anton-l's picture
anton-l HF staff
Upload README.md
71344b5
metadata
license: apache-2.0
language: sah
tags:
  - generated_from_trainer
  - robust-speech-event
  - hf-asr-leaderboard
datasets:
  - common_voice
model-index:
  - name: wav2vec2-xls-r-300m-W2V2-XLSR-300M-YAKUT-SMALL
    results:
      - task:
          name: Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: Common Voice sah
          type: common_voice
          args: sah
        metrics:
          - name: Test WER
            type: wer
            value: 79

wav2vec2-xls-r-300m-W2V2-XLSR-300M-YAKUT-SMALL

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

  • Loss: 0.9068
  • Wer: 0.7900

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: 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_steps: 500
  • num_epochs: 50
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer
4.6926 19.05 400 2.7538 1.0
0.7031 38.1 800 0.9068 0.7900

Framework versions

  • Transformers 4.11.3
  • Pytorch 1.10.0+cu111
  • Datasets 1.14.0
  • Tokenizers 0.10.3