跳到正文
HuggingFace Daily Papers(社區熱門論文)·· 24 天前AI 評分62

《我如何對齊你的模型》提出以模板擴大多語言安全訓練的方法

How I Aligned Your Model: Scaling Up Multilingual Safety Training with Templates

AI 導讀

研究團隊提出可擴展的模板式資料增強流程,並發布涵蓋 60 種語言的 MSafeTemp 語料庫,用於多語言安全對齊與評估。研究發現,僅以英語評估會誤判多語言安全:在英語中無法奏效的攻擊,翻譯後可奏效;加入文化本地化變體則會揭示單靠翻譯未能發現的失效。研究亦指出,以模板擴充安全訓練資料可提升多樣性和對齊質素,並改善英語以外的安全表現。

正文

Published on Oct 29

Authors:

Abstract

Researchers developed a scalable template-based pipeline to generate multilingual safety datasets, revealing that English-only evaluations misjudge multilingual model safety and that culturally localized training improves alignment across languages.

Safety alignment for Large Language Models is heavily focused on English with limited multilingual representation, particularly damaging low-resource language performance. Related open training datasets are often limited in size or do not follow region-localized harmful topics to broaden the scope of alignment beyond cultures. We propose a scalable template-based augmentation pipeline to produce massive multilingual refusal and benign corpora, extracting safe templates from harmful prompts, translating them and then filling in the slots with culturally authentic entities, both harmful and benign. As a result, we release MSafeTemp, a validated corpus spanning 60 languages useful for multilingual safety alignment and evaluation. Using the dataset, we show that English-only evaluation misjudges multilingual safety: for several models, attacks that fail in English succeed once translated, while culturally localized variants expose further failures that translation alone misses. We also confirm that augmenting safety training data through templates allows for higher diversity and quality of alignment, resulting in improved safety performance beyond English. MSafeTemp and the respective pipeline are aimed at facilitating research and development of safer and more responsible models for less-represented cultures.

View arXiv page View PDF Project page GitHub 1 Add to collection

Get this paper in your agent:

hf papers read 2610.40516

Don't have the latest CLI?

curl -LsSf https://hf.co/cli/install.sh | bash

Models citing this paper 0

No model linking this paper

Cite arxiv.org/abs/2610.40516 in a model README.md to link it from this page.

Datasets citing this paper 1

swiss-ai/MSafeTemp

Updated 5 days ago

•

1.1k

•

2

Spaces citing this paper 0

No Space linking this paper

Cite arxiv.org/abs/2610.40516 in a Space README.md to link it from this page.

Collections including this paper 1

來源:HuggingFace Daily Papers(社區熱門論文) · huggingface.co