<!-- 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. -->
wav2vec2-xls-r-300m-ar-3
This model is a fine-tuned version of MeshalAlamr/wav2vec2-xls-r-300m-ar-2 on the common_voice dataset. It achieves the following results on the evaluation set:
- Loss: 0.5567
- Wer: 0.3115
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: 0.001
- train_batch_size: 32
- eval_batch_size: 16
- seed: 42
- gradient_accumulation_steps: 2
- 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 |
|---|---|---|---|---|
| 0.1654 | 1.18 | 400 | 0.5815 | 0.4237 |
| 0.3412 | 2.35 | 800 | 0.5534 | 0.4479 |
| 0.4661 | 1.77 | 1200 | 0.6339 | 0.4915 |
| 0.441 | 2.36 | 1600 | 0.6435 | 0.5016 |
| 0.3273 | 5.88 | 2000 | 0.5338 | 0.4361 |
| 0.3099 | 7.06 | 2400 | 0.5570 | 0.4303 |
| 0.2833 | 8.24 | 2800 | 0.5731 | 0.4427 |
| 0.2714 | 9.41 | 3200 | 0.5551 | 0.4212 |
| 0.2598 | 10.59 | 3600 | 0.5757 | 0.4214 |
| 0.2458 | 11.76 | 4000 | 0.5269 | 0.4065 |
| 0.2316 | 12.94 | 4400 | 0.5469 | 0.4053 |
| 0.219 | 14.12 | 4800 | 0.5539 | 0.3912 |
| 0.2022 | 15.29 | 5200 | 0.5773 | 0.3887 |
| 0.1771 | 16.47 | 5600 | 0.5374 | 0.3623 |
| 0.176 | 17.65 | 6000 | 0.5545 | 0.3763 |
| 0.1645 | 18.82 | 6400 | 0.5332 | 0.3580 |
| 0.1501 | 20.0 | 6800 | 0.5496 | 0.3614 |
| 0.1372 | 21.18 | 7200 | 0.5716 | 0.3608 |
| 0.1325 | 22.35 | 7600 | 0.5476 | 0.3475 |
| 0.1233 | 23.53 | 8000 | 0.5657 | 0.3412 |
| 0.1148 | 24.71 | 8400 | 0.5399 | 0.3324 |
| 0.1058 | 25.88 | 8800 | 0.5678 | 0.3323 |
| 0.1004 | 27.06 | 9200 | 0.5648 | 0.3252 |
| 0.0953 | 28.24 | 9600 | 0.5594 | 0.3159 |
| 0.0875 | 29.41 | 10000 | 0.5567 | 0.3115 |
Framework versions
- Transformers 4.14.1
- Pytorch 1.11.0
- Datasets 1.18.3
- Tokenizers 0.10.3
