Hello I noticed some strange behavior, working in Postman and uploading PDF/JSON files to the assistant. When prompted about the file, the assistant does not get the answer. However, if we upload the file as .txt, it can read it. Is there a lag for retrieval with certain file types? Has anyone had this issue as well?
Thinking about just converting files to text but lose some context in doing so.
I’ve been sending jsons with great success with the message itself (prior to the last 24 hours), and have .txt file uploaded to create the assistant.
Which model are you using? I found gpt-4-1106 is significantly better with retrieval.
Have you compared the file sizes and data from PDF → JSON → txt? I found issues in my code a few weeks ago converting a csv → JSON which was duplicating data and causing the file size to balloon or the format to become sub-optimal.