xtreme_s_xlsr_300m_fleurs_asr_en_us
This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the GOOGLE/XTREME_S - FLEURS.EN_US dataset. It achieves the following results on the evaluation set:
- Cer: 0.1356
- Loss: 0.5599
- Wer: 0.3148
- Predict Samples: 647
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: 1
- seed: 42
- distributed_type: multi-GPU
- num_devices: 8
- total_train_batch_size: 64
- total_eval_batch_size: 8
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 200
- num_epochs: 30.0
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer | Cer |
|---|---|---|---|---|---|
| 2.8769 | 5.0 | 200 | 2.8871 | 1.0 | 0.9878 |
| 0.2458 | 10.0 | 400 | 0.5570 | 0.4899 | 0.1951 |
| 0.0762 | 15.0 | 600 | 0.5213 | 0.3727 | 0.1562 |
| 0.0334 | 20.0 | 800 | 0.5742 | 0.3666 | 0.1543 |
| 0.0244 | 25.0 | 1000 | 0.5907 | 0.3546 | 0.1499 |
| 0.0143 | 30.0 | 1200 | 0.5961 | 0.3460 | 0.1469 |
Framework versions
- Transformers 4.18.0.dev0
- Pytorch 1.10.1+cu111
- Datasets 1.18.4.dev0
- Tokenizers 0.11.6