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-training_set_variations
    results:
      - task:
          type: automatic-speech-recognition
          name: Automatic Speech Recognition
        dataset:
          name: common_voice_17_0
          type: common_voice_17_0
          config: ta
          split: validation
          args: ta
        metrics:
          - type: wer
            value: 0.4250355227574827
            name: Wer
          - type: bleu
            value: 0.3461897882903843
            name: Bleu

wav2vec2-mms-1b-CV17.0-training_set_variations

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: 0.3320
Wer: 0.4250
Cer: 0.0720
Bleu: 0.3462

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: 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: 2000
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer	Bleu
11.5156	6.25	50	5.5076	1.0	0.9701	0.0
1.8133	12.5	100	0.2759	0.4067	0.0674	0.3610
0.1751	18.75	150	0.2414	0.3828	0.0639	0.3924
0.1315	25.0	200	0.2556	0.3887	0.0649	0.3901
0.1	31.25	250	0.2842	0.4168	0.0700	0.3520
0.0842	37.5	300	0.2997	0.4133	0.0699	0.3571
0.0717	43.75	350	0.3210	0.4260	0.0732	0.3431
0.0608	50.0	400	0.3320	0.4250	0.0720	0.3462

Framework versions

Transformers 4.44.2
Pytorch 2.4.1+cu121
Datasets 3.0.0
Tokenizers 0.19.1