Closed xianghappy closed 2 weeks ago
This is being marked as wontfix
as this is intentional. Document pinning is very helpful since it does full document comprehension. However, Document pinning is also a horrible idea for a large number of documents as one or all of the following will occur:
Your model context size is not big enough for all of the full documents, so they get truncated anyway so the chat completion can complete at least - omitting documents anyway.
Your model context can fit the documents, but you are burning so many tokens forcing so many useless tokens into the context - resulting in longer inference times and higher costs (if applicable)
This defeats the purpose of RAG, making citations extraneous if you pin everything. This makes knowing what citations were used nearly impossible.
Simply, document pinning is a solution for some use-cases, which is why it is supported since it can be very handy, but something is wrong if you are trying to pin every document - so the UX is to make it deliberate to discourage this
What would you like to see?
"Can you add a bulk pin feature to select multiple documents at once?"