Azure OpenAI and OpenAI proxy support

[dlqqq]

Issues closed

• Closes Azure Open AI Support? #97
• Closes Azure OpenAI as Provider #245
• Closes Allow specifying OPENAI_API_BASE to override API location #321

Description

• Adds UI fields for other kwargs accepted by LangChain OpenAI providers: base API URL, OpenAI organization, etc.
• Provides Azure OpenAI support by implementing AzureChatOpenAIProvider class.
• Adds a new provider field model_id_label, which is used to indicate what should be used as the local model ID when using a registry provider.
  • SageMaker Endpoints: "Endpoint name"
  • Azure OpenAI: "Deployment name"

[dlqqq]

I've edited the model field labels to indicate whether each field is required or optional. There is a better way to do this, and I am tracking this in a separate issue: #323

[dlqqq]

@klauern @wanbao118 Unfortunately our team does not have an OpenAI proxy or an Azure deployment available to us. Can each of you test this branch to see if this fixes your respective issue?

I would recommend testing this branch via Conda:

# clone this branch
git clone [email protected]:jupyterlab/jupyter-ai.git
cd jupyter-ai/
git remote add dlqqq [email protected]:dlqqq/jupyter-ai.git
git fetch dlqqq
git checkout better-openai

# install locally in new Conda env
conda create -yn jai-test python=3.11
conda activate jai-test
pip install jupyterlab~=4.0
jlpm && jlpm build && jlpm dev-install
jupyter lab

[klauern]

Hello @dlqqq ! I think this works! I haven't tried all of the features, but this seems to indicate that it's using the alternate base API URL correctly:

and the logs seem to be showing fine as well:

[anammari]

I need to integrate the Azure Open AI LLMs into jupyter-ai as well. Any idea of the EDR?

Note: I've tried the steps above to test the better-openai branch but encoutered errors with uninstalling a package (urllib3 2.0.4). I'm working on Windows / WSL2.

[dlqqq]

@anammari If you could post the error trace, that would be helpful. We want to improve Windows compatibility, so any leads are appreciated. 🤗

[anammari]

@anammari If you could post the error trace, that would be helpful. We want to improve Windows compatibility, so any leads are appreciated. 🤗

Thanks @dlqqq

I couldn't log the error in WSL2 (VM crashed unexpectedly while executing the jlpm steps).

However I've tried again using Anaconda in Windows, which hasn't given me any errors this time.

I've tested the Q&A part and it seems to work.

Azure OpenAI credentials entry:

Testing Q&A: (Successful)

However the magic command has failed with:

AuthenticationError: Incorrect API key provided: ********************. You can find your API key at https://platform.openai.com/account/api-keys.

Is Azure OpenAI supported in magics?

[dlqqq]

@anammari Thank you very much for verifying the Azure's availability in the Chat UI! We're excited to make this experience available as fast as possible to our users, so we will be tracking magics support for Azure OpenAI in a separate issue. See: #341

[gdmcdonald]

Thanks for adding and merging the feature to use AzureOpenAI. I'm still confused on how exactly to get it going. Where should I look for instructions? It's not a supported model in the documentation.