GHSA-CySec：面向 Qwen3.5 的 GitHub 安全公告微调数据集

A large-scale conversational fine-tuning dataset for cybersecurity vulnerability analysis, derived from 32K+ GitHub Security Advisories (GHSA). Contains 66,393 ChatML-format samples spanning 12 package ecosystems. Each sample pairs a structured system prompt (professional cybersecurity analyst) with one of 7 query templates (overview, CVE query, severity, fix, package query, CWE query, analysis) and the corresponding advisory response including CVSS scores, affected versions, CWE classifications, and remediation steps. Designed for supervised fine-tuning of Qwen3.5 series models via LoRA. Split: 63,073 train / 3,320 validation.一个大规模网络安全漏洞分析对话微调数据集，由 32,000+ 条 GitHub 安全公告（GHSA）构建，共 66,393 条 ChatML 格式样本，覆盖 12 个包生态系统。每条样本包含专业网络安全分析师角色设定、7 种查询模板（概述、CVE 查询、严重性评估、修复方案、包查询、CWE 查询、综合研判）之一，以及对应的结构化公告回复（含 CVSS 评分、受影响版本、CWE 分类、修复措施）。适用于 Qwen3.5 系列模型的 LoRA 有监督微调。训练集 63,073 / 验证集 3,320。