learning_rate: 0.0003
train_batch_size: 32
eval_batch_size: 8
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 64
optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 1200
num_epochs: 45
mixed_precision_training: Native AMP

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
3.3927	2.0101	1200	inf	0.7921	0.3049
0.8527	4.0201	2400	inf	0.7935	0.3785
0.6921	6.0302	3600	inf	0.6959	0.2665
0.5897	8.0402	4800	inf	0.6303	0.2240
0.5151	10.0503	6000	inf	0.6588	0.2535
0.4529	12.0604	7200	inf	0.6340	0.2450
0.4	14.0704	8400	inf	0.6539	0.2383
0.357	16.0805	9600	inf	0.6479	0.2327
0.3155	18.0905	10800	inf	0.6432	0.2248
0.277	20.1006	12000	inf	0.6441	0.2361
0.2475	22.1106	13200	inf	0.6541	0.2431
0.2249	24.1207	14400	inf	0.6653	0.2500
0.2011	26.1308	15600	inf	0.6421	0.2463
0.1802	28.1408	16800	inf	0.6387	0.2323
0.1649	30.1509	18000	inf	0.6155	0.2262
0.1501	32.1609	19200	inf	0.6381	0.2331
0.1356	34.1710	20400	inf	0.6422	0.2345
0.1224	36.1811	21600	inf	0.6306	0.2305
0.1103	38.1911	22800	inf	0.6339	0.2294
0.1014	40.2012	24000	inf	0.6311	0.2275
0.0934	42.2112	25200	inf	0.6258	0.2244
0.0875	44.2213	26400	inf	0.6281	0.2261

kaarthu2003/IEEEAccess_Previous

wav2vec2-IEEEAccess-FinalRun