IonGrozea's picture
Update README.md
4df2a4d verified
metadata
library_name: transformers
license: apache-2.0
base_model: openai/whisper-small
tags:
  - generated_from_trainer
datasets:
  - generator
metrics:
  - wer
model-index:
  - name: whisper-small_ro-80mel
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: generator
          type: generator
          config: default
          split: None
          args: default
        metrics:
          - name: Wer
            type: wer
            value: 2.6589

whisper-small_ro-80mel

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

  • Loss: 0.1116
  • Wer: 2.6589
  • Cer: 3.1776

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: 2e-05
  • train_batch_size: 12
  • eval_batch_size: 12
  • seed: 42
  • gradient_accumulation_steps: 16
  • total_train_batch_size: 192
  • optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
  • lr_scheduler_type: cosine
  • lr_scheduler_warmup_steps: 1600
  • num_epochs: 6

Training results

Training Loss Epoch Step Validation Loss Wer Cer
0.1771 1.8525 2000 0.1116 3.1124 3.2257
0.0966 3.7043 4000 0.0668 3.196 3.3828
0.0675 5.5560 6000 0.0635 3.7906 3.8308

"eval_runtime": 19430.1447, "eval_samples": 27174, "eval_samples_per_second": 1.399, "eval_steps_per_second": 0.117, "test_samples": 12987, "train_samples": 207181

Framework versions

  • Transformers 4.57.0
  • Pytorch 2.9.1+cu128
  • Datasets 4.4.1
  • Tokenizers 0.22.1