deuch / python-azure-open-ai-sample

Azure ML Python sample with Azure Open AI

This repository provides very simple sample to use a notebook with Azure ML to connect and use Azure Open AI.

sample_mi.py uses a managed identity to authenticate to Open AI directly.

sample_mi_kv.py uses a managed identity to connect to a KeyVault to retrieve a secret.
This secret stores the api token of your Azure Open AI resource. This api key is used to connect to Azure OpenAI

Getting Started

1. Azure deployment

• Create an Azure ML Workspace and the compute
• Choose Python 3.8 Azure ML as kernel
• Assign an identity to the compute (system or managed)
• Add the Role Assignement to the Azure Open AI and/or KeyVault targeted resources
  • For Azure Open AI, use "Cognitive Service User" as Role
  • For KeyVault "Keyvault Secrets User" as Role
  • With System Assigned Identity, you need to use the principal named <azure_ml_workspace_name>/compute/<vm_name> for the role assignment

2. Python dependencies

Install dependencies in the notebook :

• !pip install openai
• !pip install azure-identity
• !pip install azure-keyvault-secrets

3. Set up the variables

In the notebook, please provide those informations :

• openai.api_base = "<URL of Azure OpenAI instance>"
• openai.api_version = "<API version for Azure Open AI>"
• engine = "<Name of your model deployment>"
• KVUri = "<KeyVault URL>"
• retrieved_secret = client.get_secret("<Secret name in KeyVault for azure open ai key>")