夫子•明察 (fuzi.mingcha) is a Chinese legal large language model developed by Shandong University, Inspur Cloud, and China University of Political Science and Law. It aims to provide comprehensive and accurate legal consultation and Q&A services by leveraging a ChatGLM base and extensive legal data. The model supports legal article retrieval, case analysis, syllogistic reasoning for judgments, and legal dialogue.

How It Works

The model employs a generative retrieval paradigm, first citing relevant legal articles and then verifying them against an external knowledge base. This approach ensures responses are grounded in legal authority, enhancing their reliability. For case analysis, it retrieves similar historical cases to inform its judgment generation, making the output more reasonable and understandable. A key feature is its syllogistic reasoning capability, which analyzes case facts and legal regulations to predict judgment outcomes, mirroring the logical structure of legal arguments.

Quick Start & Requirements

• Installation: Model weights are available on HuggingFace and ModelScope. Deployment details can be found in src/cli_demo.py.
• Dependencies: Based on ChatGLM. Specific hardware requirements (e.g., GPU, CUDA) are not explicitly detailed but are implied for LLM deployment.
• Resources: Requires access to model weights and potentially a robust computing environment for inference.
• Links: HuggingFace Model: https://huggingface.co/SDUIRLab/fuzi-mingcha-v1_0, ModelScope: https://www.modelscope.cn/models/furyton/fuzi-mingcha-v1_0.

Highlighted Details

• Achieved first place in the Law Specific LLMs category on the LawBench evaluation (September 2023).
• Supports legal article retrieval, case analysis, and syllogistic reasoning for judgments.
• Trained on a large corpus of Chinese legal documents and supervised fine-tuning data.
• Training data includes legal statutes, judicial interpretations, case documents, legal Q&A, case retrieval, and syllogistic judgment datasets.

Maintenance & Community

The project is associated with Shandong University, Inspur Cloud, and China University of Political Science and Law. No specific community channels (Discord/Slack) or roadmap are provided in the README.

Licensing & Compatibility

The code is licensed under Apache-2.0. The model weights follow the ChatGLM-6B Model License. The project explicitly states that the content is for academic research only and not for commercial use or purposes that could cause harm.

Limitations & Caveats

The project's output is for reference only and does not constitute legal advice. Users requiring legal assistance should consult professional legal practitioners. The specific license for model weights may have restrictions.