Andrea Cremese

A nerd with an MBA

Coming Up Next

Working list of papers and topics I want to explore. Not published to the site, but visible in the repo for interested readers.

Foundational Architecture Papers

Modern Applications

Organizational Design

One example of encoder: embedding models for a RAG

If you want to do a Retrieval Augmented Generation (HyDe too for example) you need to feed into the LLM (decoder) some of the sources for the “Augmented” part. Now, across all of your document, how do you decide what to put in? You can use a Encoder only and embed the query. So in this case the input query is transformed into a high degree vector (e.g. 3072 for text-embedding-3-large). You encode all of your documents (in chunks likely) and run a cosine similarity (how parallel are these vectors?).