Issue with populating api configs using environment variables after importing openai

[lumoslnt]

Describe the bug

I'm encountering an issue where I am unable to populate the api_key, api_base, api_type, and api_version using environment variables.

The openai package seems to read the api_key, api_base, api_type, and api_version from environment variables only at the moment it is imported. If these api configs are specified after the openai package is imported, they are not recognized.

The code is in here.

To Reproduce

Due to python's caching mechanism, to reproduce this issue, the openai package must be reloaded.

Code snippets

import openai
import os

os.environ["OPENAI_API_KEY"] = "<api_key>"
os.environ["OPENAI_API_BASE"] = "<api_base>"
os.environ["OPENAI_API_TYPE"] = "<api_type>"
os.environ["OPENAI_API_VERSION"] = "<api_version>"

openai.Completion.create(prompt="Hello", engine="text-davinci-003", max_tokens=5)

OS

Windows 11

Python version

Python v3.8.16

Library version

openai-python v0.27.8

[ehaca]

I faced the same Issue, at some point populating secrets from env seemed to stop working :(

[piiq]

yep. and this propagates to other projects that use openai under the hood

[RobertCraigie]

This issue should be fixed in our completely refactored v1 beta! Our new module-level client handles this case as expected.

While we still provide a module-level client, we highly recommend using the client instantiation based API which avoids this problem entirely.

from openai import OpenAI

client = OpenAI(
  api_key=os.environ['OPENAI_API_KEY'],  # this is also the default, it can be omitted
)

As an aside, if you're configuring environment variables like this then you should be able to update your code to directly assign values instead.

import openai

openai.api_key = "<api_key>"

[ehaca]

Nice, thank you - a client approach is exactly what I have been looking for all these months <3