Abstract
We propose DISC-LawLLM, an intelligent legal system utilizing large languagemodels (LLMs) to provide a wide range of legal services. We adopt legalsyllogism prompting strategies to construct supervised fine-tuning datasets inthe Chinese Judicial domain and fine-tune LLMs with legal reasoning capability.We augment LLMs with a retrieval module to enhance models' ability to accessand utilize external legal knowledge. A comprehensive legal benchmark,DISC-Law-Eval, is presented to evaluate intelligent legal systems from bothobjective and subjective dimensions. Quantitative and qualitative results onDISC-Law-Eval demonstrate the effectiveness of our system in serving varioususers across diverse legal scenarios. The detailed resources are available athttps://github.com/FudanDISC/DISC-LawLLM.