learning_rate: 5e-06
train_batch_size: 8
eval_batch_size: 8
seed: 42
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: 1000
num_epochs: 25
mixed_precision_training: Native AMP

Training Loss	Epoch	Step	Validation Loss	Cer	Wer
2.1281	1.0	1796	0.6781	22.4463	39.5604
0.8481	2.0	3592	0.6186	22.8864	39.1632
0.6635	3.0	5388	0.5958	22.3883	36.0063
0.5452	4.0	7184	0.5845	19.1954	33.5890
0.4535	5.0	8980	0.5879	20.0356	33.8971
0.3794	6.0	10776	0.5924	17.1588	30.0007
0.3151	7.0	12572	0.5959	16.4786	29.2885
0.2661	8.0	14368	0.6140	16.1885	28.5078
0.2212	9.0	16164	0.6200	16.4586	28.9393
0.1849	10.0	17960	0.6326	16.5526	28.8708
0.1538	11.0	19756	0.6528	15.8364	27.8984
0.1296	12.0	21552	0.6647	15.8504	28.2339
0.1087	13.0	23348	0.6783	15.6224	27.6245
0.0908	14.0	25144	0.7006	15.9265	28.1928
0.0778	15.0	26940	0.7058	15.7464	27.8778
0.0656	16.0	28736	0.7182	15.7324	27.7820
0.0568	17.0	30532	0.7351	15.9005	28.0353
0.0487	18.0	32328	0.7429	15.6484	27.6861

NgQuocThai/whisper-medium-30s

whisper-medium-30s