whisper-base-te / README.md
parambharat's picture
Update metadata with huggingface_hub
d422538
metadata
language:
  - te
license: apache-2.0
tags:
  - whisper-event
  - generated_from_trainer
metrics:
  - wer
model-index:
  - name: Whisper Base Te - Bharat Ramanathan
    results:
      - task:
          type: automatic-speech-recognition
          name: Automatic Speech Recognition
        dataset:
          name: google/fleurs
          type: google/fleurs
          config: te_in
          split: test
        metrics:
          - type: wer
            value: 39.09
            name: WER

Whisper Base Te - Bharat Ramanathan

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

  • Loss: 0.2455
  • Wer: 42.6485

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: 96
  • 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
0.6341 0.1 500 0.3894 60.7108
0.349 0.2 1000 0.3081 52.0935
0.2792 0.3 1500 0.2874 49.7079
0.2433 0.4 2000 0.2720 47.5657
0.2224 1.06 2500 0.2632 45.2288
0.2058 1.16 3000 0.2529 44.3038
0.1944 1.26 3500 0.2519 44.5959
0.1869 1.36 4000 0.2475 43.7196
0.1811 2.03 4500 0.2451 43.3301
0.1775 2.13 5000 0.2455 42.6485

Framework versions

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