custom_german
This model is a fine-tuned version of flozi00/wav2vec-xlsr-german on the None dataset. It achieves the following results on the evaluation set:
- Loss: 4.6832
- Wer: 1.0
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
- gradient_accumulation_steps: 2
- 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: 5
- num_epochs: 30
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 8.7718 | 5.0 | 5 | 8.5148 | 1.0 |
| 3.7125 | 10.0 | 10 | 5.4304 | 1.0 |
| 2.7679 | 15.0 | 15 | 5.0388 | 1.0 |
| 2.0516 | 20.0 | 20 | 4.4628 | 1.0 |
| 1.6702 | 25.0 | 25 | 4.5341 | 1.0 |
| 1.515 | 30.0 | 30 | 4.6832 | 1.0 |
Framework versions
- Transformers 4.11.3
- Pytorch 1.10.0+cu102
- Datasets 1.13.3
- Tokenizers 0.10.3