Tongyi-Zhiwen/DocQA-RL-1.6K
收藏Hugging Face2025-05-23 更新2025-05-31 收录
下载链接：
https://hf-mirror.com/datasets/Tongyi-Zhiwen/DocQA-RL-1.6K
下载链接
链接失效反馈官方服务：
资源简介：
🤗 DocQA-RL-1.6K是一个包含1600个DocQA问题的强化学习数据集，这些问题跨越三个推理领域：数学推理、逻辑推理和多层推理。数学推理问题需要对长篇专业文档进行数值推理，逻辑推理问题涉及对现实世界文档的逻辑分析，多层推理问题则侧重于跨文档推理。
🤗 DocQA-RL-1.6K is a reinforcement learning dataset consisting of 1,600 DocQA problems across three reasoning domains: Mathematical Reasoning, Logical Reasoning, and Multi-Hop Reasoning. The Mathematical Reasoning problems require numerical reasoning across long and specialized documents like financial reports. The Logical Reasoning problems involve logic analysis of real-world documents from domains such as legal, financial, insurance, and production. The Multi-Hop Reasoning problems emphasize cross-document reasoning by sampling examples from MultiHopRAG and Musique.
提供机构：
Tongyi-Zhiwen
