HalluRAG
收藏资源简介:
HalluRAG数据集由明斯特大学创建,旨在用于检测大型语言模型(LLM)在封闭领域中的幻觉现象。该数据集包含18000条数据,主要来源于维基百科的最新内容,确保信息在模型训练截止日期后产生。数据集通过生成可回答和不可回答的问题对,结合模型的内部状态,训练多层感知器(MLP)以识别幻觉。HalluRAG数据集的应用领域主要集中在提高LLM的可靠性和信任度,特别是在RAG(检索增强生成)应用中,帮助模型避免生成不基于训练数据的错误信息。
The HalluRAG dataset was developed by the University of Münster for detecting hallucinations of Large Language Models (LLMs) in closed-domain scenarios. Comprising 18,000 entries, the dataset is primarily sourced from the latest Wikipedia content, ensuring that all information was generated after the model's training cutoff date. It pairs answerable and unanswerable question sets, and trains Multi-Layer Perceptrons (MLPs) by leveraging the internal states of LLMs to identify hallucinatory outputs. The primary applications of HalluRAG focus on enhancing the reliability and trustworthiness of LLMs, particularly in Retrieval-Augmented Generation (RAG) scenarios, where it helps models avoid generating erroneous information not grounded in their training data.

- 1The HalluRAG Dataset: Detecting Closed-Domain Hallucinations in RAG Applications Using an LLM's Internal States明斯特大学 · 2024年



