|
--- |
|
language: |
|
- vi |
|
license: apache-2.0 |
|
framework_name: NeMo |
|
base_model: Conformer-CTC |
|
|
|
--- |
|
|
|
<!-- This model card has been generated automatically according to the information the Trainer had access to. You |
|
should probably proofread and complete it, then remove this comment. --> |
|
|
|
# Conformer_CTC_vi |
|
|
|
This model is a training version of Conformer-CTC model on the VinBigdata dataset. |
|
It achieves the following results on the evaluation set: |
|
|
|
|
|
## Training and evaluation data |
|
-Training loss: |
|
|
|
![image/png](https://cdn-uploads.huggingface.co/production/uploads/64e22d27eb434eea640655d7/6ctQwCWuW5yQ6DGFdGQdR.png) |
|
|
|
- WER = 0.094 |
|
|
|
![image/png](https://cdn-uploads.huggingface.co/production/uploads/64e22d27eb434eea640655d7/VBpAtNZcJDftOPH7o6q-1.png) |
|
|
|
|
|
### Training hyperparameters |
|
|
|
The following hyperparameters were used during training: |
|
- learning_rate: 5.0 |
|
- train_batch_size: 32 |
|
- eval_batch_size: 32 |
|
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08 |
|
- lr_scheduler_type: NoamAnnealing |
|
- lr_scheduler_warmup_steps: 10000 |
|
- num_epochs: 500 |