learning_rate: 1e-06
train_batch_size: 16
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: 750
training_steps: 15000
mixed_precision_training: Native AMP

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.6297	0.3047	500	0.5828	0.2892	0.2118
0.5115	0.6094	1000	0.5015	0.2713	0.2022
0.4822	0.9141	1500	0.4747	0.2612	0.1963
0.4142	1.2188	2000	0.4649	0.2427	0.1738
0.4002	1.5235	2500	0.4579	0.2336	0.1747
0.4238	1.8282	3000	0.4510	0.2306	0.1722
0.3308	2.1328	3500	0.4535	0.2229	0.1629
0.3061	2.4375	4000	0.4523	0.2370	0.1748
0.3301	2.7422	4500	0.4478	0.2266	0.1663
0.2867	3.0469	5000	0.4540	0.2457	0.1822
0.2672	3.3516	5500	0.4616	0.2449	0.1811
0.2822	3.6563	6000	0.4612	0.2256	0.1658

HouraMor/wh-ft-lr1e6-dtstf5-adm-ga1ba16-st15k-pat3-v2-evalstp500

wh-ft-lr1e6-dtstf5-adm-ga1ba16-st15k-pat3-v2-evalstp500