learning_rate: 0.001
train_batch_size: 16
eval_batch_size: 16
seed: 42
optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED 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: 25
num_epochs: 12
mixed_precision_training: Native AMP

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
3.7064	0.3876	50	0.5495	0.6254	0.1407
0.5873	0.7752	100	0.4181	0.5107	0.1042
0.487	1.1628	150	0.3632	0.4721	0.0919
0.4432	1.5504	200	0.3473	0.4594	0.0871
0.4441	1.9380	250	0.3235	0.4311	0.0810
0.4042	2.3256	300	0.3350	0.4287	0.0819
0.4132	2.7132	350	0.3321	0.4221	0.0777
0.3955	3.1008	400	0.3061	0.4156	0.0760
0.3767	3.4884	450	0.3039	0.4123	0.0754
0.3823	3.8760	500	0.2925	0.3947	0.0711
0.3385	4.2636	550	0.2915	0.3836	0.0705
0.3486	4.6512	600	0.2788	0.3672	0.0667
0.3576	5.0388	650	0.2865	0.3795	0.0678
0.3176	5.4264	700	0.2763	0.3648	0.0649
0.337	5.8140	750	0.2780	0.3689	0.0658
0.3342	6.2016	800	0.2707	0.3566	0.0644
0.306	6.5891	850	0.2735	0.3529	0.0623
0.3297	6.9767	900	0.2671	0.3418	0.0618
0.2926	7.3643	950	0.2669	0.3480	0.0616
0.3273	7.7519	1000	0.2657	0.3578	0.0640

tadgeis/mms-1b-ckt-staged-bible-radio-to-chuklang-source-eval-stage1

mms-1b-ckt-staged-bible-radio-to-chuklang-source-eval-stage1