ulysses3753/secbot-training-data

Ulysses威胁响应训练语料库是一个全面的指令调优数据集，用于训练大型语言模型（LLMs）在网络安全威胁检测、事件响应和补救方面的能力。数据集由Daniel Howard策划，当前包含约2,500对数据，目标达到35,000对。数据集覆盖了系统内部（Windows、Linux、网络协议）、云安全等多个领域，并计划扩展到MITRE ATT&CK企业版、Living Off The Land等技术。数据格式为JSON，包含指令和输出对。数据集来源于公开资源，如MITRE ATT&CK、Sigma Rules等，并遵循相应的许可协议。

The Ulysses Threat Response Training Corpus is a comprehensive instruction-tuning dataset for training LLMs on cybersecurity threat detection, incident response, and remediation. Curated by Daniel Howard, it currently contains ~2,500 pairs (building toward 35,000). The dataset covers system internals (Windows, Linux, network protocols), cloud security, and plans to expand to MITRE ATT&CK Enterprise, Living Off The Land techniques, etc. Data is in JSON format with instruction-output pairs. Derived from public sources like MITRE ATT&CK, Sigma Rules, with appropriate licenses.