A Comparison on Knowledge and Reasoning Intensive RAG-based Benchmarks

This comparison has categorized benchmarks/datasets based on their tasks include Web browsing,
such as BrowseComp; singlehop QA, such as TriviaQA; multi-hop QA, such as HotpotQA; multiple-choice QA, such as MMLU-Pro; mathematics, such as MATH etc.

本次对比分析依据任务类型对基准测试集与数据集进行了分类，涵盖网页浏览任务（如BrowseComp）、单跳问答任务（如TriviaQA）、多跳问答任务（如HotpotQA）、多项选择问答任务（如MMLU-Pro）以及数学类任务（如MATH）等。