FaultLine数据集
收藏资源简介:
FaultLine数据集是一个多语言数据集,包含100个已知的Java、C和C++项目中的安全漏洞。该数据集旨在挑战LLMs对程序极其细微特性的推理能力,代表了基于LLM的代码推理的前沿。数据集的创建过程包括收集已知漏洞,并设计一系列精心设计的推理步骤,以生成PoV测试用例。数据集的应用领域是软件安全领域,旨在解决软件安全漏洞的验证和修复问题。
The FaultLine Dataset is a multilingual dataset containing 100 known security vulnerabilities across Java, C, and C++ projects. It is designed to challenge the reasoning capabilities of large language models (LLMs) regarding the extremely subtle characteristics of programs, representing the state-of-the-art in LLM-based code reasoning. The dataset creation process involves collecting known vulnerabilities and devising a set of meticulously crafted reasoning steps to generate PoV test cases. Its application domain lies in software security, with the core goal of addressing the verification and remediation of software security vulnerabilities.
数据集概述
基本信息
- 数据集名称:icse-26
- 托管平台:GitHub
- 托管地址:https://github.com/faultline-pov/icse-26
数据集内容
(根据提供的README文件内容,该数据集未提供具体描述或相关信息)

- 1FaultLine: Automated Proof-of-Vulnerability Generation Using LLM Agents哥伦比亚大学 · 2025年



