|
--- |
|
language: fr |
|
tags: |
|
- grammar |
|
- text2text-generation |
|
license: mit |
|
datasets: |
|
- fdemelo/spelling-correction-french-news |
|
--- |
|
|
|
# T5 Spelling Correction (French) |
|
|
|
This model corrects spelling and punctuation problems of the input text (in French). |
|
It was trained on a [synthetic dataset](https://huggingface.co/datasets/fdemelo/spelling-correction-french-news) |
|
based on a French news corpus (2023) provided by the [University of Leipzig](https://wortschatz.uni-leipzig.de/en/download/French) |
|
using [Happy Transformer](https://github.com/EricFillion/happy-transformer). |
|
The base model used for training is [airKlizz/t5-base-multi-fr-wiki-news](https://huggingface.co/airKlizz/t5-base-multi-fr-wiki-news). |
|
The following article was used as reference [full article](https://www.vennify.ai/fine-tune-grammar-correction/). |
|
|
|
|
|
## Usage |
|
|
|
`pip install happytransformer ` |
|
|
|
```python |
|
from happytransformer import HappyTextToText, TTSettings |
|
|
|
happy_tt = HappyTextToText("T5", "fdemelo/t5-base-spell-correction-fr") |
|
|
|
args = TTSettings(num_beams=5, min_length=1) |
|
|
|
# Add the prefix "grammaire: " before each input |
|
result = happy_tt.generate_text("grammaire: Le vehicule a tombe encontrebas", args=args) |
|
|
|
print(result.text) # corrected sentence |
|
|
|
|
|
``` |
|
|
