QQ on vector stores. I've been playing first with the native json storage method, then chromadb, and lancedb. Using the same data set that generated a 50 mb json file, all these methods seem to take a minute to load the index, and they don't max our the memory or CPU on the machine during load.
Do these "serverless" methods all just take a long time to load at first? Is the entire dataset loaded into memory?
If I move to postgres or similar I assume the "index load" will be much faster?
Is this due to how llama index interacts with these solutions versus a real server based one? When I see these serverless solutions making claims of data sets with millions of columns being performant, seems counter intuitive from this experience...
Nice, thanks for that. Yeah trying the full self-host method for now. Setup postgres last night but wasn't able to test too much. Qdrant is on deck next.
I had to change a bunch of things to get my code to work both locally and in the cloud. (mostly non-ai stuff, but I also switched to using an API for embeddings)
