learning_rate: 5e-05
train_batch_size: 4
eval_batch_size: 4
seed: 42
gradient_accumulation_steps: 4
total_train_batch_size: 16
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: 500
training_steps: 20000

Training Loss	Epoch	Step	Validation Loss	Wer
1.3588	1.7688	500	1.3248	70.7742
1.0466	3.5350	1000	1.0844	78.4901
0.9356	5.3012	1500	0.9137	60.2303
0.8315	7.0673	2000	0.8622	64.0051
0.7841	8.8361	2500	0.8494	60.0128
0.7189	10.6023	3000	0.8459	64.2099
0.7026	12.3685	3500	0.8459	58.0422
0.6357	14.1346	4000	0.8465	59.2450
0.6708	15.9035	4500	0.8541	58.4133
0.6868	17.6696	5000	0.8520	58.5413

chengyili2005/whisper-medium-spanish-english-lora

Whisper Medium Spanish-English - Chengyi Li