learning_rate: 3e-05
train_batch_size: 4
eval_batch_size: 1
seed: 42
gradient_accumulation_steps: 8
total_train_batch_size: 32
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: cosine
lr_scheduler_warmup_steps: 2000
num_epochs: 4
mixed_precision_training: Native AMP

Training Loss	Epoch	Step	Validation Loss	Cer
1.7209	0.2561	2000	1.1863	0.2136
0.6811	0.5122	4000	0.7557	0.1281
0.5756	0.7682	6000	0.7370	0.1129
0.5212	1.0242	8000	0.7361	0.1062
0.4786	1.2803	10000	0.7466	0.0992
0.4445	1.5364	12000	0.7502	0.0936
0.4405	1.7924	14000	0.7623	0.0906
0.4073	2.0484	16000	0.7781	0.0872
0.3998	2.3045	18000	0.7862	0.0843
0.3858	2.5606	20000	0.7904	0.0831
0.3744	2.8166	22000	0.7849	0.0814
0.3628	3.0726	24000	0.7940	0.0800
0.3622	3.3287	26000	0.7972	0.0789
0.3599	3.5848	28000	0.7915	0.0788
0.3636	3.8408	30000	0.7929	0.0787

Miniijune/wav2vec2-xls-r-300m-Korean-children-pronunciation-jamo-based-V2

wav2vec2-xls-r-300m-Korean-children-pronunciation-jamo-based-V2