wav2vec2-large-xls-r-300m-basaa-cv8
This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the MOZILLA-FOUNDATION/COMMON_VOICE_8_0 - BAS dataset. It achieves the following results on the evaluation set:
- Loss: 0.4648
- Wer: 0.5472
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: 7e-05
- train_batch_size: 32
- eval_batch_size: 16
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- num_epochs: 100.0
- mixed_precision_training: Native AMP
Training results
Training Loss | Epoch | Step | Validation Loss | Wer |
---|---|---|---|---|
2.9421 | 12.82 | 500 | 2.8894 | 1.0 |
1.1872 | 25.64 | 1000 | 0.6688 | 0.7460 |
0.8894 | 38.46 | 1500 | 0.4868 | 0.6516 |
0.769 | 51.28 | 2000 | 0.4960 | 0.6507 |
0.6936 | 64.1 | 2500 | 0.4781 | 0.5384 |
0.624 | 76.92 | 3000 | 0.4643 | 0.5430 |
0.5966 | 89.74 | 3500 | 0.4530 | 0.5591 |
Framework versions
- Transformers 4.16.0.dev0
- Pytorch 1.10.1+cu102
- Datasets 1.18.3
- Tokenizers 0.11.0
- Downloads last month
- 1
Inference Providers
NEW
This model is not currently available via any of the supported third-party Inference Providers, and
the model is not deployed on the HF Inference API.
Dataset used to train infinitejoy/wav2vec2-large-xls-r-300m-basaa-cv8
Evaluation results
- Test WER on Common Voice 8self-reported38.057
- Test CER on Common Voice 8self-reported11.233