Implementation of Graph RAG from Microsoft

Env

OS

Currently, this git repo has tried under both Windows and Ubuntu.

Python version

Python 3.10 - 3.12

Wheel requirement (Ubuntu users skip this step)

C++ compilor and Visual Studio tools.

You can download VS tools from the site below:

The VS tools need about 9 GB of storage in your computer.

Python package

pip install graphrag

or

pip install -r requirement.txt

How to run Graph RAG

Step 1 : Download Dataset

The command in Microsoft tutorial cannot successfully download the text file we need, so I write a python file to download and stored the text file.

Run the command below:

python3 grab.py

This action will build a folder input and stored the text file into book.txt in this folder.

Step 2 :

Run the command below:

graphrag init --root ./ragtest

This action will build up two files .env and settings.yaml and one folder prompts.

If you clone this git repo, you have to delete .env and settings.yaml before run the command.

Step 3 :

Change the value of GRAPHRAG_API_KEY in .env to your own OpenAI API key or Azure OpenAI key.

If you use Azure OpenAI, it is suggested to follow the steps in the site below:

https://www.cnblogs.com/gaocong/p/18299221

Step 4 :

Run the command below:

python -m graphrag.index

A series of parquet files will be created according to our document book.txt after the action.

By now, our Graph RAG has been set up successfully and is ready to be used.

Source Site

GraphRAG Getting Started