ai-sec-lab/mt-sec

MT-Sec是一个用于评估大型语言模型在多轮场景中生成的代码的正确性和安全性的基准测试。它由2376个多轮编码对话组成，每个对话包含三个用户轮次，这些对话是基于安全聚焦的单轮编码任务合成的。数据集分为三种交互类型：扩展（Expansion）、编辑（Editing）和重构（Refactor），每种类型都反映了不同的真实世界开发模式。

MT-Sec is a benchmark designed to evaluate the correctness and security of code generated by large language models (LLMs) in multi-turn scenarios. It consists of 2376 multi-turn coding conversations, each containing three user turns, synthetically constructed from a base of security-focused single-turn coding tasks. The dataset is organized into three interaction types: Expansion, Editing, and Refactor, each reflecting a distinct real-world development pattern.