Not so keen about LLMs or having a use case at hand per se, but just curious if 1 provider responded vectors are actually helpful (to do things like search or similarity) with another provider's models.
(My understanding of this might be a little skewed, I'm just learning on the go)
Absolutely, you can do this. This is the underpinning of most RAG apps 😄 Where folks use different combos of embedding models + LLMs sometimes both fine-tuned to their needs.
