Is there an example for upserting into Qdrant vector indices (i.e. do we still need to manually handle the ref_doc_ids)? Also are there plans to implement metadata filtering any time soon?

looking at the source code it seems like include_summary is deprecated? Is there still some way to use/does it still currently use the actual summary text during refinement?

Has anyone ever gotten a "ValueError: Vector store is required for vector store query." from saving and then loading a ComposableGraph (Tree index built on top of a bunch of SimpleVectorStore indices) from disk?

Two questions, for Qdrant as the vector store, does creating the index also pass the extra_info from the documents into Qdrant as the payload? And for PromptHelper, what would be the configuration to use if I want the output to use as much of the tokens not taken up by the prompt as possible?

I'm a bit confused about how the interactions between indices, documents, and vector stores work. I'm using QDrant atm. So I see from QdrantReader that it basically just assumes you're adding a payload keyed with 'text' to retrieve the documents from. For the StorageContext, how would I specify that I want to use Qdrant as the Document Store?

And to clarify, Qdrant would be able to store the vectors, the documents (kinda, by hacking the payload feature), but to store the created indices, would have to use something like S3? So the update/modify path would be basically insert new document into Qdrant, and then load the associated index from s3, do the insertion separately there, and re-save onto s3?

Just wanted to clarify for PromptHelper/ServiceContext, max_input_size is how many tokens an input can be, and num_output is how many tokens the output can be, right? Seems a bit strange for docs to show max_input_size as 4096 for gpt-3.5 as that's the max context length (which should be == max(input + output tokens)), we'd actually want max_input_size + num_output to equal 4096 max, correct?