README.md

metadata

license: apache-2.0
tags:
  - generated_from_trainer
datasets:
  - mozilla-foundation/common_voice_11_0
metrics:
  - wer
language:
  - ga
model-index:
  - name: wav2vec2-large-xls-r-300m-irish-colab
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: mozilla-foundation/common_voice_11_0
          type: mozilla-foundation/common_voice_11_0
          config: ga-IE
          split: train+validation
          args: ga-IE
        metrics:
          - name: Wer
            type: wer
            value: 52.44117647058824

wav2vec2-large-xls-r-300m-irish-colab

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: 1.148
• Wer: 52.4

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: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
4.6516	12.12	400	1.2867	0.7653
0.4188	24.24	800	1.1262	0.5509

Framework versions

• Transformers 4.25.1
• Pytorch 1.10.0+cu113
• Datasets 2.0.0
• Tokenizers 0.13.2