"DeepAudit code and data"

"Since the release of the Ethereum white paper in 2013, blockchain-based smart contract technology has received significant research and industrial interest due to its automatic execution and immutability. However, security vulnerabilities in smart contracts can lead to significant economic losses and pose a serious threat to the entire ecosystem. Currently, existing vulnerability detection tools and methods mainly focus on identifying specific vulnerability types, making it difficult to effectively address diverse vulnerability detection needs. Although some research has attempted to utilize large language models for smart contract vulnerability mining, their detection results remain unsatisfactory due to hallucination and reasoning degradation issuesof large language models. To address these challenges, this study proposes the DEEPAUDIT model, which integrates multiple complementary strategies, including such as multi-agent collaboration, ensemble learning, and adversarial generation, significantly improving the accuracy and comprehensiveness of vulnerability detection. Experimental results show that DEEPAUDIT achieves significant improvements in detection performance compared to traditional detection methods. DEEPAUDIT's outstanding advantage lies in its architecture design, which is entirely driven by a large language model and does not rely on specialized smart contract domain knowledge, fully demonstrating the model's broad versatility and superior performance across different application scenarios."