Multilingual Guardrail Test Suite
收藏资源简介:
Multilingual Guardrail Test Suite是由宾夕法尼亚大学创建的综合性多语言测试套件,旨在评估大型语言模型(LLMs)在多语言环境中的安全性。该数据集包含七个子数据集,覆盖超过十种语言,主要用于测试和提升LLMs在处理多语言有害内容时的性能。数据集的创建过程包括将现有的英语安全数据集翻译成多种语言,并根据语言资源分布分为高、中、低资源组。该数据集的应用领域主要集中在提升LLMs在多语言环境中的安全性和可靠性,旨在解决多语言有害内容检测和防御的问题。
Multilingual Guardrail Test Suite is a comprehensive multilingual test suite developed by the University of Pennsylvania, which aims to evaluate the safety of Large Language Models (LLMs) in multilingual scenarios. This dataset includes seven sub-datasets covering more than ten languages, and is mainly used to test and improve the performance of LLMs when dealing with multilingual harmful content. The dataset creation process involves translating existing English safety datasets into multiple languages, and classifying them into high-, medium-, and low-resource groups based on the distribution of language resources. The application fields of this dataset mainly focus on enhancing the safety and reliability of LLMs in multilingual environments, with the goal of addressing the problems of multilingual harmful content detection and defense.




