declare-lab/OffTopicEval
收藏Hugging Face2025-10-08 更新2025-10-25 收录
下载链接：
https://hf-mirror.com/datasets/declare-lab/OffTopicEval
下载链接
链接失效反馈官方服务：
资源简介：
OffTopicEval是一个多语言的大型评估数据集，用于测量大型语言模型（LLM）在特定场景下的运营安全性。该数据集包括英语、中文和印地语三种语言，针对21个特定目的的AI代理，包含了领域内查询、直接领域外查询和自适应领域外查询。数据集旨在评估AI代理是否能够恰当地接受领域内查询以及可靠地拒绝领域外查询。
OffTopicEval is a large-scale multilingual evaluation dataset designed to measure the operational safety of large language models (LLMs) in specific scenarios. It includes English, Chinese, and Hindi languages, tailored for 21 purpose-specific AI agents, and comprises in-domain queries, direct out-of-domain queries, and adaptive out-of-domain queries. The dataset aims to evaluate whether AI agents can appropriately accept in-domain queries and reliably refuse out-of-domain queries.
提供机构：
declare-lab
