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
2.2835	0.5010	64	1.5724	137.6663	100.6584
1.2882	1.0	128	1.0519	102.4565	47.4945
0.9701	1.5010	192	0.8603	90.5834	42.3007
0.8499	2.0	256	0.7567	80.6551	34.7293
0.7095	2.5010	320	0.7002	79.8362	33.3577
0.6677	3.0	384	0.6425	74.6162	31.1448
0.5761	3.5010	448	0.6236	72.9785	30.9802
0.5718	4.0	512	0.5990	76.5609	32.3702
0.49	4.5010	576	0.5865	75.7421	31.5106
0.4757	5.0	640	0.5690	71.5455	28.9320
0.4141	5.5010	704	0.5641	71.0338	29.7732
0.4095	6.0	768	0.5655	67.4514	27.9627
0.3499	6.5010	832	0.5705	67.9632	27.5786
0.3632	7.0	896	0.5540	71.0338	30.9985
0.3078	7.5010	960	0.5613	66.5302	26.8288
0.2985	8.0	1024	0.5569	66.9396	26.7374
0.2503	8.5010	1088	0.5629	65.8137	26.5728
0.2637	9.0	1152	0.5527	65.1996	26.1339
0.2187	9.5010	1216	0.5671	65.8137	26.9386
0.2257	10.0	1280	0.5719	65.5067	26.4996
0.1856	10.5010	1344	0.5649	66.7349	26.2985
0.1971	11.0	1408	0.5745	64.4831	26.2436
0.1674	11.5010	1472	0.5771	65.8137	26.2619
0.1639	12.0	1536	0.5781	65.6090	27.5604
0.1403	12.5010	1600	0.5843	65.7114	26.8837
0.1477	13.0	1664	0.5879	65.1996	26.9203

wandererupak/whisper-base-ultimate-Augmented

Whisper base N - Augmented