I wonder at what point it will be ~as much overhead to pass through a subset of the data with a small yet capable and fast LLM vs. using a crude dot product when doing retrieval
I think a combination works quite well: first getting a small set of candidates from all the data using a lightweight model, and the using a heavy-duty model to rerank the results and get the final candidates.
This seems awesome for enabling RAG queries for on-device LLMs.
I wonder at what point it will be ~as much overhead to pass through a subset of the data with a small yet capable and fast LLM vs. using a crude dot product when doing retrieval
I think a combination works quite well: first getting a small set of candidates from all the data using a lightweight model, and the using a heavy-duty model to rerank the results and get the final candidates.
10K embeddings generated in under 700 milliseconds!!!