wav2vec2-mms-1b-l1107-mus-asr-10m

This model is a fine-tuned version of facebook/mms-1b-l1107 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 2.4432
• Wer: 0.4844
• Cer: 0.1329

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: 0.001
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.4
• num_epochs: 12
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
No log	4.0	108	3.4128	1.0	1.0
No log	5.0	135	3.2507	1.0	1.0
No log	6.0	162	2.9310	1.0	0.3287
No log	7.0	189	2.3864	0.875	0.1958
5.6606	8.0	216	2.5689	0.7188	0.1772
5.6606	9.0	243	2.2867	0.625	0.1655
5.6606	10.0	270	2.3748	0.5	0.1352
5.6606	11.0	297	2.4432	0.4844	0.1329
5.6606	12.0	324	2.5012	0.5156	0.1352

Framework versions

• Transformers 4.41.0
• Pytorch 1.10.1+cu111
• Datasets 2.19.1
• Tokenizers 0.19.1