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
9.5383	0.9445	200	1.3396	0.9940
3.8181	1.8878	400	0.5978	0.4506
2.3737	2.8312	600	0.4323	0.3510
1.9870	3.7745	800	0.3467	0.3008
1.7556	4.7178	1000	0.3180	0.2751
1.5441	5.6612	1200	0.3030	0.2551
1.3831	6.6045	1400	0.2842	0.2421
1.1620	7.5478	1600	0.2787	0.2316
1.0828	8.4911	1800	0.2659	0.2213
0.9938	9.4345	2000	0.2664	0.2107
0.9596	10.3778	2200	0.2652	0.2063
0.8609	11.3211	2400	0.2624	0.2044
0.8596	12.2645	2600	0.2553	0.1984
0.8242	13.2078	2800	0.2531	0.1957
0.8391	14.1511	3000	0.2585	0.1933
0.7841	15.0945	3200	0.2591	0.1908
0.7683	16.0378	3400	0.2558	0.1901
0.8177	16.9823	3600	0.2559	0.1897
0.7710	17.9256	3800	0.2571	0.1872
0.7162	18.8689	4000	0.2566	0.1873
0.7343	19.8123	4200	0.2555	0.1871
0.7343	20.0	4240	0.2555	0.1870

Shagufta/facebook-wav2vec2-xls-r-1b-km-indomain20-withac

facebook-wav2vec2-xls-r-1b-km-indomain20-withac