Deploy the Open-Source ChatLaw Legal LLM on the SuanWang Platform

Model Overview

ChatLaw is an open‑source large language model for legal text, trained on 936,727 real Chinese legal cases. The model is designed to understand legal terminology and reduce hallucinations.

The companion embedding model, ChatLaw‑Text2Vec, is fine‑tuned on nearly one million authentic cases. It maps statutes and user queries to high‑dimensional vectors, enabling millisecond‑level similarity search.

Typical Retrieval‑Augmented Generation (RAG) use cases include building a legal vector database, intelligent error correction, and case‑law similarity retrieval.

Deployment on SuanWang Platform

Steps to obtain a GPU instance and run the models:

Open the SuanWang website (https://sumw.com.cn/), log in with a phone number and verification code.

Enter the compute market, select a GPU instance, locate the community image list, choose the ChatLaw image, pick the desired version, and confirm the rental.

After the instance starts, open the provided JupyterLab URL and log in.

Running the Embedding Demo

The demonstration script legal_emb_demo.py performs the following pipeline:

Activate the pre‑installed Python environment that includes torch_mlu and related dependencies: source /torch/venv3/pytorch_infer/bin/activate Change to the directory containing the model and scripts (adjust the path as needed): cd /mnt/ChatLaw-Text2Vec Execute the script: python legal_emb_demo.py The script loads the model on CPU, converts weights to FP16, transfers tensors to MLU, computes embeddings, and ranks results by similarity.