trainer_compile_test
This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the librispeech_asr dataset. It achieves the following results on the evaluation set:
- Loss: 1.6771
- Cer: 0.8309
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: 1e-05
- train_batch_size: 16
- eval_batch_size: 16
- 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: cosine
- lr_scheduler_warmup_steps: 75
- num_epochs: 10
Training results
Training Loss | Epoch | Step | Validation Loss | Cer |
---|---|---|---|---|
3.2503 | 1.2 | 300 | 3.2171 | 1.0 |
2.953 | 2.4 | 600 | 2.9743 | 1.0 |
2.8961 | 3.6 | 900 | 2.9325 | 1.0 |
2.8357 | 4.8 | 1200 | 2.8215 | 1.0 |
2.2379 | 6.0 | 1500 | 2.1781 | 0.9481 |
1.8339 | 7.2 | 1800 | 1.8275 | 0.8780 |
1.6769 | 8.4 | 2100 | 1.7017 | 0.8374 |
1.642 | 9.6 | 2400 | 1.6771 | 0.8309 |
Framework versions
- Transformers 4.33.3
- Pytorch 2.1.2
- Datasets 2.16.1
- Tokenizers 0.11.0
- Downloads last month
- 1