Wav2vec2-large-arabic-no-diacs_V2
This model is a fine-tuned version of Seyfelislem/Wav2vec2-large-arabic-no-diacs_V1 on the common_voice dataset. It achieves the following results on the evaluation set:
- Loss: 0.3147
- Wer: 0.2430
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: 8
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 4
- 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
- training_steps: 5000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.3337 | 0.83 | 1000 | 0.4303 | 0.3540 |
| 0.2753 | 1.66 | 2000 | 0.3626 | 0.3026 |
| 0.1753 | 2.5 | 3000 | 0.3558 | 0.2820 |
| 0.141 | 3.33 | 4000 | 0.3333 | 0.2551 |
| 0.1449 | 4.16 | 5000 | 0.3147 | 0.2430 |
Framework versions
- Transformers 4.20.1
- Pytorch 1.10.0+cu113
- Datasets 2.1.0
- Tokenizers 0.12.1