wav2vec2-E
This model is a fine-tuned version of facebook/wav2vec2-base on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.4832
- Wer: 0.3432
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: 30
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 3.5034 | 4.0 | 500 | 1.1620 | 0.8995 |
| 0.5738 | 8.0 | 1000 | 0.4625 | 0.4396 |
| 0.2142 | 12.0 | 1500 | 0.4791 | 0.3965 |
| 0.1219 | 16.0 | 2000 | 0.4677 | 0.3703 |
| 0.0854 | 20.0 | 2500 | 0.4782 | 0.3544 |
| 0.0587 | 24.0 | 3000 | 0.4680 | 0.3516 |
| 0.044 | 28.0 | 3500 | 0.4832 | 0.3432 |
Framework versions
- Transformers 4.11.3
- Pytorch 1.10.0+cu111
- Datasets 1.18.4
- Tokenizers 0.10.3