learning_rate: 0.0003
train_batch_size: 8
eval_batch_size: 8
seed: 42
gradient_accumulation_steps: 4
total_train_batch_size: 32
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: 500
num_epochs: 30
mixed_precision_training: Native AMP

Training Loss	Epoch	Step	Validation Loss	Wer
7.3909	1.8779	400	3.4785	1.0
2.2874	3.7559	800	0.6656	0.6008
0.4781	5.6338	1200	0.3375	0.3276
0.2687	7.5117	1600	0.3546	0.3121
0.1769	9.3897	2000	0.3014	0.2611
0.1289	11.2676	2400	0.3079	0.2530
0.1064	13.1455	2800	0.3233	0.2405
0.0906	15.0235	3200	0.3141	0.2301
0.0755	16.9014	3600	0.3233	0.2301
0.0672	18.7793	4000	0.3374	0.2314
0.0588	20.6573	4400	0.3314	0.2169
0.052	22.5352	4800	0.3442	0.2153
0.0457	24.4131	5200	0.3441	0.2120
0.0409	26.2911	5600	0.3522	0.2096
0.0369	28.1690	6000	0.3501	0.2040

Mezge/wav2vec2-large-xls-r-300m-AmhEng

wav2vec2-large-xls-r-300m-AmhEng