wav2vec2-xls-r-300m-lg
This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the COMMON_VOICE - LG dataset. It achieves the following results on the evaluation set:
- Loss: 0.6989
- Wer: 0.8529
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
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 20.0
- mixed_precision_training: Native AMP
Training results
Training Loss | Epoch | Step | Validation Loss | Wer |
---|---|---|---|---|
2.9089 | 6.33 | 500 | 2.8983 | 1.0002 |
2.5754 | 12.66 | 1000 | 1.8710 | 1.0 |
1.4093 | 18.99 | 1500 | 0.7195 | 0.8547 |
Framework versions
- Transformers 4.17.0.dev0
- Pytorch 1.10.2+cu102
- Datasets 1.18.2.dev0
- Tokenizers 0.11.0
Evaluation Commands
- To evaluate on
mozilla-foundation/common_voice_7_0
with splittest
python eval.py --model_id samitizerxu/wav2vec2-xls-r-300m-lg --dataset mozilla-foundation/common_voice_7_0 --config lg --split test
- Downloads last month
- 17
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social
visibility and check back later, or deploy to Inference Endpoints (dedicated)
instead.
Dataset used to train samitizerxu/wav2vec2-xls-r-300m-lg
Evaluation results
- Test WER on Common Voice 7self-reported78.890
- Test CER on Common Voice 7self-reported15.160