README.md

metadata

language:
  - id
license: apache-2.0
tags:
  - whisper-event
  - generated_from_trainer
model-index:
  - name: Whisper Medium ID - FLEURS-CV
    results:
      - task:
          type: automatic-speech-recognition
          name: Automatic Speech Recognition
        dataset:
          name: google/fleurs
          type: google/fleurs
          config: id_id
          split: test
        metrics:
          - type: wer
            value: 7.8
            name: WER
          - type: cer
            value: 2.43
            name: CER
      - task:
          type: automatic-speech-recognition
          name: Automatic Speech Recognition
        dataset:
          name: mozilla-foundation/common_voice_11_0
          type: mozilla-foundation/common_voice_11_0
          config: id
          split: test
        metrics:
          - type: wer
            value: 8.67
            name: WER
          - type: cer
            value: 2.71
            name: CER

Whisper Medium ID - FLEURS-CV

This model is a fine-tuned version of openai/whisper-medium on the None dataset. It achieves the following results on the evaluation set:

• eval_loss: 0.2563
• eval_wer: 8.4690
• eval_runtime: 2961.9108
• eval_samples_per_second: 1.453
• eval_steps_per_second: 0.091
• epoch: 14.29
• step: 5000

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: 32
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 10000
• mixed_precision_training: Native AMP

Framework versions

• Transformers 4.26.0.dev0
• Pytorch 1.13.1+cu117
• Datasets 2.7.1.dev0
• Tokenizers 0.13.2