OpenDCAI/dataflow-demo-multihop-AgenticRAG

该数据集是DataFlow项目中Agentic Retrieval-Augmented Generation (AgenticRAG)管道的演示数据集，用于展示单跳输入问题与其对应的多跳、代理驱动推理输出之间的比较。数据集分为输入和输出两个部分：输入部分包含单跳问答样本，作为AgenticRAG代理的起点；输出部分展示了AgenticRAG管道如何将初始问题转化为多跳推理轨迹。每个样本包含多个跳（hop_1, hop_2, ...），每跳包含问题、答案、支持文档、最终问题、最终答案等信息。

This dataset serves as a demo for showcasing the Agentic Retrieval-Augmented Generation (AgenticRAG) pipeline of the DataFlow Project. It provides a clear comparison between single-hop input questions and their corresponding multi-hop, agent-driven reasoning outputs. The dataset is divided into two splits: the input split contains single-hop QA samples that serve as the starting point for the AgenticRAG agent, and the output split demonstrates how the AgenticRAG pipeline transforms an initial question into a multi-hop reasoning trajectory. Each sample consists of multiple hops (hop_1, hop_2, ...), where each hop includes the question, answer, supporting document, final question, final answer, etc.