google/frames-benchmark

FRAMES数据集是一个综合评估检索增强生成系统的数据集，用于测试系统在事实性、检索准确性和推理方面的能力。该数据集包含824个涉及多个领域知识的多跳问题，每个问题都附有推理类型标签，并提供黄金答案和相关维基百科文章。

FRAMES is a comprehensive evaluation dataset designed to assess the capabilities of Retrieval-Augmented Generation (RAG) systems in terms of factuality, retrieval accuracy, and reasoning. The dataset consists of 824 multi-hop questions that require information from 2 to 15 Wikipedia articles, spanning diverse topics such as history, sports, science, animals, and health. Each question is labeled with reasoning types including numerical, tabular, multiple constraints, temporal, and post-processing, along with gold answers and relevant Wikipedia articles.