|
--- |
|
language: |
|
- multilingual |
|
- en |
|
- ko |
|
- ar |
|
- bg |
|
- de |
|
- el |
|
- es |
|
- fr |
|
- hi |
|
- ru |
|
- sw |
|
- th |
|
- tr |
|
- ur |
|
- vi |
|
- zh |
|
tags: |
|
- deberta |
|
- deberta-v3 |
|
- mdeberta |
|
- korean |
|
- pretraining |
|
license: mit |
|
--- |
|
|
|
# mDeBERTa-v3-base-kor-further |
|
|
|
> ๐ก ์๋ ํ๋ก์ ํธ๋ย KPMG Lighthouse Korea์์ ์งํํ์์ต๋๋ค. |
|
> KPMG Lighthouse Korea์์๋, Financial area์ ๋ค์ํ ๋ฌธ์ ๋ค์ ํด๊ฒฐํ๊ธฐ ์ํด Edge Technology์ NLP/Vision AI๋ฅผ ๋ชจ๋ธ๋งํ๊ณ ์์ต๋๋ค. |
|
> https://kpmgkr.notion.site/ |
|
|
|
## What is DeBERTa? |
|
- [DeBERTa](https://arxiv.org/abs/2006.03654)๋ `Disentangled Attention` + `Enhanced Mask Decoder` ๋ฅผ ์ ์ฉํ์ฌ ๋จ์ด์ positional information์ ํจ๊ณผ์ ์ผ๋ก ํ์ตํฉ๋๋ค. ์ด์ ๊ฐ์ ์์ด๋์ด๋ฅผ ํตํด, ๊ธฐ์กด์ BERT, RoBERTa์์ ์ฌ์ฉํ๋ absolute position embedding๊ณผ๋ ๋ฌ๋ฆฌ DeBERTa๋ ๋จ์ด์ ์๋์ ์ธ ์์น ์ ๋ณด๋ฅผ ํ์ต ๊ฐ๋ฅํ ๋ฒกํฐ๋ก ํํํ์ฌ ๋ชจ๋ธ์ ํ์ตํ๊ฒ ๋ฉ๋๋ค. ๊ฒฐ๊ณผ์ ์ผ๋ก, BERT, RoBERTA ์ ๋น๊ตํ์ ๋ ๋ ์ค์ํ ์ฑ๋ฅ์ ๋ณด์ฌ์ฃผ์์ต๋๋ค. |
|
- [DeBERTa-v3](https://arxiv.org/abs/2111.09543)์์๋, ์ด์ ๋ฒ์ ์์ ์ฌ์ฉํ๋ MLM (Masked Language Model) ์ RTD (Replaced Token Detection) Task ๋ก ๋์ฒดํ ELECTRA ์คํ์ผ์ ์ฌ์ ํ์ต ๋ฐฉ๋ฒ๊ณผ, Gradient-Disentangled Embedding Sharing ์ ์ ์ฉํ์ฌ ๋ชจ๋ธ ํ์ต์ ํจ์จ์ฑ์ ๊ฐ์ ํ์์ต๋๋ค. |
|
- DeBERTa์ ์ํคํ
์ฒ๋ก ํ๋ถํ ํ๊ตญ์ด ๋ฐ์ดํฐ๋ฅผ ํ์ตํ๊ธฐ ์ํด์, `mDeBERTa-v3-base-kor-further` ๋ microsoft ๊ฐ ๋ฐํํ `mDeBERTa-v3-base` ๋ฅผ ์ฝ 40GB์ ํ๊ตญ์ด ๋ฐ์ดํฐ์ ๋ํด์ **์ถ๊ฐ์ ์ธ ์ฌ์ ํ์ต**์ ์งํํ ์ธ์ด ๋ชจ๋ธ์
๋๋ค. |
|
|
|
## How to Use |
|
- Requirements |
|
``` |
|
pip install transformers |
|
pip install sentencepiece |
|
``` |
|
- Huggingface Hub |
|
```python |
|
from transformers import AutoModel, AutoTokenizer |
|
|
|
model = AutoModel.from_pretrained("lighthouse/mdeberta-v3-base-kor-further") # DebertaV2ForModel |
|
tokenizer = AutoTokenizer.from_pretrained("lighthouse/mdeberta-v3-base-kor-further") # DebertaV2Tokenizer (SentencePiece) |
|
``` |
|
|
|
## Pre-trained Models |
|
- ๋ชจ๋ธ์ ์ํคํ
์ฒ๋ ๊ธฐ์กด microsoft์์ ๋ฐํํ `mdeberta-v3-base`์ ๋์ผํ ๊ตฌ์กฐ์
๋๋ค. |
|
|
|
| | Vocabulary(K) | Backbone Parameters(M) | Hidden Size | Layers | Note | |
|
| --- | --- | --- | --- | --- | --- | |
|
| mdeberta-v3-base-kor-further (mdeberta-v3-base์ ๋์ผ) | 250 | 86 | 768 | 12 | 250K new SPM vocab | |
|
|
|
## Further Pretraing Details (MLM Task) |
|
- `mDeBERTa-v3-base-kor-further` ๋ `microsoft/mDeBERTa-v3-base` ๋ฅผ ์ฝ 40GB์ ํ๊ตญ์ด ๋ฐ์ดํฐ์ ๋ํด์ MLM Task๋ฅผ ์ ์ฉํ์ฌ ์ถ๊ฐ์ ์ธ ์ฌ์ ํ์ต์ ์งํํ์์ต๋๋ค. |
|
|
|
| | Max length | Learning Rate | Batch Size | Train Steps | Warm-up Steps | |
|
| --- | --- | --- | --- | --- | --- | |
|
| mdeberta-v3-base-kor-further | 512 | 2e-5 | 8 | 5M | 50k | |
|
|
|
|
|
## Datasets |
|
- ๋ชจ๋์ ๋ง๋ญ์น(์ ๋ฌธ, ๊ตฌ์ด, ๋ฌธ์ด), ํ๊ตญ์ด Wiki, ๊ตญ๋ฏผ์ฒญ์ ๋ฑ ์ฝ 40 GB ์ ํ๊ตญ์ด ๋ฐ์ดํฐ์
์ด ์ถ๊ฐ์ ์ธ ์ฌ์ ํ์ต์ ์ฌ์ฉ๋์์ต๋๋ค. |
|
- Train: 10M lines, 5B tokens |
|
- Valid: 2M lines, 1B tokens |
|
- cf) ๊ธฐ์กด mDeBERTa-v3์ XLM-R ๊ณผ ๊ฐ์ด [cc-100 ๋ฐ์ดํฐ์
](https://data.statmt.org/cc-100/)์ผ๋ก ํ์ต๋์์ผ๋ฉฐ, ๊ทธ ์ค ํ๊ตญ์ด ๋ฐ์ดํฐ์
์ ํฌ๊ธฐ๋ 54GB์
๋๋ค. |
|
|
|
|
|
## Fine-tuning on NLU Tasks - Base Model |
|
| Model | Size | NSMC(acc) | Naver NER(F1) | PAWS (acc) | KorNLI (acc) | KorSTS (spearman) | Question Pair (acc) | KorQuaD (Dev) (EM/F1) | Korean-Hate-Speech (Dev) (F1) | |
|
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | |
|
| XLM-Roberta-Base | 1.03G | 89.03 | 86.65 | 82.80 | 80.23 | 78.45 | 93.80 | 64.70 / 88.94 | 64.06 | |
|
| mdeberta-base | 534M | 90.01 | 87.43 | 85.55 | 80.41 | **82.65** | 94.06 | 65.48 / 89.74 | 62.91 | |
|
| mdeberta-base-kor-further (Ours) | 534M | **90.52** | **87.87** | **85.85** | **80.65** | 81.90 | **94.98** | **66.07 / 90.35** | **68.16** | |
|
|
|
|
|
## KPMG Lighthouse KR |
|
https://kpmgkr.notion.site/ |
|
|
|
|
|
## Citation |
|
``` |
|
@misc{he2021debertav3, |
|
title={DeBERTaV3: Improving DeBERTa using ELECTRA-Style Pre-Training with Gradient-Disentangled Embedding Sharing}, |
|
author={Pengcheng He and Jianfeng Gao and Weizhu Chen}, |
|
year={2021}, |
|
eprint={2111.09543}, |
|
archivePrefix={arXiv}, |
|
primaryClass={cs.CL} |
|
} |
|
``` |
|
|
|
``` |
|
@inproceedings{ |
|
he2021deberta, |
|
title={DEBERTA: DECODING-ENHANCED BERT WITH DISENTANGLED ATTENTION}, |
|
author={Pengcheng He and Xiaodong Liu and Jianfeng Gao and Weizhu Chen}, |
|
booktitle={International Conference on Learning Representations}, |
|
year={2021}, |
|
url={https://openreview.net/forum?id=XPZIaotutsD} |
|
} |
|
``` |
|
|
|
## Reference |
|
- [mDeBERTa-v3-base-kor-further](https://github.com/kpmg-kr/mDeBERTa-v3-base-kor-further) |
|
- [DeBERTa](https://github.com/microsoft/DeBERTa) |
|
- [Huggingface Transformers](https://github.com/huggingface/transformers) |
|
- [๋ชจ๋์ ๋ง๋ญ์น](https://corpus.korean.go.kr/) |
|
- [Korpora: Korean Corpora Archives](https://github.com/ko-nlp/Korpora) |
|
- [sooftware/Korean PLM](https://github.com/sooftware/Korean-PLM) |