whisper-small-hr / README.md
mapau's picture
End of training
20da83b verified
metadata
language:
  - hr
license: apache-2.0
base_model: openai/whisper-small
tags:
  - generated_from_trainer
datasets:
  - parlaSmall
metrics:
  - wer
model-index:
  - name: Whisper Small Croatian
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: parlaSmall_subset
          type: parlaSmall
          config: default
          split: None
          args: 'config: hr, split: test'
        metrics:
          - name: Wer
            type: wer
            value: 25.440806045340054

Whisper Small Croatian

This model is a fine-tuned version of openai/whisper-small on the parlaSmall_subset dataset. It achieves the following results on the evaluation set:

  • Loss: 0.5739
  • Wer: 25.4408

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: 1e-05
  • train_batch_size: 2
  • eval_batch_size: 8
  • seed: 42
  • gradient_accumulation_steps: 8
  • total_train_batch_size: 16
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 500
  • training_steps: 4000
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer
0.0003 32.52 1000 0.5073 25.0630
0.0001 65.04 2000 0.5470 25.5668
0.0001 97.56 3000 0.5668 25.0630
0.0 130.08 4000 0.5739 25.4408

Framework versions

  • Transformers 4.39.3
  • Pytorch 2.0.0
  • Datasets 2.19.1
  • Tokenizers 0.15.1