learning_rate: 1.5e-05
train_batch_size: 4
eval_batch_size: 4
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_steps: 150
num_epochs: 20
mixed_precision_training: Native AMP

Training Loss	Epoch	Step	Validation Loss	Wer
11.6650	0.9445	200	2.9165	1.0
1.9822	1.8878	400	0.3216	0.2960
1.4406	2.8312	600	0.2689	0.2619
1.2779	3.7745	800	0.2356	0.2238
1.1193	4.7178	1000	0.2190	0.2080
1.0423	5.6612	1200	0.2154	0.2038
0.9447	6.6045	1400	0.2070	0.1944
0.8678	7.5478	1600	0.2039	0.1877
0.8488	8.4911	1800	0.1984	0.1817
0.7765	9.4345	2000	0.2037	0.1793
0.7410	10.3778	2200	0.1944	0.1726
0.7008	11.3211	2400	0.1983	0.1720
0.7001	12.2645	2600	0.1961	0.1741
0.6553	13.2078	2800	0.1967	0.1684
0.6469	14.1511	3000	0.1953	0.1682
0.6271	15.0945	3200	0.1921	0.1641
0.6198	16.0378	3400	0.1921	0.1628
0.6432	16.9823	3600	0.1924	0.1639
0.6262	17.9256	3800	0.1926	0.1627
0.5928	18.8689	4000	0.1926	0.1624
0.5891	19.8123	4200	0.1921	0.1624
0.5891	20.0	4240	0.1921	0.1624

Shagufta/facebook-mms-1b-all-km-indomain20-withac

facebook-mms-1b-all-km-indomain20-withac