learning_rate: 3e-05
train_batch_size: 4
eval_batch_size: 4
seed: 42
gradient_accumulation_steps: 8
total_train_batch_size: 32
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: cosine
lr_scheduler_warmup_steps: 150
num_epochs: 30
mixed_precision_training: Native AMP

Training Loss	Epoch	Step	Validation Loss	Wer
27.2471	1.3344	100	2.9130	0.9997
13.0177	2.6689	200	0.3523	0.3643
3.0029	4.0	300	0.2441	0.2595
2.4318	5.3344	400	0.2140	0.2180
2.1110	6.6689	500	0.2050	0.2123
1.9648	8.0	600	0.1851	0.2039
1.7470	9.3344	700	0.1939	0.1951
1.6039	10.6689	800	0.1821	0.1843
1.4432	12.0	900	0.1799	0.1779
1.3283	13.3344	1000	0.1795	0.1755
1.3047	14.6689	1100	0.1866	0.1741
1.1873	16.0	1200	0.1842	0.1693
1.1327	17.3344	1300	0.1914	0.1689
1.0576	18.6689	1400	0.1869	0.1665
1.0515	20.0	1500	0.1860	0.1620
0.9851	21.3344	1600	0.1842	0.1599
0.9491	22.6689	1700	0.1871	0.1588
0.9504	24.0	1800	0.1856	0.1591
0.9398	25.3344	1900	0.1871	0.1581
0.9291	26.6689	2000	0.1848	0.1567
0.8734	28.0	2100	0.1857	0.1571
0.9234	29.3344	2200	0.1853	0.1568
0.8981	30.0	2250	0.1854	0.1570

Shagufta/facebook-mms-1b-all-km-indomain15-withac

facebook-mms-1b-all-km-indomain15-withac