metadata
license: apache-2.0
tags:
- automatic-speech-recognition
- robust-speech-event
- generated_from_trainer
datasets:
- common_voice
model-index:
- name: wav2vec2-large-xls-r-300m-hi
results: []
wav2vec2-large-xls-r-300m-hi
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: 2.5039
- Wer: 0.8877
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: 7.5e-05
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 4
- total_train_batch_size: 32
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- num_epochs: 50
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 9.4071 | 4.76 | 400 | 3.5871 | 1.0 |
| 3.5056 | 9.52 | 800 | 3.4414 | 1.0 |
| 2.9652 | 14.28 | 1200 | 2.1936 | 0.9573 |
| 1.3822 | 19.05 | 1600 | 2.1039 | 0.9157 |
| 0.9906 | 23.81 | 2000 | 2.2512 | 0.8960 |
| 0.8405 | 28.57 | 2400 | 2.2878 | 0.8931 |
| 0.7686 | 33.33 | 2800 | 2.3291 | 0.8884 |
| 0.7092 | 38.09 | 3200 | 2.4806 | 0.8921 |
| 0.6757 | 42.85 | 3600 | 2.4675 | 0.8847 |
| 0.6606 | 47.62 | 4000 | 2.5039 | 0.8877 |
Framework versions
- Transformers 4.11.3
- Pytorch 1.10.1+cu102
- Datasets 1.17.1.dev0
- Tokenizers 0.10.3
Built during robust-speech-challenge. Will keep updating the same !
Thanks Patrick, Anton for the wonderful event.