ls-timit-100percent-supervised-meta
This model was trained from scratch on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.0649
- Wer: 0.0253
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: 0.0001
- train_batch_size: 32
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 1000
- num_epochs: 20
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.0964 | 7.04 | 1000 | 0.0706 | 0.0342 |
| 0.0445 | 14.08 | 2000 | 0.0649 | 0.0253 |
Framework versions
- Transformers 4.16.2
- Pytorch 1.10.2
- Datasets 1.18.2
- Tokenizers 0.10.3