wav2vec2-large-xlsr-53-demo-colab
This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the common_voice dataset. It achieves the following results on the evaluation set:
- Loss: 0.4253
- Wer: 0.4880
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: 0.0003
- train_batch_size: 16
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 32
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- num_epochs: 30
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 5.2135 | 4.21 | 400 | 2.5232 | 1.0 |
| 0.8323 | 8.42 | 800 | 0.4673 | 0.6142 |
| 0.3247 | 12.63 | 1200 | 0.4087 | 0.5536 |
| 0.217 | 16.84 | 1600 | 0.3950 | 0.5237 |
| 0.166 | 21.05 | 2000 | 0.4294 | 0.5075 |
| 0.141 | 25.26 | 2400 | 0.4219 | 0.4944 |
| 0.1193 | 29.47 | 2800 | 0.4253 | 0.4880 |
Framework versions
- Transformers 4.18.0
- Pytorch 1.11.0+cu102
- Datasets 2.1.0
- Tokenizers 0.12.1