DISC-LawLLM
DISC-LawLLM is an intelligent legal system developed by the Fudan-DISC laboratory, powered by large language models to deliver professional and comprehensive legal services. Featuring deep legal text processing, logical reasoning based on syllogistic theory, and retrieval-augmented knowledge adherence, it supports diverse scenarios including legal consultation, contract drafting, judicial tool assistance, exam preparation, and statute retrieval. The project provides high-quality resources such as the DISC-Law-SFT dataset for supervised fine-tuning, covering tasks like information extraction, judgment prediction, document summarization, and question answering. Two model variants, LawLLM-7B and DISC-LawLLM-13B, are available for deployment and fine-tuning. Performance benchmarks on LawBench demonstrate that DISC-LawLLM outperforms existing legal models and rivals GPT-4 in zero-shot and one-shot evaluations. The system is designed to assist legal professionals and the general public by combining robust legal rea