Is there a way to cusomize the scoring

Only if the vector db exposes a way to configure that

@Logan M Thank you. Is there documentation on how the scoring works when using the default VectorStoreIndex? Are the name, or metadata of a Node used when comparing to the query string, or just the content of a Node? Thx.

node.get_content(metadata_mode="embed") is what is sent to the embedding model

by default its just pairwise cosine similarity, nothing too fancy

Thank you, that is very helpful. It would be nice if we could customize what is sent to the embedding model, and what similarity score is used.

You can customize what is sent to both the embed model and llm
https://docs.llamaindex.ai/en/stable/module_guides/loading/documents_and_nodes/usage_documents.html#advanced-metadata-customization

Scoring function depends on the vector db. Some use cosine, some use distance, some use HNSW, some let you pick/change it

This is exactly what I was looking for, thank you. Btw, I don't know how you manage to address all these questions at all hours of the day, but it is much appreciated!

hahaha somehow I am good at multitasking