Add support for Azure Open AI Embeddings to Jupyter AI

[gsrikant7]

This PR adds Azure Open AI embeddings provider to be able to use embedding use cases like /learn from jupyter-ai UI.

Related issue - #349

Caveats: It expects that the model_id and the deployment_name are same for the embeddings model. It currently does not support custom deployment names.

For example, to use text-embedding-ada-002 embedding model, create a deployment with name text-embedding-ada-002 and use it.

[gsrikant7]

Can someone please add the label to this PR? I am not seeing the option. Maybe because I do not have contributor/owner access.

[gsrikant7]

It would also be nice if someone can point to the necessary literature to add support for deployment names.

[srdas]

@gsrikant7 Thanks so much for opening this PR! I have reviewed the code changes and tested your branch and checked that the drop down menus now correctly show the required fields and options.

However, I do not have an API key for an Azure account, so cannot test if it works as intended with /learn and then with '/ask. Would you be able to add screenshots to demonstrate that the enhancement works as intended? (Show that the embedding model has learned a single doc and that /ask` delivers answers showing that the document information has been retrieved and presented in the response.)

[gsrikant7]

Thanks for reviewing. I have tested this and have shared the screenshot @srdas

[bjornarfjelldal]

Hi!

I am getting this error when trying /learn on Azure:

Learn documents in /home/jovyan/docs failed. 1 validation error for AzureOpenAIEmbeddingsProvider root Must provide one of the base_url or azure_endpoint arguments, or the AZURE_OPENAI_ENDPOINT environment variable (type=value_error).

I have created a deployment of text-embedding-ada-002 named text-embedding-ada-002.

Normal interaction works, so API key and endpoint is correctly set.

[albertmichaelj]

@bjornarfjelldal I had the same issue. If you set the environmental variable AZURE_OPENAI_ENDPOINT to the endpoint, then it will work. This feature is not fully properly integrated, but the environmental variable workaround seems to be functional at this point.

[bjornarfjelldal]

Hi, @albertmichaelj ! I forgot to mention that in the previous post, but I've tried to set

export AZURE_OPENAI_ENDPOINT=myendpointurl in the terminal. But I guess this might not catched by jupyter-ai allready running. Is the only solution then to add it to my dockerfile to be pre-loaded?

[albertmichaelj]

You will have to set the environmental variable when launching the Docker container. This can be done from the docker run command. The environmental variable won't take effect if you set it from the command line within the docker container (or if you set it in your environment before launching the docker container, since environmental variables are not inherited by docker containers).