wav2vec2-large-xls-r-300m-as
This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the common_voice dataset. It achieves the following results on the evaluation set:
- Loss: 0.8318
- Wer: 0.5174
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_ratio: 0.12
- num_epochs: 120
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 4.882 | 25.0 | 400 | 1.2290 | 0.8182 |
| 0.8275 | 50.0 | 800 | 0.6835 | 0.5398 |
| 0.337 | 75.0 | 1200 | 0.7789 | 0.5107 |
| 0.2113 | 100.0 | 1600 | 0.8318 | 0.5174 |
Framework versions
- Transformers 4.16.2
- Pytorch 1.10.1+cu102
- Datasets 1.17.1.dev0
- Tokenizers 0.10.3
Test Evaluation
Common Voice Assamese Test Set (v7.0)
- WER: 0.7224
- CER: 0.2882