CodeLMSec Benchmark
收藏资源简介:
CodeLMSec Benchmark是一个用于评估和比较代码语言模型生成易受攻击代码倾向性的数据集。该数据集由CISPA Helmholtz Center for Information Security创建,包含280个非安全提示,用于触发模型生成包含特定安全漏洞的代码。数据集的创建过程涉及使用先进的代码模型自动生成提示,并通过静态分析工具CodeQL验证生成的代码是否包含指定的安全弱点。该数据集旨在帮助研究人员和开发者评估和改进代码生成模型的安全性,解决自动代码生成中可能引入的安全漏洞问题。
CodeLMSec Benchmark is a dataset designed to evaluate and compare the propensity of code language models to generate vulnerable code. Created by the CISPA Helmholtz Center for Information Security, this dataset includes 280 non-security prompts used to trigger models to generate code with specific security vulnerabilities. The dataset's creation process involves automatically generating prompts using state-of-the-art code models, and verifying whether the generated code contains designated security weaknesses via the static analysis tool CodeQL. This benchmark aims to help researchers and developers evaluate and improve the security of code generation models, addressing the issue of potential security vulnerabilities introduced by automated code generation.




