Arabic Dataset for LLM Safeguard Evaluation
收藏资源简介:
阿拉伯语大型语言模型安全评估数据集由MBZUAI等机构创建,包含5799个问题,旨在评估阿拉伯语环境下大型语言模型的安全性。数据集内容涵盖直接攻击、间接攻击和无害请求,涉及阿拉伯世界的文化和社会背景。数据集的创建过程包括翻译和本地化中国安全评估数据集的问题,并添加了3000多个区域特定的敏感问题。该数据集的应用领域主要集中在大型语言模型的安全评估,旨在解决阿拉伯语环境下模型生成有害内容的问题。
The Arabic Large Language Model (LLM) Safety Evaluation Dataset was developed by institutions including MBZUAI. Containing 5,799 questions, this dataset is designed to assess the safety of large language models within Arabic-speaking contexts. It covers direct attacks, indirect attacks, and harmless requests, and incorporates the cultural and social backgrounds of the Arab world. The dataset construction process involved translating and localizing questions sourced from a Chinese safety evaluation dataset, as well as adding over 3,000 region-specific sensitive questions. Primarily focused on the safety evaluation of large language models, this dataset aims to address the issue of harmful content generated by models in Arabic-speaking environments.
Arabic_safety_evaluation
概述
- 名称: Arabic_safety_evaluation
- 描述: 一个用于评估阿拉伯语大型语言模型(LLM)安全性的基准和评估框架。




