For standard datasets we use OpenAI ada embeddings, for hybrid its instructor + SPLADE. The HyDE toggle in the context variable feeds the query to a prompt firs

For standard datasets we use OpenAI ada embeddings, for hybrid its instructor + SPLADE. The HyDE toggle in the context variable feeds the query to a prompt first ("Generate a document that answers..."), before embedding. I think in the paper they use the contriever embedding model but we just use the ones supported on our platform