morenolq's picture
Update metadata with huggingface_hub
67d8e4a
|
raw
history blame
2.74 kB
metadata
language:
  - hu
license: apache-2.0
tags:
  - whisper-event
  - generated_from_trainer
datasets:
  - mozilla-foundation/common_voice_11_0
metrics:
  - wer
model-index:
  - name: Whisper Small Hungarian - Robust
    results:
      - task:
          type: automatic-speech-recognition
          name: Automatic Speech Recognition
        dataset:
          name: mozilla-foundation/common_voice_11_0 hu
          type: mozilla-foundation/common_voice_11_0
          config: hu
          split: test
          args: hu
        metrics:
          - type: wer
            value: 30.904239549362583
            name: Wer
          - type: wer
            value: 26.15
            name: WER
      - task:
          type: automatic-speech-recognition
          name: Automatic Speech Recognition
        dataset:
          name: google/fleurs
          type: google/fleurs
          config: hu_hu
          split: test
        metrics:
          - type: wer
            value: 35.49
            name: WER

Whisper Small Hungarian - Robust

This model is a fine-tuned version of openai/whisper-small on the mozilla-foundation/common_voice_11_0 hu dataset. It achieves the following results on the evaluation set:

  • Loss: 0.5243
  • Wer: 30.9042

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: 5e-05
  • train_batch_size: 128
  • eval_batch_size: 64
  • 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: 5000
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer
0.0763 7.46 500 0.5268 40.5099
0.0147 14.93 1000 0.5233 36.1429
0.0064 22.39 1500 0.5467 35.0934
0.0045 29.85 2000 0.5434 34.2929
0.0019 37.31 2500 0.5348 32.7868
0.0008 44.78 3000 0.5314 32.0605
0.0008 52.24 3500 0.5438 32.6920
0.0005 59.7 4000 0.5428 32.0931
0.0003 67.16 4500 0.5328 31.2511
0.0004 74.63 5000 0.5292 31.1236

Framework versions

  • Transformers 4.26.0.dev0
  • Pytorch 1.13.0+cu117
  • Datasets 2.7.1.dev0
  • Tokenizers 0.13.2