Update Azure OpenAI fields

[cloutier]

Description

I noticed Azure OpenAI wasn't working in our jupyterhub after an update, and I traced it to us having upgraded to openai 1.x.

I made the changes recommanded by the microsoft doc and I noticed there was also a regression regarding the env var. Unfortunately langchain uses the same env var for both azure and regular openai (sample code) which means people can't use both azure and openai at the same time with jupyter ai, although I think that is a very niche use case.

I have tested my patch in my org and it works now with these changes.

Related issues

• Closes As of openai>=1.0.0, Azure endpoints should be specified via the azure_endpoint param not openai_api_base #734

[welcome]

Thanks for submitting your first pull request! You are awesome! 🤗

If you haven't done so already, check out Jupyter's Code of Conduct. Also, please make sure you followed the pull request template, as this will help us review your contribution more quickly.

You can meet the other Jovyans by joining our Discourse forum. There is also a intro thread there where you can stop by and say Hi! 👋

Welcome to the Jupyter community! 🎉

[3coins]

Unfortunately langchain uses the same env var for both azure and regular openai (sample code) which means people can't use both azure and openai at the same time with jupyter ai, although I think that is a very niche use case.

@cloutier Thanks for making this update. Do you think this needs a fix upstream on the LangChain side? Is there a related issue opened there?

[cloutier]

I looked deeper into it, and I wasn't quite right, langchain uses both env var: https://github.com/langchain-ai/langchain/blob/f2579096993ae460516a0aae1d3e09f3eb5c1772/libs/partners/openai/langchain_openai/llms/azure.py#L85

I have ammended my commit to not change the env var and added a comment about that, but upstream still has the limitation that you can't use both openai and azure openai. They do have a TODO in the code about it.

[dlqqq]

@cloutier I reviewed the source code you linked, and think I understand the issue. Basically we assume the expected keyword argument to be the same as the environment variable name, i.e. AZURE_OPENAI_API_KEY should be passed as azure_openai_api_key=... to the LangChain provider class. However, for authentication, the Azure provider expects the keyword argument openai_api_key, despite preferring AZURE_OPENAI_API_KEY environment variable.

My question to you is: how are you currently authenticating when using Azure in Jupyter AI? Are you setting the AZURE_OPENAI_API_KEY as an environment variable to the jupyter lab process when starting it? I believe I can fix this issue, but I need to know what is currently working for you first.

[dlqqq]

Thank you for fixing this issue! Approved. ✅

My thoughts:

• The fields removed in this PR were deprecated in the openai>=1.0 upgrade in LangChain.
• The issue about AZURE_OPENAI_API_KEY is an existing issue with Jupyter AI; thank you for bringing this to my attention again.
• Therefore I'm merging this PR, and the API key issue can be addressed in a follow-up PR. 👍

[welcome]

Congrats on your first merged pull request in this project! 🎉

Thank you for contributing, we are very proud of you! ❤️

[dlqqq]

@meeseeksdev please backport to 1.x