Forge
收藏资源简介:
Forge是一个基于大型语言模型(LLM)的自动化框架,用于从真实世界的智能合约审计报告中提取高质量的安全漏洞信息,并将其分类到通用弱点枚举(CWE)分类中。该框架采用分而治之的策略和思维树技术,从非结构化的审计报告中提取结构化且独立的漏洞信息,并实现漏洞信息的层次分类。Forge框架通过分析6,454份真实世界的审计报告,构建了一个包含81,390个Solidity文件和27,497个漏洞发现的大规模数据集,覆盖了296个CWE类别。该数据集的构建过程高效且自动化,显著提高了数据集的规模和质量,为智能合约安全研究提供了重要的数据资源。
Forge is an automated framework based on a Large Language Model (LLM) that is designed to extract high-quality security vulnerability information from real-world smart contract audit reports and classify them into the Common Weakness Enumeration (CWE) categories. Utilizing a divide-and-conquer strategy and mind-tree technology, the framework extracts structured and independent vulnerability information from unstructured audit reports, achieving hierarchical classification of vulnerability information. Through the analysis of 6,454 real-world audit reports, Forge has constructed a large-scale dataset containing 81,390 Solidity files and 27,497 vulnerability discoveries, covering 296 CWE categories. The construction process of this dataset is efficient and automated, significantly enhancing the scale and quality of the dataset, providing an important data resource for smart contract security research.

- 1FORGE: An LLM-driven Framework for Large-Scale Smart Contract Vulnerability Dataset Construction中山大学,珠海,中国 · 2025年



