Using Azure and OpenAI at the same time

[rmilkowski]

Describe the bug

Currently it seems there is no strightforward way to be able to send requests to both OpenAI and Azure from same python program. Ideally this should be possible:

openai = OpenAI(key=key)
azureai = OpenAI(key=key, api_base=xxx, ....)

While currently given key, api_base, etc. are global variables to the module, the above won't work.

To Reproduce

Code snippets

No response

OS

Linux

Python version

Python 3

Library version

latest

[ellis-jc]

+1 on this.

Right now we have different models available on each side and using the two providers in tandem is a huge pain

[fabswt]

This is indeed annoying… I need Whisper from OpenAI but ChatCompletion from Azure.

@ellis-jc How do you guys handle it?

FYI, I posted this on StackOverflow: How to use the Python openai client with both Azure and OpenAI at the same time?

[sevenjay]

+1 on this.
We run both and a load balance in front of them, need this teature, too.

[Yaakov-Belch]

According to this response, there is a non-documented feature: Many API features accept named arguments such as:

import openai
openai.ChatCompletion.create(model=gpt_model_config.model_engine,
                             api_key=SECRET
                             api_base=BASE,
                             api_type=TYPE,
                             api_version=VERSION,
                           ...)

Setting the api_key and api_base per request may address your issue.

[rmilkowski]

According to this response, there is a non-documented feature: Many API features accept named arguments such as:

import openai
openai.ChatCompletion.create(model=gpt_model_config.model_engine,
                             api_key=SECRET
                             api_base=BASE,
                             api_type=TYPE,
                             api_version=VERSION,
                           ...)

Setting the api_key and api_base per request may address your issue.

I just tried it with openai.Image.create() and it works.
Thank you.

btw: looks like the upcoming 1.0.0 module version will finally support client instantiation. See #631