wav2vec2-300m-ft-soft-skill

This model is a fine-tuned version of glob-asr/xls-r-es-test-lm on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.7447
• Accuracy: 0.6827
• F1 Micro: 0.3514
• F1 Macro: 0.6827
• Precision Micro: 0.6827

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 8
• eval_batch_size: 10
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 10
• num_epochs: 10
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1 Micro	F1 Macro	Precision Micro
0.823	0.51	100	0.6821	0.7589	0.2876	0.7589	0.7589
0.7122	1.02	200	0.6767	0.7589	0.2876	0.7589	0.7589
0.6706	1.52	300	0.6768	0.7589	0.2876	0.7589	0.7589
0.7096	2.03	400	0.6791	0.7589	0.2876	0.7589	0.7589
0.6909	2.54	500	0.6780	0.7589	0.2876	0.7589	0.7589
0.6861	3.05	600	0.6779	0.7589	0.2876	0.7589	0.7589
0.6842	3.55	700	0.6773	0.7589	0.2876	0.7589	0.7589
0.6887	4.06	800	0.6764	0.7589	0.2876	0.7589	0.7589
0.6766	4.57	900	0.6803	0.7589	0.2876	0.7589	0.7589
0.6964	5.08	1000	0.6819	0.7589	0.2876	0.7589	0.7589
0.6515	5.58	1100	0.6788	0.7589	0.2876	0.7589	0.7589
0.6608	6.09	1200	0.6864	0.7589	0.2876	0.7589	0.7589
0.6171	6.6	1300	0.6980	0.7589	0.2876	0.7589	0.7589
0.6292	7.11	1400	0.7172	0.7386	0.3119	0.7386	0.7386
0.6015	7.61	1500	0.6988	0.7462	0.3212	0.7462	0.7462
0.6236	8.12	1600	0.7493	0.6954	0.3432	0.6954	0.6954
0.5643	8.63	1700	0.7250	0.7107	0.3466	0.7107	0.7107
0.6134	9.14	1800	0.7561	0.6751	0.3565	0.6751	0.6751
0.5642	9.64	1900	0.7447	0.6827	0.3514	0.6827	0.6827

Framework versions

• Transformers 4.21.3
• Pytorch 1.8.1+cu111
• Datasets 2.4.0
• Tokenizers 0.12.1