graphrag-accelerator
graphrag-accelerator copied to clipboard
[BUG] Local search makes two LLM calls
I recently noticed that when executing a local search, there are two calls made to my azure openai model. While it is clear to me that the global search invokes multiple LLM calls, I am puzzled why the local search makes more than one single LLM call. This finding was robust across different indexes that I built.
Can you explain to me why the local search involves two instead of one LLM call?