<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->
This model is a fine-tuned version of facebook/wav2vec2-xls-r-1b on the MOZILLA-FOUNDATION/COMMON_VOICE_8_0 - GA-IE dataset. It achieves the following results on the evaluation set:
- Loss: 0.9810
- Wer: 0.4761
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: 16
- eval_batch_size: 16
- 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: 1000
- num_epochs: 100.0
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 2.2427 | 15.15 | 500 | 1.4632 | 0.9481 |
| 1.3128 | 30.3 | 1000 | 0.8662 | 0.6195 |
| 0.9403 | 45.45 | 1500 | 0.8163 | 0.5169 |
| 0.6868 | 60.61 | 2000 | 0.8661 | 0.4858 |
| 0.563 | 75.76 | 2500 | 0.9447 | 0.4867 |
| 0.4887 | 90.91 | 3000 | 0.9650 | 0.4823 |
Framework versions
- Transformers 4.17.0.dev0
- Pytorch 1.10.2+cu102
- Datasets 1.18.2.dev0
- Tokenizers 0.11.0
