samples_per_second": 39.079,
+      "eval_steps_per_second": 4.885,
+      "eval_wer": 0.336858660589623,
+      "step": 132000
+    },
+    {
+      "epoch": 5.206491414200951,
+      "grad_norm": 0.6387330889701843,
+      "learning_rate": 0.00010157894736842104,
+      "loss": 0.1682,
+      "step": 132500
+    },
+    {
+      "epoch": 5.22613855161303,
+      "grad_norm": 0.5514143705368042,
+      "learning_rate": 0.00010082706766917292,
+      "loss": 0.1777,
+      "step": 133000
+    },
+    {
+      "epoch": 5.22613855161303,
+      "eval_loss": 0.4189823567867279,
+      "eval_runtime": 145.1159,
+      "eval_samples_per_second": 38.976,
+      "eval_steps_per_second": 4.872,
+      "eval_wer": 0.33265394553128663,
+      "step": 133000
+    },
+    {
+      "epoch": 5.245785689025109,
+      "grad_norm": 0.49433717131614685,
+      "learning_rate": 0.00010007669172932331,
+      "loss": 0.1757,
+      "step": 133500
+    },
+    {
+      "epoch": 5.265432826437188,
+      "grad_norm": 14.663381576538086,
+      "learning_rate": 9.932481203007518e-05,
+      "loss": 0.18,
+      "step": 134000
+    },
+    {
+      "epoch": 5.265432826437188,
+      "eval_loss": 0.42281797528266907,
+      "eval_runtime": 145.0781,
+      "eval_samples_per_second": 38.986,
+      "eval_steps_per_second": 4.873,
+      "eval_wer": 0.33376129415352024,
+      "step": 134000
+    },
+    {
+      "epoch": 5.285079963849267,
+      "grad_norm": 0.3960479497909546,
+      "learning_rate": 9.857293233082706e-05,
+      "loss": 0.1773,
+      "step": 134500
+    },
+    {
+      "epoch": 5.304727101261347,
+      "grad_norm": 0.48836782574653625,
+      "learning_rate": 9.782105263157894e-05,
+      "loss": 0.1747,
+      "step": 135000
+    },
+    {
+      "epoch": 5.304727101261347,
+      "eval_loss": 0.4267714023590088,
+      "eval_runtime": 144.4858,
+      "eval_samples_per_second": 39.146,
+      "eval_steps_per_second": 4.893,
+      "eval_wer": 0.3322687807931184,
+      "step": 135000
+    },
+    {
+      "epoch": 5.324374238673426,
+      "grad_norm": 0.7414509654045105,
+      "learning_rate": 9.706917293233082e-05,
+      "loss": 0.1804,
+      "step": 135500
+    },
+    {
+      "epoch": 5.344021376085505,
+      "grad_norm": 0.3100612461566925,
+      "learning_rate": 9.63172932330827e-05,
+      "loss": 0.1737,
+      "step": 136000
+    },
+    {
+      "epoch": 5.344021376085505,
+      "eval_loss": 0.41930150985717773,
+      "eval_runtime": 145.0977,
+      "eval_samples_per_second": 38.981,
+      "eval_steps_per_second": 4.873,
+      "eval_wer": 0.3324774116929595,
+      "step": 136000
+    },
+    {
+      "epoch": 5.363668513497584,
+      "grad_norm": 2.2844786643981934,
+      "learning_rate": 9.556541353383459e-05,
+      "loss": 0.1779,
+      "step": 136500
+    },
+    {
+      "epoch": 5.383315650909663,
+      "grad_norm": 0.7908081412315369,
+      "learning_rate": 9.481503759398495e-05,
+      "loss": 0.1709,
+      "step": 137000
+    },
+    {
+      "epoch": 5.383315650909663,
+      "eval_loss": 0.4228932559490204,
+      "eval_runtime": 145.0454,
+      "eval_samples_per_second": 38.995,
+      "eval_steps_per_second": 4.874,
+      "eval_wer": 0.3278714833656979,
+      "step": 137000
+    },
+    {
+      "epoch": 5.402962788321742,
+      "grad_norm": 1.6749204397201538,
+      "learning_rate": 9.406315789473683e-05,
+      "loss": 0.1745,
+      "step": 137500
+    },
+    {
+      "epoch": 5.422609925733821,
+      "grad_norm": 0.25723955035209656,
+      "learning_rate": 9.331127819548871e-05,
+      "loss": 0.1726,
+      "step": 138000
+    },
+    {
+      "epoch": 5.422609925733821,
+      "eval_loss": 0.4178549647331238,
+      "eval_runtime": 145.1876,
+      "eval_samples_per_second": 38.957,
+      "eval_steps_per_second": 4.87,
+      "eval_wer": 0.32714929948163246,
+      "step": 138000
+    },
+    {
+      "epoch": 5.4422570631459,
+      "grad_norm": 0.43192166090011597,
+      "learning_rate": 9.255939849624058e-05,
+      "loss": 0.1699,
+      "step": 138500
+    },
+    {
+      "epoch": 5.461904200557979,
+      "grad_norm": 0.4252433776855469,
+      "learning_rate": 9.180751879699246e-05,
+      "loss": 0.1741,
+      "step": 139000
+    },
+    {
+      "epoch": 5.461904200557979,
+      "eval_loss": 0.42049652338027954,
+      "eval_runtime": 145.3425,
+      "eval_samples_per_second": 38.915,
+      "eval_steps_per_second": 4.864,
+      "eval_wer": 0.3254963008136605,
+      "step": 139000
+    },
+    {
+      "epoch": 5.481551337970058,
+      "grad_norm": 0.6398211717605591,
+      "learning_rate": 9.105563909774435e-05,
+      "loss": 0.1675,
+      "step": 139500
+    },
+    {
+      "epoch": 5.501198475382137,
+      "grad_norm": 2.678009510040283,
+      "learning_rate": 9.030375939849623e-05,
+      "loss": 0.1723,
+      "step": 140000
+    },
+    {
+      "epoch": 5.501198475382137,
+      "eval_loss": 0.4140247702598572,
+      "eval_runtime": 145.8316,
+      "eval_samples_per_second": 38.784,
+      "eval_steps_per_second": 4.848,
+      "eval_wer": 0.32944423937988476,
+      "step": 140000
+    },
+    {
+      "epoch": 5.520845612794216,
+      "grad_norm": 0.42189884185791016,
+      "learning_rate": 8.955187969924811e-05,
+      "loss": 0.167,
+      "step": 140500
+    },
+    {
+      "epoch": 5.540492750206295,
+      "grad_norm": 0.6850213408470154,
+      "learning_rate": 8.8803007518797e-05,
+      "loss": 0.1676,
+      "step": 141000
+    },
+    {
+      "epoch": 5.540492750206295,
+      "eval_loss": 0.42560333013534546,
+      "eval_runtime": 145.0938,
+      "eval_samples_per_second": 38.982,
+      "eval_steps_per_second": 4.873,
+      "eval_wer": 0.32540000962911847,
+      "step": 141000
+    },
+    {
+      "epoch": 5.560139887618374,
+      "grad_norm": 0.46668741106987,
+      "learning_rate": 8.805112781954888e-05,
+      "loss": 0.1674,
+      "step": 141500
+    },
+    {
+      "epoch": 5.579787025030453,
+      "grad_norm": 0.38750043511390686,
+      "learning_rate": 8.729924812030075e-05,
+      "loss": 0.1769,
+      "step": 142000
+    },
+    {
+      "epoch": 5.579787025030453,
+      "eval_loss": 0.41800424456596375,
+      "eval_runtime": 144.6116,
+      "eval_samples_per_second": 39.112,
+      "eval_steps_per_second": 4.889,
+      "eval_wer": 0.3279196289579689,
+      "step": 142000
+    },
+    {
+      "epoch": 5.599434162442532,
+      "grad_norm": 0.47452759742736816,
+      "learning_rate": 8.654887218045112e-05,
+      "loss": 0.1704,
+      "step": 142500
+    },
+    {
+      "epoch": 5.619081299854611,
+      "grad_norm": 1.3760634660720825,
+      "learning_rate": 8.579699248120299e-05,
+      "loss": 0.1718,
+      "step": 143000
+    },
+    {
+      "epoch": 5.619081299854611,
+      "eval_loss": 0.4158097207546234,
+      "eval_runtime": 144.8323,
+      "eval_samples_per_second": 39.052,
+      "eval_steps_per_second": 4.882,
+      "eval_wer": 0.3203928680329316,
+      "step": 143000
+    },
+    {
+      "epoch": 5.63872843726669,
+      "grad_norm": 1.2168941497802734,
+      "learning_rate": 8.504511278195487e-05,
+      "loss": 0.1763,
+      "step": 143500
+    },
+    {
+      "epoch": 5.658375574678769,
+      "grad_norm": 0.6660623550415039,
+      "learning_rate": 8.429323308270675e-05,
+      "loss": 0.1735,
+      "step": 144000
+    },
+    {
+      "epoch": 5.658375574678769,
+      "eval_loss": 0.41737955808639526,
+      "eval_runtime": 145.161,
+      "eval_samples_per_second": 38.964,
+      "eval_steps_per_second": 4.87,
+      "eval_wer": 0.3209385180786699,
+      "step": 144000
+    },
+    {
+      "epoch": 5.678022712090849,
+      "grad_norm": 0.7844908237457275,
+      "learning_rate": 8.354135338345864e-05,
+      "loss": 0.1696,
+      "step": 144500
+    },
+    {
+      "epoch": 5.697669849502928,
+      "grad_norm": 1.7285536527633667,
+      "learning_rate": 8.278947368421052e-05,
+      "loss": 0.1693,
+      "step": 145000
+    },
+    {
+      "epoch": 5.697669849502928,
+      "eval_loss": 0.416604220867157,
+      "eval_runtime": 143.979,
+      "eval_samples_per_second": 39.284,
+      "eval_steps_per_second": 4.91,
+      "eval_wer": 0.3197669753334082,
+      "step": 145000
+    },
+    {
+      "epoch": 5.717316986915007,
+      "grad_norm": 0.3506734073162079,
+      "learning_rate": 8.20375939849624e-05,
+      "loss": 0.1811,
+      "step": 145500
+    },
+    {
+      "epoch": 5.736964124327086,
+      "grad_norm": 0.9915302395820618,
+      "learning_rate": 8.128571428571428e-05,
+      "loss": 0.1745,
+      "step": 146000
+    },
+    {
+      "epoch": 5.736964124327086,
+      "eval_loss": 0.41646912693977356,
+      "eval_runtime": 143.6976,
+      "eval_samples_per_second": 39.36,
+      "eval_steps_per_second": 4.92,
+      "eval_wer": 0.32445314631445493,
+      "step": 146000
+    },
+    {
+      "epoch": 5.756611261739165,
+      "grad_norm": 0.4368499219417572,
+      "learning_rate": 8.053383458646616e-05,
+      "loss": 0.1757,
+      "step": 146500
+    },
+    {
+      "epoch": 5.776258399151244,
+      "grad_norm": 0.8709374070167542,
+      "learning_rate": 7.978195488721803e-05,
+      "loss": 0.1692,
+      "step": 147000
+    },
+    {
+      "epoch": 5.776258399151244,
+      "eval_loss": 0.4147648215293884,
+      "eval_runtime": 144.5484,
+      "eval_samples_per_second": 39.129,
+      "eval_steps_per_second": 4.891,
+      "eval_wer": 0.3230408756078381,
+      "step": 147000
+    },
+    {
+      "epoch": 5.795905536563323,
+      "grad_norm": 16.672887802124023,
+      "learning_rate": 7.903007518796991e-05,
+      "loss": 0.1633,
+      "step": 147500
+    },
+    {
+      "epoch": 5.815552673975402,
+      "grad_norm": 0.7690948247909546,
+      "learning_rate": 7.82781954887218e-05,
+      "loss": 0.1641,
+      "step": 148000
+    },
+    {
+      "epoch": 5.815552673975402,
+      "eval_loss": 0.4115670621395111,
+      "eval_runtime": 145.0143,
+      "eval_samples_per_second": 39.003,
+      "eval_steps_per_second": 4.875,
+      "eval_wer": 0.3216446534319783,
+      "step": 148000
+    },
+    {
+      "epoch": 5.835199811387481,
+      "grad_norm": 1.9833319187164307,
+      "learning_rate": 7.752781954887217e-05,
+      "loss": 0.1646,
+      "step": 148500
+    },
+    {
+      "epoch": 5.85484694879956,
+      "grad_norm": 0.38222184777259827,
+      "learning_rate": 7.677593984962405e-05,
+      "loss": 0.173,
+      "step": 149000
+    },
+    {
+      "epoch": 5.85484694879956,
+      "eval_loss": 0.40414321422576904,
+      "eval_runtime": 148.2393,
+      "eval_samples_per_second": 38.155,
+      "eval_steps_per_second": 4.769,
+      "eval_wer": 0.32366676830736146,
+      "step": 149000
+    },
+    {
+      "epoch": 5.874494086211639,
+      "grad_norm": 2.3978090286254883,
+      "learning_rate": 7.602556390977442e-05,
+      "loss": 0.1669,
+      "step": 149500
+    },
+    {
+      "epoch": 5.894141223623718,
+      "grad_norm": 0.7286165952682495,
+      "learning_rate": 7.52736842105263e-05,
+      "loss": 0.1664,
+      "step": 150000
+    },
+    {
+      "epoch": 5.894141223623718,
+      "eval_loss": 0.4038516581058502,
+      "eval_runtime": 145.7264,
+      "eval_samples_per_second": 38.812,
+      "eval_steps_per_second": 4.852,
+      "eval_wer": 0.3184349472805765,
+      "step": 150000
+    },
+    {
+      "epoch": 5.913788361035797,
+      "grad_norm": 0.6666128635406494,
+      "learning_rate": 7.45218045112782e-05,
+      "loss": 0.1631,
+      "step": 150500
+    },
+    {
+      "epoch": 5.933435498447876,
+      "grad_norm": 3.139840841293335,
+      "learning_rate": 7.376992481203008e-05,
+      "loss": 0.1648,
+      "step": 151000
+    },
+    {
+      "epoch": 5.933435498447876,
+      "eval_loss": 0.4072332978248596,
+      "eval_runtime": 144.1568,
+      "eval_samples_per_second": 39.235,
+      "eval_steps_per_second": 4.904,
+      "eval_wer": 0.31657331771276337,
+      "step": 151000
+    },
+    {
+      "epoch": 5.953082635859955,
+      "grad_norm": 0.2758707106113434,
+      "learning_rate": 7.301804511278196e-05,
+      "loss": 0.1616,
+      "step": 151500
+    },
+    {
+      "epoch": 5.972729773272034,
+      "grad_norm": 0.5328942537307739,
+      "learning_rate": 7.226616541353382e-05,
+      "loss": 0.1709,
+      "step": 152000
+    },
+    {
+      "epoch": 5.972729773272034,
+      "eval_loss": 0.40219077467918396,
+      "eval_runtime": 144.786,
+      "eval_samples_per_second": 39.065,
+      "eval_steps_per_second": 4.883,
+      "eval_wer": 0.3205854504020157,
+      "step": 152000
+    },
+    {
+      "epoch": 5.992376910684113,
+      "grad_norm": 0.5073242783546448,
+      "learning_rate": 7.15142857142857e-05,
+      "loss": 0.1651,
+      "step": 152500
+    },
+    {
+      "epoch": 6.012024048096192,
+      "grad_norm": 0.4045845866203308,
+      "learning_rate": 7.076390977443608e-05,
+      "loss": 0.151,
+      "step": 153000
+    },
+    {
+      "epoch": 6.012024048096192,
+      "eval_loss": 0.4034076929092407,
+      "eval_runtime": 144.9751,
+      "eval_samples_per_second": 39.014,
+      "eval_steps_per_second": 4.877,
+      "eval_wer": 0.31882011201874466,
+      "step": 153000
+    },
+    {
+      "epoch": 6.031671185508271,
+      "grad_norm": 1.1703969240188599,
+      "learning_rate": 7.001203007518797e-05,
+      "loss": 0.1397,
+      "step": 153500
+    },
+    {
+      "epoch": 6.05131832292035,
+      "grad_norm": 0.3152583837509155,
+      "learning_rate": 6.926015037593985e-05,
+      "loss": 0.1353,
+      "step": 154000
+    },
+    {
+      "epoch": 6.05131832292035,
+      "eval_loss": 0.41277533769607544,
+      "eval_runtime": 144.9149,
+      "eval_samples_per_second": 39.03,
+      "eval_steps_per_second": 4.879,
+      "eval_wer": 0.32572098024425866,
+      "step": 154000
+    },
+    {
+      "epoch": 6.070965460332429,
+      "grad_norm": 0.5021807551383972,
+      "learning_rate": 6.850827067669173e-05,
+      "loss": 0.1429,
+      "step": 154500
+    },
+    {
+      "epoch": 6.090612597744508,
+      "grad_norm": 0.4375011622905731,
+      "learning_rate": 6.77563909774436e-05,
+      "loss": 0.1476,
+      "step": 155000
+    },
+    {
+      "epoch": 6.090612597744508,
+      "eval_loss": 0.4197489619255066,
+      "eval_runtime": 145.056,
+      "eval_samples_per_second": 38.992,
+      "eval_steps_per_second": 4.874,
+      "eval_wer": 0.3200398003562774,
+      "step": 155000
+    },
+    {
+      "epoch": 6.110259735156587,
+      "grad_norm": 0.4859500527381897,
+      "learning_rate": 6.700451127819548e-05,
+      "loss": 0.1456,
+      "step": 155500
+    },
+    {
+      "epoch": 6.129906872568666,
+      "grad_norm": 0.4906657636165619,
+      "learning_rate": 6.625263157894736e-05,
+      "loss": 0.1465,
+      "step": 156000
+    },
+    {
+      "epoch": 6.129906872568666,
+      "eval_loss": 0.40734121203422546,
+      "eval_runtime": 144.5712,
+      "eval_samples_per_second": 39.123,
+      "eval_steps_per_second": 4.89,
+      "eval_wer": 0.3167338030203335,
+      "step": 156000
+    },
+    {
+      "epoch": 6.149554009980746,
+      "grad_norm": 0.7306200861930847,
+      "learning_rate": 6.550075187969924e-05,
+      "loss": 0.1414,
+      "step": 156500
+    },
+    {
+      "epoch": 6.169201147392825,
+      "grad_norm": 0.35837283730506897,
+      "learning_rate": 6.474887218045112e-05,
+      "loss": 0.139,
+      "step": 157000
+    },
+    {
+      "epoch": 6.169201147392825,
+      "eval_loss": 0.42275404930114746,
+      "eval_runtime": 144.9153,
+      "eval_samples_per_second": 39.03,
+      "eval_steps_per_second": 4.879,
+      "eval_wer": 0.321179246040025,
+      "step": 157000
+    },
+    {
+      "epoch": 6.188848284804904,
+      "grad_norm": 0.5820499658584595,
+      "learning_rate": 6.39984962406015e-05,
+      "loss": 0.1408,
+      "step": 157500
+    },
+    {
+      "epoch": 6.208495422216983,
+      "grad_norm": 0.2785002291202545,
+      "learning_rate": 6.324812030075188e-05,
+      "loss": 0.1404,
+      "step": 158000
+    },
+    {
+      "epoch": 6.208495422216983,
+      "eval_loss": 0.4117072522640228,
+      "eval_runtime": 144.7738,
+      "eval_samples_per_second": 39.068,
+      "eval_steps_per_second": 4.883,
+      "eval_wer": 0.3244691948452119,
+      "step": 158000
+    },
+    {
+      "epoch": 6.228142559629062,
+      "grad_norm": 0.9491069912910461,
+      "learning_rate": 6.249624060150375e-05,
+      "loss": 0.1443,
+      "step": 158500
+    },
+    {
+      "epoch": 6.247789697041141,
+      "grad_norm": 0.6151573657989502,
+      "learning_rate": 6.174436090225563e-05,
+      "loss": 0.1338,
+      "step": 159000
+    },
+    {
+      "epoch": 6.247789697041141,
+      "eval_loss": 0.41795113682746887,
+      "eval_runtime": 144.7948,
+      "eval_samples_per_second": 39.062,
+      "eval_steps_per_second": 4.883,
+      "eval_wer": 0.3153054837829597,
+      "step": 159000
+    },
+    {
+      "epoch": 6.26743683445322,
+      "grad_norm": 1.4104067087173462,
+      "learning_rate": 6.099248120300751e-05,
+      "loss": 0.1458,
+      "step": 159500
+    },
+    {
+      "epoch": 6.287083971865299,
+      "grad_norm": 0.4986151158809662,
+      "learning_rate": 6.024060150375939e-05,
+      "loss": 0.1436,
+      "step": 160000
+    },
+    {
+      "epoch": 6.287083971865299,
+      "eval_loss": 0.42644599080085754,
+      "eval_runtime": 145.4284,
+      "eval_samples_per_second": 38.892,
+      "eval_steps_per_second": 4.861,
+      "eval_wer": 0.31670170595881947,
+      "step": 160000
+    },
+    {
+      "epoch": 6.306731109277378,
+      "grad_norm": 1.0388261079788208,
+      "learning_rate": 5.9488721804511266e-05,
+      "loss": 0.1382,
+      "step": 160500
+    },
+    {
+      "epoch": 6.326378246689457,
+      "grad_norm": 1.0425645112991333,
+      "learning_rate": 5.873834586466165e-05,
+      "loss": 0.1317,
+      "step": 161000
+    },
+    {
+      "epoch": 6.326378246689457,
+      "eval_loss": 0.4117776155471802,
+      "eval_runtime": 144.9416,
+      "eval_samples_per_second": 39.023,
+      "eval_steps_per_second": 4.878,
+      "eval_wer": 0.31524128965993164,
+      "step": 161000
+    },
+    {
+      "epoch": 6.346025384101536,
+      "grad_norm": 0.47523021697998047,
+      "learning_rate": 5.798646616541353e-05,
+      "loss": 0.1386,
+      "step": 161500
+    },
+    {
+      "epoch": 6.365672521513615,
+      "grad_norm": 0.27745115756988525,
+      "learning_rate": 5.7234586466165414e-05,
+      "loss": 0.1395,
+      "step": 162000
+    },
+    {
+      "epoch": 6.365672521513615,
+      "eval_loss": 0.42685896158218384,
+      "eval_runtime": 145.206,
+      "eval_samples_per_second": 38.952,
+      "eval_steps_per_second": 4.869,
+      "eval_wer": 0.3118390011394457,
+      "step": 162000
+    },
+    {
+      "epoch": 6.385319658925694,
+      "grad_norm": 0.3224891126155853,
+      "learning_rate": 5.6484210526315785e-05,
+      "loss": 0.1335,
+      "step": 162500
+    },
+    {
+      "epoch": 6.404966796337773,
+      "grad_norm": 0.2714509665966034,
+      "learning_rate": 5.5732330827067666e-05,
+      "loss": 0.1267,
+      "step": 163000
+    },
+    {
+      "epoch": 6.404966796337773,
+      "eval_loss": 0.4240754544734955,
+      "eval_runtime": 144.9066,
+      "eval_samples_per_second": 39.032,
+      "eval_steps_per_second": 4.879,
+      "eval_wer": 0.31345990274590363,
+      "step": 163000
+    },
+    {
+      "epoch": 6.4246139337498525,
+      "grad_norm": 0.3742597997188568,
+      "learning_rate": 5.498045112781954e-05,
+      "loss": 0.1438,
+      "step": 163500
+    },
+    {
+      "epoch": 6.4442610711619315,
+      "grad_norm": 1.6135519742965698,
+      "learning_rate": 5.422857142857142e-05,
+      "loss": 0.1334,
+      "step": 164000
+    },
+    {
+      "epoch": 6.4442610711619315,
+      "eval_loss": 0.40579110383987427,
+      "eval_runtime": 144.6174,
+      "eval_samples_per_second": 39.11,
+      "eval_steps_per_second": 4.889,
+      "eval_wer": 0.31686219126638954,
+      "step": 164000
+    },
+    {
+      "epoch": 6.4639082085740105,
+      "grad_norm": 0.7605300545692444,
+      "learning_rate": 5.3476691729323304e-05,
+      "loss": 0.1371,
+      "step": 164500
+    },
+    {
+      "epoch": 6.4835553459860895,
+      "grad_norm": 0.44126906991004944,
+      "learning_rate": 5.2724812030075185e-05,
+      "loss": 0.1369,
+      "step": 165000
+    },
+    {
+      "epoch": 6.4835553459860895,
+      "eval_loss": 0.40502265095710754,
+      "eval_runtime": 145.4039,
+      "eval_samples_per_second": 38.899,
+      "eval_steps_per_second": 4.862,
+      "eval_wer": 0.31296239829243633,
+      "step": 165000
+    },
+    {
+      "epoch": 6.5032024833981685,
+      "grad_norm": 0.32450059056282043,
+      "learning_rate": 5.197293233082706e-05,
+      "loss": 0.1352,
+      "step": 165500
+    },
+    {
+      "epoch": 6.522849620810248,
+      "grad_norm": 1.38713538646698,
+      "learning_rate": 5.122105263157894e-05,
+      "loss": 0.1322,
+      "step": 166000
+    },
+    {
+      "epoch": 6.522849620810248,
+      "eval_loss": 0.40965744853019714,
+      "eval_runtime": 144.8647,
+      "eval_samples_per_second": 39.043,
+      "eval_steps_per_second": 4.88,
+      "eval_wer": 0.31403764985315596,
+      "step": 166000
+    },
+    {
+      "epoch": 6.5424967582223275,
+      "grad_norm": 0.7151561379432678,
+      "learning_rate": 5.046917293233082e-05,
+      "loss": 0.1385,
+      "step": 166500
+    },
+    {
+      "epoch": 6.5621438956344065,
+      "grad_norm": 0.46481749415397644,
+      "learning_rate": 4.9717293233082705e-05,
+      "loss": 0.1358,
+      "step": 167000
+    },
+    {
+      "epoch": 6.5621438956344065,
+      "eval_loss": 0.41421449184417725,
+      "eval_runtime": 144.9831,
+      "eval_samples_per_second": 39.011,
+      "eval_steps_per_second": 4.876,
+      "eval_wer": 0.3129142527001653,
+      "step": 167000
+    },
+    {
+      "epoch": 6.5817910330464855,
+      "grad_norm": 0.4189301133155823,
+      "learning_rate": 4.896541353383458e-05,
+      "loss": 0.1359,
+      "step": 167500
+    },
+    {
+      "epoch": 6.6014381704585645,
+      "grad_norm": 0.7608076333999634,
+      "learning_rate": 4.821353383458646e-05,
+      "loss": 0.1345,
+      "step": 168000
+    },
+    {
+      "epoch": 6.6014381704585645,
+      "eval_loss": 0.40090152621269226,
+      "eval_runtime": 144.6628,
+      "eval_samples_per_second": 39.098,
+      "eval_steps_per_second": 4.887,
+      "eval_wer": 0.31230440853139896,
+      "step": 168000
+    },
+    {
+      "epoch": 6.6210853078706435,
+      "grad_norm": 0.23644275963306427,
+      "learning_rate": 4.746165413533834e-05,
+      "loss": 0.1329,
+      "step": 168500
+    },
+    {
+      "epoch": 6.6407324452827226,
+      "grad_norm": 0.5338233709335327,
+      "learning_rate": 4.6711278195488714e-05,
+      "loss": 0.1321,
+      "step": 169000
+    },
+    {
+      "epoch": 6.6407324452827226,
+      "eval_loss": 0.4004514813423157,
+      "eval_runtime": 144.9848,
+      "eval_samples_per_second": 39.011,
+      "eval_steps_per_second": 4.876,
+      "eval_wer": 0.3092712362183242,
+      "step": 169000
+    },
+    {
+      "epoch": 6.660379582694802,
+      "grad_norm": 0.5386209487915039,
+      "learning_rate": 4.5959398496240595e-05,
+      "loss": 0.1324,
+      "step": 169500
+    },
+    {
+      "epoch": 6.680026720106881,
+      "grad_norm": 0.7969732880592346,
+      "learning_rate": 4.5207518796992477e-05,
+      "loss": 0.1299,
+      "step": 170000
+    },
+    {
+      "epoch": 6.680026720106881,
+      "eval_loss": 0.39957067370414734,
+      "eval_runtime": 144.9466,
+      "eval_samples_per_second": 39.021,
+      "eval_steps_per_second": 4.878,
+      "eval_wer": 0.305387491775128,
+      "step": 170000
+    },
+    {
+      "epoch": 6.69967385751896,
+      "grad_norm": 0.7069671154022217,
+      "learning_rate": 4.445563909774436e-05,
+      "loss": 0.1381,
+      "step": 170500
+    },
+    {
+      "epoch": 6.719320994931039,
+      "grad_norm": 0.8022767305374146,
+      "learning_rate": 4.370375939849623e-05,
+      "loss": 0.1345,
+      "step": 171000
+    },
+    {
+      "epoch": 6.719320994931039,
+      "eval_loss": 0.40409377217292786,
+      "eval_runtime": 145.3133,
+      "eval_samples_per_second": 38.923,
+      "eval_steps_per_second": 4.865,
+      "eval_wer": 0.30705653897385693,
+      "step": 171000
+    },
+    {
+      "epoch": 6.738968132343118,
+      "grad_norm": 0.9058027863502502,
+      "learning_rate": 4.2951879699248114e-05,
+      "loss": 0.1314,
+      "step": 171500
+    },
+    {
+      "epoch": 6.758615269755197,
+      "grad_norm": 0.4458518326282501,
+      "learning_rate": 4.2199999999999996e-05,
+      "loss": 0.1328,
+      "step": 172000
+    },
+    {
+      "epoch": 6.758615269755197,
+      "eval_loss": 0.3997325003147125,
+      "eval_runtime": 145.3079,
+      "eval_samples_per_second": 38.924,
+      "eval_steps_per_second": 4.866,
+      "eval_wer": 0.3069762963200719,
+      "step": 172000
+    },
+    {
+      "epoch": 6.778262407167276,
+      "grad_norm": 0.5749480128288269,
+      "learning_rate": 4.144812030075188e-05,
+      "loss": 0.135,
+      "step": 172500
+    },
+    {
+      "epoch": 6.797909544579355,
+      "grad_norm": 0.3367716073989868,
+      "learning_rate": 4.069624060150375e-05,
+      "loss": 0.1245,
+      "step": 173000
+    },
+    {
+      "epoch": 6.797909544579355,
+      "eval_loss": 0.3974212110042572,
+      "eval_runtime": 145.9176,
+      "eval_samples_per_second": 38.762,
+      "eval_steps_per_second": 4.845,
+      "eval_wer": 0.3044566769912215,
+      "step": 173000
+    },
+    {
+      "epoch": 6.817556681991434,
+      "grad_norm": 0.546405553817749,
+      "learning_rate": 3.9944360902255633e-05,
+      "loss": 0.1312,
+      "step": 173500
+    },
+    {
+      "epoch": 6.837203819403513,
+      "grad_norm": 0.38214609026908875,
+      "learning_rate": 3.9192481203007515e-05,
+      "loss": 0.1356,
+      "step": 174000
+    },
+    {
+      "epoch": 6.837203819403513,
+      "eval_loss": 0.39992156624794006,
+      "eval_runtime": 144.9546,
+      "eval_samples_per_second": 39.019,
+      "eval_steps_per_second": 4.877,
+      "eval_wer": 0.3008939031631654,
+      "step": 174000
+    },
+    {
+      "epoch": 6.856850956815592,
+      "grad_norm": 0.21237680315971375,
+      "learning_rate": 3.8442105263157886e-05,
+      "loss": 0.1335,
+      "step": 174500
+    },
+    {
+      "epoch": 6.876498094227671,
+      "grad_norm": 0.4656332731246948,
+      "learning_rate": 3.769022556390977e-05,
+      "loss": 0.1208,
+      "step": 175000
+    },
+    {
+      "epoch": 6.876498094227671,
+      "eval_loss": 0.39532560110092163,
+      "eval_runtime": 145.4346,
+      "eval_samples_per_second": 38.89,
+      "eval_steps_per_second": 4.861,
+      "eval_wer": 0.301921009131614,
+      "step": 175000
+    },
+    {
+      "epoch": 6.89614523163975,
+      "grad_norm": 0.6751464605331421,
+      "learning_rate": 3.693834586466165e-05,
+      "loss": 0.1282,
+      "step": 175500
+    },
+    {
+      "epoch": 6.915792369051829,
+      "grad_norm": 1.1535145044326782,
+      "learning_rate": 3.618646616541353e-05,
+      "loss": 0.1316,
+      "step": 176000
+    },
+    {
+      "epoch": 6.915792369051829,
+      "eval_loss": 0.39738306403160095,
+      "eval_runtime": 146.048,
+      "eval_samples_per_second": 38.727,
+      "eval_steps_per_second": 4.841,
+      "eval_wer": 0.3056442682672401,
+      "step": 176000
+    },
+    {
+      "epoch": 6.935439506463908,
+      "grad_norm": 0.8314586877822876,
+      "learning_rate": 3.543458646616541e-05,
+      "loss": 0.1271,
+      "step": 176500
+    },
+    {
+      "epoch": 6.955086643875987,
+      "grad_norm": 0.7973750233650208,
+      "learning_rate": 3.4682706766917294e-05,
+      "loss": 0.1232,
+      "step": 177000
+    },
+    {
+      "epoch": 6.955086643875987,
+      "eval_loss": 0.39205384254455566,
+      "eval_runtime": 146.1083,
+      "eval_samples_per_second": 38.711,
+      "eval_steps_per_second": 4.839,
+      "eval_wer": 0.30333327983823083,
+      "step": 177000
+    },
+    {
+      "epoch": 6.974733781288066,
+      "grad_norm": 0.3950521647930145,
+      "learning_rate": 3.393082706766917e-05,
+      "loss": 0.1344,
+      "step": 177500
+    },
+    {
+      "epoch": 6.994380918700146,
+      "grad_norm": 0.4100574851036072,
+      "learning_rate": 3.317894736842105e-05,
+      "loss": 0.1261,
+      "step": 178000
+    },
+    {
+      "epoch": 6.994380918700146,
+      "eval_loss": 0.39850306510925293,
+      "eval_runtime": 145.9873,
+      "eval_samples_per_second": 38.743,
+      "eval_steps_per_second": 4.843,
+      "eval_wer": 0.3034616680842869,
+      "step": 178000
+    },
+    {
+      "epoch": 7.014028056112225,
+      "grad_norm": 0.5746680498123169,
+      "learning_rate": 3.242857142857143e-05,
+      "loss": 0.1105,
+      "step": 178500
+    },
+    {
+      "epoch": 7.033675193524304,
+      "grad_norm": 0.5409959554672241,
+      "learning_rate": 3.16766917293233e-05,
+      "loss": 0.1184,
+      "step": 179000
+    },
+    {
+      "epoch": 7.033675193524304,
+      "eval_loss": 0.40056413412094116,
+      "eval_runtime": 145.9106,
+      "eval_samples_per_second": 38.763,
+      "eval_steps_per_second": 4.845,
+      "eval_wer": 0.3061096756591934,
+      "step": 179000
+    },
+    {
+      "epoch": 7.053322330936383,
+      "grad_norm": 0.7439139485359192,
+      "learning_rate": 3.092631578947368e-05,
+      "loss": 0.1132,
+      "step": 179500
+    },
+    {
+      "epoch": 7.072969468348462,
+      "grad_norm": 1.1852874755859375,
+      "learning_rate": 3.0174436090225562e-05,
+      "loss": 0.1115,
+      "step": 180000
+    },
+    {
+      "epoch": 7.072969468348462,
+      "eval_loss": 0.4096328318119049,
+      "eval_runtime": 145.6721,
+      "eval_samples_per_second": 38.827,
+      "eval_steps_per_second": 4.853,
+      "eval_wer": 0.3049541814446887,
+      "step": 180000
+    },
+    {
+      "epoch": 7.092616605760541,
+      "grad_norm": 0.6158276796340942,
+      "learning_rate": 2.9422556390977444e-05,
+      "loss": 0.1032,
+      "step": 180500
+    },
+    {
+      "epoch": 7.11226374317262,
+      "grad_norm": 1.272557258605957,
+      "learning_rate": 2.867067669172932e-05,
+      "loss": 0.1109,
+      "step": 181000
+    },
+    {
+      "epoch": 7.11226374317262,
+      "eval_loss": 0.41377753019332886,
+      "eval_runtime": 146.2393,
+      "eval_samples_per_second": 38.676,
+      "eval_steps_per_second": 4.835,
+      "eval_wer": 0.3038147357609411,
+      "step": 181000
+    },
+    {
+      "epoch": 7.131910880584699,
+      "grad_norm": 0.4577464163303375,
+      "learning_rate": 2.7918796992481203e-05,
+      "loss": 0.1157,
+      "step": 181500
+    },
+    {
+      "epoch": 7.151558017996778,
+      "grad_norm": 1.748535394668579,
+      "learning_rate": 2.716691729323308e-05,
+      "loss": 0.1113,
+      "step": 182000
+    },
+    {
+      "epoch": 7.151558017996778,
+      "eval_loss": 0.41194456815719604,
+      "eval_runtime": 146.2502,
+      "eval_samples_per_second": 38.673,
+      "eval_steps_per_second": 4.834,
+      "eval_wer": 0.3052270064675579,
+      "step": 182000
+    },
+    {
+      "epoch": 7.171205155408857,
+      "grad_norm": 0.8288829326629639,
+      "learning_rate": 2.6416541353383456e-05,
+      "loss": 0.1114,
+      "step": 182500
+    },
+    {
+      "epoch": 7.190852292820936,
+      "grad_norm": 0.5038288235664368,
+      "learning_rate": 2.5664661654135334e-05,
+      "loss": 0.1075,
+      "step": 183000
+    },
+    {
+      "epoch": 7.190852292820936,
+      "eval_loss": 0.41699934005737305,
+      "eval_runtime": 145.6145,
+      "eval_samples_per_second": 38.842,
+      "eval_steps_per_second": 4.855,
+      "eval_wer": 0.30066922373256727,
+      "step": 183000
+    },
+    {
+      "epoch": 7.210499430233015,
+      "grad_norm": 0.41699087619781494,
+      "learning_rate": 2.4912781954887215e-05,
+      "loss": 0.1155,
+      "step": 183500
+    },
+    {
+      "epoch": 7.230146567645094,
+      "grad_norm": 0.9346128702163696,
+      "learning_rate": 2.4162406015037593e-05,
+      "loss": 0.1081,
+      "step": 184000
+    },
+    {
+      "epoch": 7.230146567645094,
+      "eval_loss": 0.4134830832481384,
+      "eval_runtime": 145.6714,
+      "eval_samples_per_second": 38.827,
+      "eval_steps_per_second": 4.853,
+      "eval_wer": 0.3031246489383897,
+      "step": 184000
+    },
+    {
+      "epoch": 7.249793705057173,
+      "grad_norm": 1.0166319608688354,
+      "learning_rate": 2.341052631578947e-05,
+      "loss": 0.1173,
+      "step": 184500
+    },
+    {
+      "epoch": 7.269440842469252,
+      "grad_norm": 0.8515588045120239,
+      "learning_rate": 2.2658646616541353e-05,
+      "loss": 0.1108,
+      "step": 185000
+    },
+    {
+      "epoch": 7.269440842469252,
+      "eval_loss": 0.41293400526046753,
+      "eval_runtime": 146.3235,
+      "eval_samples_per_second": 38.654,
+      "eval_steps_per_second": 4.832,
+      "eval_wer": 0.3003161560559131,
+      "step": 185000
+    },
+    {
+      "epoch": 7.289087979881331,
+      "grad_norm": 0.5291551351547241,
+      "learning_rate": 2.190676691729323e-05,
+      "loss": 0.1064,
+      "step": 185500
+    },
+    {
+      "epoch": 7.30873511729341,
+      "grad_norm": 1.0743286609649658,
+      "learning_rate": 2.1154887218045113e-05,
+      "loss": 0.1044,
+      "step": 186000
+    },
+    {
+      "epoch": 7.30873511729341,
+      "eval_loss": 0.41300591826438904,
+      "eval_runtime": 145.5862,
+      "eval_samples_per_second": 38.85,
+      "eval_steps_per_second": 4.856,
+      "eval_wer": 0.3022740768082682,
+      "step": 186000
+    },
+    {
+      "epoch": 7.328382254705489,
+      "grad_norm": 0.4959530532360077,
+      "learning_rate": 2.040300751879699e-05,
+      "loss": 0.1063,
+      "step": 186500
+    },
+    {
+      "epoch": 7.348029392117568,
+      "grad_norm": 0.6196532845497131,
+      "learning_rate": 1.9651127819548872e-05,
+      "loss": 0.1121,
+      "step": 187000
+    },
+    {
+      "epoch": 7.348029392117568,
+      "eval_loss": 0.40789899230003357,
+      "eval_runtime": 145.8295,
+      "eval_samples_per_second": 38.785,
+      "eval_steps_per_second": 4.848,
+      "eval_wer": 0.2992890500874645,
+      "step": 187000
+    },
+    {
+      "epoch": 7.367676529529647,
+      "grad_norm": 1.7419555187225342,
+      "learning_rate": 1.889924812030075e-05,
+      "loss": 0.1092,
+      "step": 187500
+    },
+    {
+      "epoch": 7.387323666941727,
+      "grad_norm": 0.931948721408844,
+      "learning_rate": 1.814736842105263e-05,
+      "loss": 0.1052,
+      "step": 188000
+    },
+    {
+      "epoch": 7.387323666941727,
+      "eval_loss": 0.40476053953170776,
+      "eval_runtime": 145.5337,
+      "eval_samples_per_second": 38.864,
+      "eval_steps_per_second": 4.858,
+      "eval_wer": 0.301904960600857,
+      "step": 188000
+    },
+    {
+      "epoch": 7.406970804353806,
+      "grad_norm": 0.3558327853679657,
+      "learning_rate": 1.739548872180451e-05,
+      "loss": 0.112,
+      "step": 188500
+    },
+    {
+      "epoch": 7.426617941765885,
+      "grad_norm": 0.48914971947669983,
+      "learning_rate": 1.6643609022556388e-05,
+      "loss": 0.103,
+      "step": 189000
+    },
+    {
+      "epoch": 7.426617941765885,
+      "eval_loss": 0.415385365486145,
+      "eval_runtime": 145.9476,
+      "eval_samples_per_second": 38.754,
+      "eval_steps_per_second": 4.844,
+      "eval_wer": 0.3015197958626888,
+      "step": 189000
+    },
+    {
+      "epoch": 7.446265079177964,
+      "grad_norm": 0.5291373133659363,
+      "learning_rate": 1.5893233082706766e-05,
+      "loss": 0.1073,
+      "step": 189500
+    },
+    {
+      "epoch": 7.465912216590043,
+      "grad_norm": 0.6397764086723328,
+      "learning_rate": 1.514285714285714e-05,
+      "loss": 0.1105,
+      "step": 190000
+    },
+    {
+      "epoch": 7.465912216590043,
+      "eval_loss": 0.4119686484336853,
+      "eval_runtime": 145.6307,
+      "eval_samples_per_second": 38.838,
+      "eval_steps_per_second": 4.855,
+      "eval_wer": 0.30187286353934295,
+      "step": 190000
+    },
+    {
+      "epoch": 7.485559354002122,
+      "grad_norm": 0.45867177844047546,
+      "learning_rate": 1.439097744360902e-05,
+      "loss": 0.1079,
+      "step": 190500
+    },
+    {
+      "epoch": 7.505206491414201,
+      "grad_norm": 1.0139355659484863,
+      "learning_rate": 1.36390977443609e-05,
+      "loss": 0.1093,
+      "step": 191000
+    },
+    {
+      "epoch": 7.505206491414201,
+      "eval_loss": 0.4104667901992798,
+      "eval_runtime": 146.3292,
+      "eval_samples_per_second": 38.653,
+      "eval_steps_per_second": 4.832,
+      "eval_wer": 0.3007494663863523,
+      "step": 191000
+    },
+    {
+      "epoch": 7.52485362882628,
+      "grad_norm": 0.35021767020225525,
+      "learning_rate": 1.288721804511278e-05,
+      "loss": 0.1108,
+      "step": 191500
+    },
+    {
+      "epoch": 7.544500766238359,
+      "grad_norm": 0.7307072281837463,
+      "learning_rate": 1.2136842105263156e-05,
+      "loss": 0.1058,
+      "step": 192000
+    },
+    {
+      "epoch": 7.544500766238359,
+      "eval_loss": 0.41022607684135437,
+      "eval_runtime": 146.2108,
+      "eval_samples_per_second": 38.684,
+      "eval_steps_per_second": 4.835,
+      "eval_wer": 0.3011025340630065,
+      "step": 192000
+    },
+    {
+      "epoch": 7.564147903650438,
+      "grad_norm": 0.46207743883132935,
+      "learning_rate": 1.1384962406015036e-05,
+      "loss": 0.1053,
+      "step": 192500
+    },
+    {
+      "epoch": 7.583795041062517,
+      "grad_norm": 0.47636836767196655,
+      "learning_rate": 1.0633082706766916e-05,
+      "loss": 0.1043,
+      "step": 193000
+    },
+    {
+      "epoch": 7.583795041062517,
+      "eval_loss": 0.41014641523361206,
+      "eval_runtime": 145.8628,
+      "eval_samples_per_second": 38.776,
+      "eval_steps_per_second": 4.847,
+      "eval_wer": 0.2994495353950346,
+      "step": 193000
+    },
+    {
+      "epoch": 7.603442178474596,
+      "grad_norm": 1.0540902614593506,
+      "learning_rate": 9.881203007518796e-06,
+      "loss": 0.1072,
+      "step": 193500
+    },
+    {
+      "epoch": 7.623089315886675,
+      "grad_norm": 0.8974863290786743,
+      "learning_rate": 9.129323308270676e-06,
+      "loss": 0.1098,
+      "step": 194000
+    },
+    {
+      "epoch": 7.623089315886675,
+      "eval_loss": 0.408490389585495,
+      "eval_runtime": 146.1703,
+      "eval_samples_per_second": 38.695,
+      "eval_steps_per_second": 4.837,
+      "eval_wer": 0.29980260307168877,
+      "step": 194000
+    },
+    {
+      "epoch": 7.642736453298754,
+      "grad_norm": 0.49042123556137085,
+      "learning_rate": 8.377443609022555e-06,
+      "loss": 0.1035,
+      "step": 194500
+    },
+    {
+      "epoch": 7.662383590710833,
+      "grad_norm": 0.7251204252243042,
+      "learning_rate": 7.625563909774436e-06,
+      "loss": 0.1057,
+      "step": 195000
+    },
+    {
+      "epoch": 7.662383590710833,
+      "eval_loss": 0.40715456008911133,
+      "eval_runtime": 146.2248,
+      "eval_samples_per_second": 38.68,
+      "eval_steps_per_second": 4.835,
+      "eval_wer": 0.2982137985267449,
+      "step": 195000
+    },
+    {
+      "epoch": 7.682030728122912,
+      "grad_norm": 0.9783725142478943,
+      "learning_rate": 6.8751879699248115e-06,
+      "loss": 0.1078,
+      "step": 195500
+    },
+    {
+      "epoch": 7.701677865534991,
+      "grad_norm": 0.66826331615448,
+      "learning_rate": 6.123308270676691e-06,
+      "loss": 0.1021,
+      "step": 196000
+    },
+    {
+      "epoch": 7.701677865534991,
+      "eval_loss": 0.4079470634460449,
+      "eval_runtime": 146.5661,
+      "eval_samples_per_second": 38.59,
+      "eval_steps_per_second": 4.824,
+      "eval_wer": 0.2973792749273804,
+      "step": 196000
+    },
+    {
+      "epoch": 7.72132500294707,
+      "grad_norm": 0.34865960478782654,
+      "learning_rate": 5.371428571428571e-06,
+      "loss": 0.108,
+      "step": 196500
+    },
+    {
+      "epoch": 7.740972140359149,
+      "grad_norm": 0.6881831884384155,
+      "learning_rate": 4.619548872180451e-06,
+      "loss": 0.0994,
+      "step": 197000
+    },
+    {
+      "epoch": 7.740972140359149,
+      "eval_loss": 0.4088830053806305,
+      "eval_runtime": 145.6213,
+      "eval_samples_per_second": 38.84,
+      "eval_steps_per_second": 4.855,
+      "eval_wer": 0.29871130298021215,
+      "step": 197000
+    },
+    {
+      "epoch": 7.760619277771228,
+      "grad_norm": 0.7812435030937195,
+      "learning_rate": 3.867669172932331e-06,
+      "loss": 0.1017,
+      "step": 197500
+    },
+    {
+      "epoch": 7.780266415183307,
+      "grad_norm": 0.23445354402065277,
+      "learning_rate": 3.118796992481203e-06,
+      "loss": 0.1065,
+      "step": 198000
+    },
+    {
+      "epoch": 7.780266415183307,
+      "eval_loss": 0.4065949022769928,
+      "eval_runtime": 146.17,
+      "eval_samples_per_second": 38.695,
+      "eval_steps_per_second": 4.837,
+      "eval_wer": 0.2973792749273804,
+      "step": 198000
+    },
+    {
+      "epoch": 7.799913552595386,
+      "grad_norm": 0.4873931407928467,
+      "learning_rate": 2.366917293233083e-06,
+      "loss": 0.1052,
+      "step": 198500
+    },
+    {
+      "epoch": 7.8195606900074655,
+      "grad_norm": 0.24652531743049622,
+      "learning_rate": 1.6150375939849622e-06,
+      "loss": 0.1111,
+      "step": 199000
+    },
+    {
+      "epoch": 7.8195606900074655,
+      "eval_loss": 0.40712785720825195,
+      "eval_runtime": 145.6053,
+      "eval_samples_per_second": 38.845,
+      "eval_steps_per_second": 4.856,
+      "eval_wer": 0.2981817014652309,
+      "step": 199000
+    },
+    {
+      "epoch": 7.839207827419545,
+      "grad_norm": 0.5532709956169128,
+      "learning_rate": 8.631578947368421e-07,
+      "loss": 0.106,
+      "step": 199500
+    },
+    {
+      "epoch": 7.858854964831624,
+      "grad_norm": 0.4496346116065979,
+      "learning_rate": 1.1127819548872179e-07,
+      "loss": 0.1065,
+      "step": 200000
+    },
+    {
+      "epoch": 7.858854964831624,
+      "eval_loss": 0.4064280092716217,
+      "eval_runtime": 146.2071,
+      "eval_samples_per_second": 38.685,
+      "eval_steps_per_second": 4.836,
+      "eval_wer": 0.298422429426586,
+      "step": 200000
+    },
+    {
+      "epoch": 7.858854964831624,
+      "step": 200000,
+      "total_flos": 2.4880981924796708e+20,
+      "train_loss": 0.2853552089881897,
+      "train_runtime": 103513.8909,
+      "train_samples_per_second": 15.457,
+      "train_steps_per_second": 1.932
     }
   ],
   "logging_steps": 500,
-  "max_steps": 100000,
+  "max_steps": 200000,
   "num_input_tokens_seen": 0,
-  "num_train_epochs": 4,
+  "num_train_epochs": 8,
   "save_steps": 4000,
   "stateful_callbacks": {
     "TrainerControl": {
@@ -2335,7 +4635,7 @@
       "attributes": {}
     }
   },
-  "total_flos": 1.24400740487767e+20,
+  "total_flos": 2.4880981924796708e+20,
   "train_batch_size": 8,
   "trial_name": null,
   "trial_params": null