Struggling a lot with my basic pipeline. I have a milvus db that I already stored embeddings in. I checked using the milvus client that it does in fact have data in there. I can't for the life of me however figure out how to do basic rag using llamaindex, the concept sprawl and version fragility isn't as bad as langchain but it's slowly getting there. Can't find one basic end to end example of this anywhere. Can anyone explain why this is returning an empty response?
Not man vector stores work out of the box with existing data -- you'll need to either set a few variables, subclass the query method, or just write your own vector store tbh
Hey dude I got it working thanks for the help! One thing that might not be super helpful because my view of this is quite myopic, is that with the text_field vs text_key problem and the other problem I had was it was accepting system_prompt as an argument where it really wasn't a valid one, I think the ability to add kwargs arbitrarily is going to cause a lot of potential confusion. I wonder if instead of the interface was stricter and then additional arguments had to be strictly passed in via a dict etc. But anyways just happy rambling. Thanks for your help again
Yea it could be stricter -- although it can be a combination of stuff
pydantic classes allow arbitrary kwargs, which is annoying for pydantic models
many models actually use kwargs in some way (llm arguments, vector store constructor arguments, etc. -- it can be hard to type out every possible option sometimes)
