Assistant can't read files

API
1

Hello, I’m trying to make my assistant read my csv OR json files since two days but he’s always telling me “sorry there’s currently an issue while reading files”. Web GPT 4 can read files so I’m puzzled. Uploaded files do not work both via API and from Assistant playground. No difference if the file is sent via user chat or attached to assistant directly. Assistant model doesn’t make any difference. Code interpreter and Retrieval are naturally on.

Am I the only one with these issues? (I know there are many threads like this but kinda 6+ days old mostly)

2

What is the chain of messages you use to when trying to read these files?

3

I simply go into the playground assistant, upload my json file and ask a question in the same chat. like “tell me the value for november 14th”. Web GPT4 can reply no issues. In the playground I only get an error message. Same File. Not sure if I replied to your question

4

It is very important what you put into the system prompt. Try to describe the file in the system prompt and also indicate its ID there.

5

I have problems with assistants with retrieval enabled, no code interpreter, has been for days. With the 3.5-based assistant delays are super long and I get empty replies. And it sometimes says it has trouble with retrieval.

I have a GPT-4 assistant with identical prompting, and the GPT-4 version maybe works better.
Both prompts casually mention the file by its original name. BUT: the 3.5-based assistant, originally copied from the GPT-4 version, lists the file in its resources with a long id, while the original assistant (GPT-4 version) lists the file with its original name.

Can it really be that mentioning files by id or name in the prompt is relevant for retrieval, or is it? When I asked the model last week, it seemed like it didn’t know the file ids or names from which the retrieval came, and it had no means to point to single files.

1 Like