SDK - Added re-ranking into vector search #1516
Conversation
| "boost": 1.0 | ||
| }, | ||
| } | ||
| }, |
There was a problem hiding this comment.
@montanalow How does this "rerank" key look?
query is the text to compare against.
model is the model to use
num_documents_to_rerank are the number of results to return from vector search and rerank against before limiting it to the limit parameter defined in the next section
| "rerank": { | ||
| "query": "Test document 2", | ||
| "model": "mixedbread-ai/mxbai-rerank-base-v1", | ||
| "num_documents_to_rerank": 100 |
There was a problem hiding this comment.
What about calling this just limit. Does llamaindex or transformers have a similarly named parameter name?
There was a problem hiding this comment.
Oh sorry missed this before merging. I think it might be a little confusing if we make it limit as we already have a limit key, and this isn't actually the limit. We already defined limit with llama index to mean the final number of items returned, but I'm not sure if they or langchain use it elsewhere.
| } | ||
| }, | ||
| "rerank": { | ||
| "query": "Test document 2", |
There was a problem hiding this comment.
Seems like query is being repeated a few places in this example, which may be pretty typical. One enhancement would be to move the query string out and reuse it everywhere, and make passing specific sub clause query strings optional. Not a launch blocker though.
There was a problem hiding this comment.
Got it, I will think more on making that optional and reusing it, but will merge this and get it out in the meantime.
No description provided.