QuixiAI/refusal-taxonomy
收藏Hugging Face2025-07-13 更新2025-08-09 收录
下载链接：
https://hf-mirror.com/datasets/QuixiAI/refusal-taxonomy
下载链接
链接失效反馈官方服务：
资源简介：
Quixi AI Refusal Taxonomy是一个全面的生产级拒绝分类系统，基于MLCommons Hazard Taxonomy和Llama Guard的示例进行了显著扩展和重构，适用于现实世界的部署场景。该分类系统提供了一种详细的分类方法，用于识别和分类有害用户提示，其类别和示例反映了在生产AI系统中观察到的实际威胁模式。该框架包含16个主要类别和300多个细粒度子类别，以支持开发高度准确和高效的安全模型。
Quixi AI Refusal Taxonomy is a comprehensive, production-grade refusal classification system based on the significantly expanded and restructured MLCommons Hazard Taxonomy and examples from Llama Guard, tailored for real-world deployment scenarios. This classification system provides a detailed classification method for identifying and categorizing harmful user prompts, with categories and examples reflecting actual threat patterns observed in production AI systems. The framework includes 16 major categories and over 300 granular subcategories to support the development of highly accurate and efficient safety models.
提供机构：
QuixiAI
