I had a fully working chatbot using the Assistants API (code available in my github xamat/Xavibot but I can’t link to it here it seems). All of the sudden, yesterday I started to get weird answers. In fact I was getting the exact same response regardless of the message I was sending. A bit later in the day the bot stopped working. I have debugged the error and all I can see is that the Assistant Run comes back as “failed”. However run.last_error is NULL. So, there is no information about what failed in the run.
Again, this has happened without me changing anything on the server code, so I can only imagine that something has changed on the API/model side. I am using gpt-4-1106-preview as instructed btw. Has anything maybe changed on that model?
This sounds like a rate limit issue. Curious to know if it was resolved on its own the next day. I’m not entirely sure about the expected behavior once you hit past your usage limits.
I’m also getting a rate-limit for a single call on setting up the assisstant. This happens via python or in the open ai ‘assisstants’ console
Update here. The only way I could get my Assistant to “work again” was creating a new account with a new API key. It could be something on my older account having hit some limits. Unfortunately, the quotes on “work again” are because I have managed to get it to consistently work when debugging locally and calling the API from localhost, but my remote deployment still fails with the same issue (run fails with null error code). Still debugging the remote behavior. Will report back.
The reason that the remote deployment was not working was simply because the old key was cached. With the new key (meaning new OpenAI account) it does work. Not super stable, and very slow when doing RAG with GPT4, but it works. I still don’t know why the other account does not work since I had not hit any quota or spend limit, but at least I got it working.