learning_rate: 1e-05
train_batch_size: 4
eval_batch_size: 8
seed: 42
gradient_accumulation_steps: 4
total_train_batch_size: 16
optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: cosine
lr_scheduler_warmup_ratio: 0.1
num_epochs: 20
mixed_precision_training: Native AMP

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
1.7466	0.5010	64	1.1037	96.4176	46.9276
0.9332	1.0	128	0.7805	83.2139	35.9912
0.7327	1.5010	192	0.6800	80.3480	33.4492
0.6564	2.0	256	0.6217	72.1597	28.6576
0.5303	2.5010	320	0.5789	68.8843	27.2312
0.5045	3.0	384	0.5451	69.2938	36.8142
0.381	3.5010	448	0.5374	62.8454	25.3841
0.3915	4.0	512	0.5134	62.6407	25.2195
0.2779	4.5010	576	0.5229	62.6407	25.9144
0.2892	5.0	640	0.5157	62.4360	25.0183
0.2008	5.5010	704	0.5193	60.9007	28.8405
0.206	6.0	768	0.5236	59.3654	23.8113
0.1396	6.5010	832	0.5356	61.7195	27.0300
0.1494	7.0	896	0.5288	58.3419	24.8903
0.1075	7.5010	960	0.5469	61.1054	27.9078
0.0991	8.0	1024	0.5673	57.9324	22.4396
0.0744	8.5010	1088	0.5787	60.2866	26.2436
0.0762	9.0	1152	0.5787	61.3101	26.6459
0.0528	9.5010	1216	0.6053	59.1607	23.0249
0.0577	10.0	1280	0.5880	58.1372	22.8420

wandererupak/whisper-small-ultimate-Augmented

Whisper Small N - Augmented