learning_rate: 1e-05
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
0.9389	0.1994	1000	0.9970	0.9766	0.8008
0.8663	0.3989	2000	0.9427	0.6904	0.5569
0.8929	0.5983	3000	0.9267	0.6321	0.5140
0.7371	0.7978	4000	0.8962	0.7803	0.7050
0.7111	0.9972	5000	0.8790	0.5932	0.4867
0.5275	1.1966	6000	0.8841	0.5782	0.4390
0.5699	1.3961	7000	0.8773	0.6604	0.5288
0.5308	1.5955	8000	0.8682	0.5300	0.4096
0.5061	1.7950	9000	0.8599	0.5820	0.4616
0.5071	1.9944	10000	0.8542	0.5609	0.4449
0.3288	2.1939	11000	0.8972	0.4607	0.3564
0.3188	2.3933	12000	0.8936	0.4290	0.3281
0.3095	2.5927	13000	0.8997	0.4600	0.3552

HouraMor/wh-ft-lre5-adm-ga1ba16-st15k-pat3-v2

wh-ft-lre5-adm-ga1ba16-st15k-pat3-v2