You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
This repository has been archived by the owner on Nov 13, 2024. It is now read-only.
I was wondering how I could figure out which parts of the debug info > context > query result documents can be considered as relevant to the question asked. I assume that the debug info at hand is just the top_k results for a certain query it generated from the original question. Is there any kind of cosine score threshold that could be considered as "good" or is there by any chance knowledge about it in the knowledge base or context builder?
The LLM generates responses using the "sources", but I was wondering if there could be some kind of structured way of having a context to present the user with: "these are related sources to the question/answer".
Edit: This is not the case, the documents in the query results of the debug_info property are the top_k documents retrieved afaik, they're not necessarily good matches, just the best matching documents. So a sort of similarity threshold must be applied in order to keep the "good ones".
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
-
Hi there!
I was wondering how I could figure out which parts of the debug info > context > query result documents can be considered as relevant to the question asked. I assume that the debug info at hand is just the top_k results for a certain query it generated from the original question. Is there any kind of cosine score threshold that could be considered as "good" or is there by any chance knowledge about it in the knowledge base or context builder?
The LLM generates responses using the "sources", but I was wondering if there could be some kind of structured way of having a context to present the user with: "these are related sources to the question/answer".
Edit: This is not the case, the documents in the query results of the debug_info property are the top_k documents retrieved afaik, they're not necessarily good matches, just the best matching documents. So a sort of similarity threshold must be applied in order to keep the "good ones".
Beta Was this translation helpful? Give feedback.
All reactions