Poor performance and regular errors lately

I’ve noticed that the performance and reliability of the code-davinci-002 model has deteriorated quite a bit lately. API calls seem to be much slower and I’m quite often getting these two errors:

  • openai.error.RateLimitError: That model is currently overloaded with other requests
  • openai.error.APIError: The server experienced an error while processing your request
    Both with the python API. I also see similar errors using the HTTP API directly from C#.

Granted I’m using a lot of tokens in my queries (since fine-tuning is not available on codex yet) and am right up against the limit, so it could be related and maybe why I haven’t found this reported much lately on here. However, using the same queries even a few weeks ago was much more reliable and performant, so it feels like something has changed (since ChatGPT release maybe?).

Are other people having this problem? Is there maybe something on my end I can try?

Quick update on this, I noticed that the engine parameter was deprecated in favour of the model parameter when using openai.Completion.create. I’m not 100% sure yet, but at first glance it looks like the performance / reliability is much much better when using model. Can anyone confirm this is the case?

Been using the model parameter but started facing performance issues and errors two weeks ago. The model seems to be doing a lot better today. Possible they fixed issues / added more support for codex after a few busy weeks preparing for GPT-4

That’s interesting, probably just coincidence then - thank you :slight_smile: