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: 500
num_epochs: 3.0
mixed_precision_training: Native AMP

Training Loss	Epoch	Step	Validation Loss	Wer
0.7172	0.1028	1000	0.7280	40.3137
0.6097	0.2055	2000	0.6292	35.9280
0.6005	0.3083	3000	0.5747	39.8925
0.5383	0.4111	4000	0.5368	51.1182
0.5332	0.5138	5000	0.5145	49.7822
0.4759	0.6166	6000	0.4885	70.9410
0.4824	0.7194	7000	0.4662	123.4679
0.437	0.8221	8000	0.4551	77.9988
0.4217	0.9249	9000	0.4411	60.1220
0.3637	1.0276	10000	0.4334	78.2602
0.3461	1.1304	11000	0.4260	46.1226
0.3613	1.2332	12000	0.4175	47.5748
0.3478	1.3359	13000	0.4104	32.2829
0.3413	1.4387	14000	0.4033	30.0465
0.3331	1.5415	15000	0.4022	24.7023
0.3325	1.6442	16000	0.3935	24.2085
0.3142	1.7470	17000	0.3921	24.8185
0.3158	1.8498	18000	0.3847	26.9532
0.305	1.9525	19000	0.3817	26.2416
0.2887	2.0553	20000	0.3822	26.4304
0.2723	2.1581	21000	0.3793	28.8411
0.3019	2.2608	22000	0.3756	33.7206
0.2693	2.3636	23000	0.3757	32.0796
0.2721	2.4663	24000	0.3760	31.7746
0.2715	2.5691	25000	0.3727	34.9695
0.272	2.6719	26000	0.3716	34.1708
0.2557	2.7746	27000	0.3702	34.7517
0.2661	2.8774	28000	0.3676	34.0546
0.2522	2.9802	29000	0.3667	33.5754

linl03/vibase-fineturn

Whisper small vi - Ox