graphrag-accelerator being archived?

[JFolberth]

The main page references leveraging the graphrag-accelerator to get started. However that solution has been listed as read only and is set to archive. What are the accurate steps to get started?

https://github.com/Azure-Samples/graphrag-accelerator

[ntsarb]

I had openned the same question the day the graphrag-accelerator was archived and did not receive any response. The graphrag-accelerator is a script for enterprise-scale deployment of graphrag, which is meant to be used as an example, as a starting point. It looks like it won't be maintained any further, but solution providers can still use it as a point of reference.

For a standalone GraphRAG, e.g. in WSL2 environment, first reserve the required llm and embeddings models on OpenAI or Azure OpenAI platforms. You will need the relevant details (base URL, API Key) for the graphrag configuration files.

Use conda to create an environment, and a folder where graphrag will initialise the settings (this is referred to as the root folder). Graphrag can be installed and initialised as explained here.

graphrag init --root <your_graphrag_root_folder>

To ingest documents (only text), save the text files in <your_graphrag_root_folder>/input. Then invoke graphrag index.

graphrag index --root <your_graphrag_root_folder>

It takes quite long to index documents. When the process is completed, you can run a query, like:

graphrag query --root <your_graphrag_root_folder> --method global --query "What are the top themes in this library?"

The results include citations in the form of IDs found in the Graphrag parquet files, in the output subfolder. For a large dataset, a proper, scalable database is needed. This is what the graphrag-accelerator script is useful for.

Currently, the local and global search methods work for me, the drift search method isn't. Not sure why; I've reported the issue here.

I hope the above helps.