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
training_steps: 1000
mixed_precision_training: Native AMP

Training Loss	Epoch	Step	Validation Loss	Bleu
9.5028	12.5	50	7.3272	0.0
6.3107	25.0	100	5.5738	0.0
5.3612	37.5	150	5.0667	0.0
4.9035	50.0	200	4.5926	0.0
4.3264	62.5	250	3.9335	0.0
3.5805	75.0	300	3.1526	0.0
2.7738	87.5	350	2.4034	0.0045
2.1907	100.0	400	2.0046	0.0099
1.8861	112.5	450	1.7903	0.0095
1.6974	125.0	500	1.5375	0.0072
1.4036	137.5	550	1.2622	0.0118
1.1448	150.0	600	1.0376	0.0024
0.95	162.5	650	0.8782	0.0208
0.8016	175.0	700	0.7391	0.0800
0.6971	187.5	750	0.6580	0.0836
0.6094	200.0	800	0.5693	0.1410
0.5455	212.5	850	0.5185	0.1336
0.4971	225.0	900	0.4791	0.1210
0.4654	237.5	950	0.4571	0.1636
0.4471	250.0	1000	0.4484	0.1506

Marialab/whisper-large-v3-turbo-swish

whisper-large-v3-turbo-darija-st