FutureProofGlitch - Whisper Small - Fine Tuned on Lectures
This model is a fine-tuned version of futureProofGlitch/whisper-small-v2 on the TBK's Treasured Lectures dataset. It achieves the following results on the evaluation set:
- Loss: 0.3574
- Wer Ortho: 0.1834
- Wer: 0.0562
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: 1.1e-05
- train_batch_size: 16
- eval_batch_size: 16
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: constant_with_warmup
- lr_scheduler_warmup_steps: 10
- training_steps: 100
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer Ortho | Wer |
|---|---|---|---|---|---|
| No log | 0.21 | 25 | 0.8342 | 0.2377 | 0.0939 |
| 3.0694 | 0.42 | 50 | 0.4413 | 0.2100 | 0.0651 |
| 3.0694 | 0.64 | 75 | 0.3754 | 0.1859 | 0.0557 |
| 0.3126 | 0.85 | 100 | 0.3574 | 0.1834 | 0.0562 |
Framework versions
- Transformers 4.39.3
- Pytorch 2.2.1+cu121
- Datasets 2.18.0
- Tokenizers 0.15.2
- Downloads last month
- 9