My use case is to do a synthesis of relevant articles on a certain topic taken from a data base of articles. The document summary index sounds like the most adapted format. However itās not clear in the doc if you cache the document summary embedding instead of computing it every time. Any insight on this ?
Thanks thatās what I felt alsoā¦ but to me it looks like a major topic as you need to recalculate everything for every request which makes the data structure impractical >> where do you suggest I post the idea / comment ?
I mean, embeddings are fast and cheap to calcualte, so not terribly impractical š¤·āāļø but yea, could be improved. Feel free to post on our github issues, or even attempt a PR š
