ymoslem's picture
Update README.md
5664a3f verified
metadata
license: apache-2.0
base_model: openai/whisper-tiny
tags:
  - generated_from_trainer
metrics:
  - bleu
  - wer
model-index:
  - name: Whisper Tiny GA-EN Speech Translation
    results: []

Whisper Tiny GA-EN Speech Translation

This model is a fine-tuned version of openai/whisper-tiny on an unknown dataset. The best model (this version) is at checkpoint 700, epoch 5.19, and it achieves the following results on the evaluation set:

  • Loss: 2.7090
  • Bleu: 17.56
  • Chrf: 32.23
  • Wer: 90.4998

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Experiment

Data Augmentation (processing)

Training hyperparameters

The following hyperparameters were used during training:

  • learning_rate: 0.0001
  • train_batch_size: 128
  • eval_batch_size: 128
  • seed: 42
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 0.03
  • training_steps: 1000
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Bleu Chrf Wer
0.9302 0.74 100 2.1831 8.7 24.4 114.2729
0.3304 1.48 200 2.2829 11.34 28.63 115.6686
0.1095 2.22 300 2.4409 11.35 29.53 117.4696
0.0781 2.96 400 2.5060 12.55 29.32 110.1306
0.0506 3.7 500 2.6288 16.65 31.61 91.9856
0.041 4.44 600 2.6641 15.82 31.65 96.1279
0.0335 5.19 700 2.7090 17.56 32.23 90.4998
0.0296 5.93 800 2.7257 16.61 32.11 93.0662
0.0278 6.67 900 2.7528 15.38 31.57 95.7226
0.0217 7.41 1000 2.7758 15.81 32.03 96.3980

Framework versions

  • Transformers 4.39.2
  • Pytorch 2.2.1+cu121
  • Datasets 2.18.0
  • Tokenizers 0.15.2