learning_rate: 0.0005
train_batch_size: 1
eval_batch_size: 6
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 2
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: 100
num_epochs: 5
mixed_precision_training: Native AMP

Training Loss	Epoch	Step	Validation Loss	Cer	Wer
0.7489	0.3019	200	0.6951	0.1857	0.5369
0.7537	0.6038	400	0.7556	0.1931	0.5521
0.7548	0.9057	600	0.7266	0.1930	0.5591
0.6896	1.2068	800	0.7408	0.1888	0.5317
0.74	1.5087	1000	0.7624	0.1876	0.5322
0.722	1.8106	1200	0.7312	0.1878	0.5284
0.6754	2.1117	1400	0.6957	0.1859	0.5425
0.6713	2.4136	1600	0.7077	0.1875	0.5366
0.6927	2.7155	1800	0.6985	0.1858	0.5273
0.6448	3.0166	2000	0.6932	0.1839	0.5253
0.6077	3.3185	2200	0.7229	0.1831	0.5187
0.6141	3.6204	2400	0.6936	0.1826	0.5220
0.6388	3.9223	2600	0.7130	0.1826	0.5196
0.604	4.2234	2800	0.6911	0.1817	0.5238
0.6071	4.5253	3000	0.6963	0.1818	0.5174
0.5633	4.8272	3200	0.6904	0.1816	0.5178

ctaguchi/ssc-rwm-mms-model-mix-adapt-max3-devtrain

ssc-rwm-mms-model-mix-adapt-max3-devtrain