newmindai/siu-rag-data

SIU-RAG数据集是一个专为评估检索增强生成（RAG）系统性能而设计的专用评估数据集，特别关注于分析使用引导解码方法的RAG性能。该数据集包含507条训练数据，每条数据包括问题、上下文、答案和文档ID等关键特征。数据集的创建采用了类似于mezura-eval-data数据集的方法，目的是为了评估引导解码方法在RAG系统中的性能，提供对系统利用提供上下文的效果的基准测量，以及分析生成的回答中文档引用的准确性，支持改进语言模型输出的事实基础研究。

The SIU-RAG dataset is a specialized evaluation dataset designed for benchmarking the performance of Retrieval-Augmented Generation (RAG) systems, with a particular focus on analyzing the performance of RAG with guided decoding methods. The dataset contains 507 training examples, each with key features such as question, context, answer, and document IDs. The dataset was created using a methodology similar to mezura-eval-data, aiming to evaluate the performance of guided decoding in RAG systems, provide a benchmark for measuring the effectiveness of systems in utilizing the provided context, analyze the accuracy of document references in generated responses, and support research on improving the factual grounding of language model outputs.