AZURE_HOWTO.md

How to setup the app with Azure OpenAI Service

In addition to (mostly US-hosted) API model endpoints, we created an app version that runs with Azure OpenAI Services (AOAI). Find the docs here: https://learn.microsoft.com/en-us/azure/ai-services/openai/

Benefits

• OpenAI models can be deployed within the Swiss Azure Cloud Region (e.g. switzerlandnorth) or regions within the EU (like westeurope).
• This may help with achieving compliance and adress data residency considerations and governance.
• Some good reference documents:
  • Azure Product terms
  • Cloud Services in the Swiss Public Sector (Overview page)
  • Azure Services in the Public Sector in Switzerland - DE

Important information Always ensure appropriate classification and handling of your data, especially for sensitive and/or personal data. Check with your Microsoft contact or Cloud Service Provider (Partner) representative for further questions.

Prerequisites

• An active Azure account. If you don't have one, click here.
• In your account, an Azure Subscription that has been enabled for Azure OpenAI Services.
  • Apply here: https://aka.ms/oaiapply (takes ~24hrs to activate)

Walkthrough

• Go to https://portal.azure.com
• From the Catalog, choose Azure OpenAI service

• You need to select (or create) a new Resource Group, choose the Cloud Region and a unique service Name (becomes part of the endpoint url)
• Note: Not all model versions are available in all regions, see here. The region can't be changed afterwards (you'll need to delete & create a new service instance in that case).

• Follow the wizard to the end, and create the service.
• Once the service instance is ready, find the overview and Go to Azure OpenAI Studio.

• In the Azure Open AI Studio, go to Deployments and create a new model deployment.
• Note: If you can't create a particular model type, that model may be unavailable in your region, or your quota may be exceeded.

• Take note of the deployment name - this is needed later for making API calls

• Now go back and find the Endpoint and API Key. There are multiple ways to find these, but one is clicking on the gear icon on top of OpenAI Studio:

• Now add these credentials to your _streamlit_app/.env file:

AZURE_OPENAI_ENDPOINT=....
AZURE_OPENAI_API_KEY=....
AZURE_OPENAI_DEPLOYMENT=....

• Start the application (`cd _streamlit_app/ && python -m streamlit run ./sprache-vereinfachen_azure.py)

Congratulations - your good to go... 🎉

Known limitations

• While latest OpenAI models find their way to Azure quite fast, some may be restricted to higher tiers for capacity reasons or arrive later (e.g. GPT-4o as of June 2024). Open a support ticket or talk to your Microsot representative if in doubt.