learning_rate: 5e-06
train_batch_size: 4
eval_batch_size: 8
seed: 42
optimizer: Use OptimizerNames.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: 500
num_epochs: 25
mixed_precision_training: Native AMP

Training Loss	Epoch	Step	Validation Loss	Cer	Wer
1.879	1.0	2697	0.7137	26.9809	44.6664
0.9114	2.0	5394	0.6623	28.2322	44.8033
0.7286	3.0	8091	0.6491	23.1257	37.7489
0.6074	4.0	10788	0.6287	19.7227	33.1919
0.5108	5.0	13485	0.6356	19.7446	33.8488
0.4354	6.0	16182	0.6505	19.1947	32.7403
0.3691	7.0	18879	0.6518	21.5776	34.8820
0.3144	8.0	21576	0.6746	18.5412	31.6182
0.268	9.0	24273	0.6791	18.9337	31.8508
0.2296	10.0	26970	0.6913	16.9632	29.4287
0.1976	11.0	29667	0.7053	17.9933	31.0571
0.1678	12.0	32364	0.7112	17.2979	29.7503
0.1432	13.0	35061	0.7255	16.0088	28.1697
0.1239	14.0	37758	0.7477	17.0788	29.5792
0.1062	15.0	40455	0.7648	16.3535	28.6965
0.0934	16.0	43152	0.7731	16.5667	28.8950
0.082	17.0	45849	0.7804	16.5886	28.9702
0.0726	18.0	48546	0.7936	16.7181	29.1481

NgQuocThai/whisper-medium-korean-finetuned

whisper-medium-korean-finetuned