LlamaIndex
Log in

Find answers from the community

Updated 7 months ago

Hi,

HHammad
·
Hi,

Can anyone explain the concept behind retriever in the LlamaIndex, as per my knowledge the retriever retrieve the information from the documents by matching it with the query, and as per my knowledge this matching is done my cosine similarity. Please highlight if I ma right or wrong. The RAG concept was first given in the research paper "Retrieval-Augmented Generation for
Knowledge-Intensive NLP Tasks" in the research paper 2 formulas are used for retrieving which I have attached as pictures. I want to know that LlamaIndex (retriever = index.as_retriever(similarity_top_k=3)) works on which formula.
Attachments
image.png
image.png
L
H
4 comments
LLogan M
yea it uses cosine similarity to compare a query embedding to the embeddings of text chunks
LLogan M
other vector db integrations may use more approximate methods to help with scaling, like HNSW
HHammad
Hi Logan,
So, if I want to use dense passage retrieval what will be its code or class of retrieval in LlamaIndex?
LLogan M
That's just a default vector index 👍
Add a reply
Sign up and join the conversation on Discord