<!-- 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. -->
84rry-xlsr-53-arabic
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: 1.0025
- Wer: 0.4977
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: 0.0005
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 8
- total_train_batch_size: 64
- 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 |
|---|---|---|---|---|
| 1.4906 | 2.25 | 500 | 1.3179 | 0.8390 |
| 0.8851 | 4.5 | 1000 | 0.7385 | 0.6221 |
| 0.6884 | 6.76 | 1500 | 0.7005 | 0.5765 |
| 0.5525 | 9.01 | 2000 | 0.6931 | 0.5610 |
| 0.474 | 11.26 | 2500 | 0.7977 | 0.5560 |
| 0.3976 | 13.51 | 3000 | 0.7750 | 0.5375 |
| 0.343 | 15.76 | 3500 | 0.7553 | 0.5206 |
| 0.2838 | 18.02 | 4000 | 0.8162 | 0.5099 |
| 0.2369 | 20.27 | 4500 | 0.8574 | 0.5124 |
| 0.2298 | 22.52 | 5000 | 0.8848 | 0.5057 |
| 0.1727 | 24.77 | 5500 | 0.9193 | 0.5070 |
| 0.1675 | 27.03 | 6000 | 0.9959 | 0.4988 |
| 0.1457 | 29.28 | 6500 | 1.0025 | 0.4977 |
Framework versions
- Transformers 4.19.2
- Pytorch 1.11.0+cu113
- Datasets 2.2.1
- Tokenizers 0.12.1
