Find answers from the community

Updated 2 months ago

Embeddings

Is it possible to use a Dense Passage Retriever (i.e. separate embedding models for queries and the indexed chunks/passages)

This would require a separate embedding model for the chunks/passages and the queries

Hypothetical document embeddings (HyDE) is trying to solve the same problem but seems like a hack. Unlikely to work on rare phrases, since the generative model wouldn't know the answer to the query
L
M
4 comments
Wouldn't using different embedding models mean that the vectors can't be compared? Because they live in completely different Semantic spaces?
These two models are trained to map different things (query and passage) to the same space
Oh neat, wasn't aware 😎
Add a reply
Sign up and join the conversation on Discord