metadata

base_model: facebook/mms-1b-all
datasets:
  - common_voice_17_0
library_name: transformers
license: cc-by-nc-4.0
metrics:
  - wer
  - bleu
tags:
  - generated_from_trainer
model-index:
  - name: wav2vec2-mms-1b-CV17.0
    results:
      - task:
          type: automatic-speech-recognition
          name: Automatic Speech Recognition
        dataset:
          name: common_voice_17_0
          type: common_voice_17_0
          config: ta
          split: test[:5%]+test[20%:25%]+test[60%:65%]+test[90%:]
          args: ta
        metrics:
          - type: wer
            value: 0.39458150446496143
            name: Wer
          - type: bleu
            value: 0.3894086227361947
            name: Bleu

wav2vec2-mms-1b-CV17.0

This model is a fine-tuned version of facebook/mms-1b-all on the common_voice_17_0 dataset. It achieves the following results on the evaluation set:

Loss: 0.2560
Wer: 0.3946
Cer: 0.0667
Bleu: 0.3894

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: 0.0001
train_batch_size: 16
eval_batch_size: 8
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 32
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
lr_scheduler_warmup_ratio: 0.15
training_steps: 5000
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer	Bleu
5.2362	0.3509	500	0.3039	0.4233	0.0716	0.3579
0.2003	0.7018	1000	0.2726	0.4072	0.0694	0.3806
0.1821	1.0526	1500	0.2632	0.3993	0.0677	0.3890
0.1792	1.4035	2000	0.2602	0.4011	0.0677	0.3850
0.1764	1.7544	2500	0.2572	0.3984	0.0674	0.3850
0.1771	2.1053	3000	0.2570	0.3955	0.0672	0.3918
0.1777	2.4561	3500	0.2562	0.3970	0.0670	0.3866
0.1721	2.8070	4000	0.2547	0.3975	0.0669	0.3841
0.177	3.1579	4500	0.2560	0.3949	0.0668	0.3910
0.1765	3.5088	5000	0.2560	0.3946	0.0667	0.3894

Framework versions

Transformers 4.44.2
Pytorch 2.4.0+cu121
Datasets 3.0.0
Tokenizers 0.19.1