Please help explain the indexing pipeline and workflow of GraphRAG

[michaelnny]

We want to adapt the idea of GraghRAG to processing some private documents, all using local LLM + embedding model, and maybe make some changes to the overall pipeline. However, I have to be honest, after spending almost a week looking at the code, we still have no idea how the overall pipeline works. I've read the paper and have a basic idea of building the index, but the code make sense to me.

For example, under index\graph module, we have lots of different submodules to do entity extraction and building the graph,
however, I didn't find how we can put these different submodules together to build an end-to-end pipeline, that's from taking in some .txt of .csv input, to build the index files ready to run query. I found the json code in workdlows\default_workflows.py seems to be related to define the workflow, but it just don't make any sense to me.