This is a DistilBERT model pre-trained on 131 GB of Japanese web text. The teacher model is BERT-base that built in-house at LINE. The model was trained by LINE Corporation.
https://huggingface.co/line-corporation/line-distilbert-base-japanese
README_ja.md is written in Japanese.
from transformers import AutoTokenizer, AutoModel
tokenizer = AutoTokenizer.from_pretrained("line-corporation/line-distilbert-base-japanese", trust_remote_code=True)
model = AutoModel.from_pretrained("line-corporation/line-distilbert-base-japanese")
sentence = "LINE株式会社で[MASK]の研究・開発をしている。"
print(model(**tokenizer(sentence, return_tensors="pt")))
fugashi
sentencepiece
unidic-lite
The model architecture is the DitilBERT base model; 6 layers, 768 dimensions of hidden states, 12 attention heads, 66M parameters.
The evaluation by JGLUE is as follows:
model name | #Params | Marc_ja | JNLI | JSTS | JSQuAD | JCommonSenseQA |
---|---|---|---|---|---|---|
acc | acc | Pearson/Spearman | EM/F1 | acc | ||
LINE-DistilBERT | 68M | 95.6 | 88.9 | 89.2/85.1 | 87.3/93.3 | 76.1 |
Laboro-DistilBERT | 68M | 94.7 | 82.0 | 87.4/82.7 | 70.2/87.3 | 73.2 |
BandaiNamco-DistilBERT | 68M | 94.6 | 81.6 | 86.8/82.1 | 80.0/88.0 | 66.5 |
The texts are first tokenized by MeCab with the Unidic dictionary and then split into subwords by the SentencePiece algorithm. The vocabulary size is 32768.
The pretrained models are distributed under the terms of the Apache License, Version 2.0.
We haven't published any paper on this work. Please cite this GitHub repository:
@article{LINE DistilBERT Japanese,
title = {LINE DistilBERT Japanese},
author = {"Koga, Kobayashi and Li, Shengzhe and Nakamachi, Akifumi and Sato, Toshinori"},
year = {2023},
howpublished = {\url{http://github.com/line/LINE-DistilBERT-Japanese}}
}