## Korean Grammatical Error Correction Model
maintainer: [Soyoung Yoon](https://soyoung97.github.io/profile/)

Official repository: [link](https://github.com/soyoung97/GEC-Korean)

Dataset request form: [link](https://forms.gle/kF9pvJbLGvnh8ZnQ6)

Demo: [link](https://huggingface.co/spaces/Soyoung97/gec-korean-demo)

Colab demo: [link](https://colab.research.google.com/drive/1CL__3CpkhBzxWUbvsQmPTQWWu1cWmJHa?usp=sharing)


### Sample code
```
import torch
from transformers import PreTrainedTokenizerFast
from transformers import BartForConditionalGeneration

tokenizer = PreTrainedTokenizerFast.from_pretrained('Soyoung97/gec_kr')
model = BartForConditionalGeneration.from_pretrained('Soyoung97/gec_kr')

text = '한국어는어렵다.'

raw_input_ids = tokenizer.encode(text)
input_ids = [tokenizer.bos_token_id] + raw_input_ids + [tokenizer.eos_token_id]
corrected_ids = model.generate(torch.tensor([input_ids]),
                                max_length=128,
                                eos_token_id=1, num_beams=4,
                                early_stopping=True, repetition_penalty=2.0)
output_text = tokenizer.decode(corrected_ids.squeeze().tolist(), skip_special_tokens=True)


output_text
>>> '한국어는 어렵다.'
```

Special thanks to the [KoBART-summarization repository](https://huggingface.co/gogamza/kobart-summarization) (referenced from it)