Jdd12/amnesty_qa

Amnesty QA数据集是一个基于Amnesty国际组织收集的报告构建的问答数据集，专门用于评估检索增强生成（RAG）系统。该数据集包含真实世界的人权内容，旨在测试和评估RAG流程。每个数据样本包含用户输入的问题、用于评估的真实答案、系统生成的答案以及检索到的相关上下文段落列表。数据集支持英语、印地语和马拉雅拉姆语三种语言，每种语言都有v3版本（推荐使用），同时包含已弃用的v1和v2版本（仅用于向后兼容）。数据集仅包含评估分割（eval split），包含20个精心策划的问题-答案对。

The Amnesty QA Dataset is a question-answering dataset constructed based on reports collected by Amnesty International, specifically designed for evaluating Retrieval-Augmented Generation (RAG) systems. This dataset contains real-world human rights content, aiming to test and assess RAG workflows. Each data sample includes the user's input question, the ground-truth answer for evaluation, the system-generated answer, and a list of retrieved relevant contextual passages. The dataset supports three languages: English, Hindi, and Malayalam, each with a v3 version (recommended for use), as well as deprecated v1 and v2 versions (only for backward compatibility). The dataset only includes the evaluation split (eval split), containing 20 carefully curated question-answer pairs.