MOS: Towards Effective Smart Contract Vulnerability Detection through Mixture-of-Experts Tuning of Large Language Models

This is dataset and source code for 'MOS'. MOS is a smart contract vulnerability detection framework based on mixture-of-experts tuning of large language models. The framework features three key innovations: continual pre-training on smart contract datasets, a MOE-Tuning dataset construction pipeline combining LLM generation with expert verification, and a vulnerability-aware routing mechanism directing analysis to specialized expert networks. This design aims to overcome the limitations of existing methods while providing reliable vulnerability detection and explanations.