Hello, has anyone tried a ComposableGraph with two levels of vector indexes for many long documents (e.g. 100 documents, 100 nodes each).

Hello, may I ask why loading a pdf with SimpleDirectoryReader produces a list of documents per page? This seems counter intuitive and really the wrong behaviour. It should load a single document containing all of the pages?

Oh I see embeddings are used, but text DaVinci in service context is used for other LLM functions like multi step?