|
--- |
|
license: apache-2.0 |
|
tags: |
|
- generated_from_trainer |
|
model-index: |
|
- name: bart-base-spelling-nl |
|
results: [] |
|
--- |
|
|
|
# bart-base-spelling-nl |
|
|
|
This model is a Dutch fine-tuned version of [facebook/bart-base](https://huggingface.co./facebook/bart-base). |
|
|
|
It achieves the following results on the evaluation set: |
|
- Loss: 0.0276 |
|
- Cer: 0.0147 |
|
|
|
## Model description |
|
|
|
This is a text-to-text fine-tuned version of [facebook/bart-base](https://huggingface.co./facebook/bart-base) trained on spelling correction. It leans on the excellent work by Oliver Guhr ([github](https://github.com/oliverguhr/spelling), [huggingface](https://huggingface.co./oliverguhr/spelling-correction-english-base)). Training was performed on an AWS EC2 instance (g5.xlarge) on a single GPU in about 4 hours. |
|
|
|
## Intended uses & limitations |
|
|
|
The intended use for this model is to be a component of the [Valkuil.net](https://valkuil.net) context-sensitive spelling checker. A next version of the model will be trained on more data. |
|
|
|
## Training and evaluation data |
|
|
|
The model was trained on a Dutch dataset composed of 300,000 lines of text from three public Dutch sources, downloaded from the [Opus corpus](https://opus.nlpl.eu/): |
|
|
|
- nl-europarlv7.100k.txt |
|
- nl-opensubtitles2016.100k.txt |
|
- nl-wikipedia.100k.txt |
|
|
|
|
|
## Training procedure |
|
|
|
### Training hyperparameters |
|
|
|
The following hyperparameters were used during training: |
|
- learning_rate: 0.0003 |
|
- train_batch_size: 2 |
|
- eval_batch_size: 4 |
|
- seed: 42 |
|
- gradient_accumulation_steps: 16 |
|
- total_train_batch_size: 32 |
|
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08 |
|
- lr_scheduler_type: linear |
|
- num_epochs: 2.0 |
|
|
|
### Training results |
|
|
|
| Training Loss | Epoch | Step | Validation Loss | Cer | |
|
|:-------------:|:-----:|:-----:|:---------------:|:------:| |
|
| 0.1617 | 0.11 | 1000 | 0.0986 | 0.9241 | |
|
| 0.1326 | 0.21 | 2000 | 0.0676 | 0.9240 | |
|
| 0.09 | 0.32 | 3000 | 0.0586 | 0.9241 | |
|
| 0.0891 | 0.43 | 4000 | 0.0530 | 0.9240 | |
|
| 0.0753 | 0.54 | 5000 | 0.0491 | 0.9239 | |
|
| 0.069 | 0.64 | 6000 | 0.0459 | 0.9238 | |
|
| 0.0615 | 0.75 | 7000 | 0.0435 | 0.9238 | |
|
| 0.0494 | 0.86 | 8000 | 0.0409 | 0.9237 | |
|
| 0.0671 | 0.97 | 9000 | 0.0388 | 0.9238 | |
|
| 0.0425 | 1.07 | 10000 | 0.0367 | 0.9237 | |
|
| 0.0394 | 1.18 | 11000 | 0.0356 | 0.9237 | |
|
| 0.0399 | 1.29 | 12000 | 0.0344 | 0.9236 | |
|
| 0.0375 | 1.4 | 13000 | 0.0333 | 0.9235 | |
|
| 0.0409 | 1.5 | 14000 | 0.0315 | 0.9237 | |
|
| 0.0291 | 1.61 | 15000 | 0.0304 | 0.9236 | |
|
| 0.0268 | 1.72 | 16000 | 0.0293 | 0.9236 | |
|
| 0.0309 | 1.83 | 17000 | 0.0284 | 0.9235 | |
|
| 0.0362 | 1.93 | 18000 | 0.0276 | 0.9235 | |
|
|
|
|
|
### Framework versions |
|
|
|
- Transformers 4.27.3 |
|
- Pytorch 2.0.0+cu117 |
|
- Datasets 2.10.1 |
|
- Tokenizers 0.13.2 |
|
|
