Heyy Folks,
Have been encountering a few ‘You requested a model that is not compatible with this engine.’ errors since last night when calling the gpt-4 API via python.
Haven’t seen them since the days of da-vinci002. Anyone else having the same problems and any idea why they might be cropping up again ?
Can you share the API call request with all the hyper-parameters?
Function call →
response = openai.ChatCompletion.create(model=model, temperature=0, messages=messages)
where model=‘gpt-4’.
It’s a pretty standard call, been working for a long time, just got these error yesterday, have not encountered them since
OpenAI package has changed significantly in v1 upgrade. Did you upgrade the OpenAI the package recently?
I would change the generic gpt-4 model to point directly at the model name gpt-4-0613.
It is possible there are some shenanigans afoot in messing with the older model and what actual unpublished revision it points to. That could affect the alias differently than the model name that includes a version number.
No i have not. I’m still working with the older version by choice, but to your point, the v1 has been out since the dev conference and have not encountered this error until yesterday, but i’ll follow through on this.
Thanks on the suggestion
Yeah. That feels like that possible reasoning to me as well, though I would’ve expected any model running on the gpt-4 base to have a similar inference engine aside from the turbo maybe, which should not have been referencable via gpt-4 call but who knows what happens during a possible testing in the backend tbf
I’ve just started getting this error for the first time today. It’s not happening often- perhaps one out of ever few hundred calls- but it’s running in an iterative environment so it seems like most calls succeed and calls are identical except for the prompt.
We are also seeing this issue today. We’re using gpt-4-1106-preview.
Same. Started about 30 minutes ago for us. No change to model that we’re requesting.
Same here, also seeing it with gpt-4-1106-preview even though our code didn’t change.
This looks like a new issue.
I also have the same problem, no code change. But the error happens today.
Same here with gpt-4-0125-preview. It worked for the last 2-3 hours but then suddenly popped the error.
Same here, with gpt-4-0125-preview
+1 here using gpt-4-0125-preview
+1 here using gpt-4-0125-preview
+1 here using gpt-4-0125-preview and gpt-4-1106-preview. It’s flaky.
+1 on API calls for gpt-4-1106-preview
python
openai 0.28
Also it only appears to be happening on longer requests
Workaround: Retrying a failed request seems to work. So a simple retry loop is sufficient.
Sharing details like OpenAI library version anf language version would be helpful for OpenAI staff to help fix this.