wav2vec2-large-xls-r-300m-chuvash

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the MOZILLA-FOUNDATION/COMMON_VOICE_7_0 - CV dataset. It achieves the following results on the evaluation set:

• Loss: 0.7651
• Wer: 0.6166

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: 0.0003
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 100.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.8032	8.77	500	0.8059	0.8352
1.2608	17.54	1000	0.5828	0.6769
1.1337	26.32	1500	0.6892	0.6908
1.0457	35.09	2000	0.7077	0.6781
0.97	43.86	2500	0.5993	0.6228
0.8767	52.63	3000	0.7213	0.6604
0.8223	61.4	3500	0.8161	0.6968
0.7441	70.18	4000	0.7057	0.6184
0.7011	78.95	4500	0.7027	0.6024
0.6542	87.72	5000	0.7092	0.5979
0.6081	96.49	5500	0.7917	0.6324

Framework versions

• Transformers 4.16.0.dev0
• Pytorch 1.10.1+cu102
• Datasets 1.17.1.dev0
• Tokenizers 0.11.0