Byt5 Korean Base

🚀 ByT5-Korean - base

ByT5-Korean is a Korean-specific extension of Google's ByT5. It addresses the unique characteristics of the Korean language, enhancing the encoding and processing capabilities for Korean text.

Key Information

🚀 Quick Start

ByT5-Korean is a Korean specific extension of Google's ByT5.

A Korean syllable has three components (called Jamo): a beginning consonant, a middle vowel, and an optional final consonant; they are like individual characters of alphabet. While the ByT5's utf-8 encoding allows generic encoding for multiple languages, it is unnatural for Korean because it splits the bits representation of each Jamo in the middle.

ByT5-Korean extends ByT5's utf-8 encoding with special care for Korean syllables; each Jamo is represented with an extra token. ByT5-Korean was pre-trained on mC4 with 70% Korean and 30% English.

✨ Features

• Special Encoding for Korean: ByT5-Korean extends ByT5's utf - 8 encoding, representing each Jamo with an extra token, which is more natural for the Korean language.
• Multilingual Pre - training: Pre - trained on mC4 dataset with a mix of 70% Korean and 30% English.

📚 Documentation

Encoding Scheme

id: token
0: <pad>
1: <eos>
2: <unk>
3~258: utf-8 encoding
259~277: beginning consonants(초성), 19개(ㄱㄲㄴㄷㄸㄹㅁㅂㅃㅅㅆㅇㅈㅉㅊㅋㅌㅍㅎ)
278~298: middle vowel(중성), 21개(ㅏㅐㅑㅒㅓㅔㅕㅖㅗㅘㅙㅚㅛㅜㅝㅞㅟㅠㅡㅢㅣ)
299~326: final consonant(종성), 무종성+27개(ㄱㄲㄳㄴㄵㄶㄷㄹㄺㄻㄼㄽㄾㄿㅀㅁㅂㅄㅅㅆㅇㅈㅊㅋㅌㅍㅎ)
327~384: from <extra_id_0> to <extra_id_57>

💻 Usage Examples

Basic Usage

import torch
from tokenizer import ByT5KoreanTokenizer # https://huggingface.co/everdoubling/byt5-Korean-base/blob/main/tokenizer.py
from transformers import T5ForConditionalGeneration

tokenizer_jamo = ByT5KoreanTokenizer()
model = T5ForConditionalGeneration.from_pretrained('everdoubling/byt5-Korean-base')

input_sentence = '한국어 위키백과(영어: Korean Wikipedia)는 한국어로 운영되는 위키백과의 다언어판 가운데 하나로서, 2002년 10월 11일에 <extra_id_0>. 또한 현재 한국어 위키백과에는 넘겨주기, 토론, 그림 등 페이지로 불리는 모든 문서를 포함하면 총 2,629,860개가 <extra_id_1>되어 있으며, 넘겨주기를 포함한 일반 문서 수는 1,278,560개,[1] 그중 넘겨주기, 막다른 문서를 제외한 일반 문서 수는 573,149개이다.'

input_ids_jamo = tokenizer_jamo(input_sentence).input_ids
outputs_jamo = model_jamo.generate(torch.tensor([input_ids_jamo]))
print(tokenizer_jamo.decode(outputs_jamo[0]))
# <pad><extra_id_0>설립되었다<extra_id_1>đě

📄 License

This project is licensed under the Apache-2.0 license.