whisper-tiny-cit-nt / README.md
Makkoen's picture
End of training
cb3e00b verified
metadata
language:
  - en
license: apache-2.0
base_model: openai/whisper-tiny
tags:
  - generated_from_trainer
metrics:
  - wer
model-index:
  - name: Whisper tiny En - SF test
    results: []

Whisper tiny En - SF test

This model is a fine-tuned version of openai/whisper-tiny on the SF 200 dataset. It achieves the following results on the evaluation set:

  • Loss: 1.7031
  • Wer: 62.9291

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
  • distributed_type: multi-GPU
  • 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: 1000
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer
0.5292 8.8889 100 1.0957 59.9542
0.0462 17.7778 200 1.2314 57.6659
0.0058 26.6667 300 1.4473 63.6156
0.0029 35.5556 400 1.5361 63.6156
0.0017 44.4444 500 1.6016 60.6407
0.0012 53.3333 600 1.6367 62.9291
0.0009 62.2222 700 1.6670 63.6156
0.0008 71.1111 800 1.6875 63.6156
0.0007 80.0 900 1.6992 63.6156
0.0007 88.8889 1000 1.7031 62.9291

Framework versions

  • Transformers 4.41.0
  • Pytorch 2.3.0+cu121
  • Datasets 2.19.1
  • Tokenizers 0.19.1