AIFFELTHON Final Project Repository

saturi translator

• last update : MAR 31, 2023

---

Team name : Saturi

Participants :

name	role
1. SY	(lead)
2. SH	(model)
3. SA	(data)
4. JI	(model)
5. DS	(coordinator)
6. HW	(coordinator)

Description :

• Eng-to-Korean dialect translation(5 different dialect)

• We trained our model using kor-dialect corpus data downloaded from AIHUB(aihub.or.kr), and paired with their english corpus by translation using hugging face NMT checkpoint.

• Number of data pair for training : kangwon 300k, jeju 200k, jd 170k, cc 110k, gs 110k.

How to use :

• download checkpoint -> please email me if you want to try(seuyon0101@gmail.com).

• python main.py

• insert tokenizer type

• vocab size small? Yes(8k vocab) , No(16k vocab)

• add region tag before input text

Example :

• basic transformer model inference :