kyoro1/AzureML_HuggingFace

AzureML_HuggingFace

This repository intends to guide how we populate HuggingFace API with pre-defined models in Azure ML with sample codes. You find the basic description.

Assumption

• The following Azure resources are already provisioned:

  • Azure ML and related resources
  • User managed identity
  • python environment is prepared with these libraries. This repository was tested with python 3.9.

• Configuration prequisites: Please prepare the following contents for configuration as ./config.yml.

Model:
    registry_name: 'HuggingFace'
    model_name: 'bert-base-multilingual-uncased' ## Please find appropriate model in Model Catalogue in Azure ML

Azure:
    subscription_id: <YOUR SUBSCRIPTION ID in AZURE>
    resource_group: <YOUR RESOURCE GROUP>
    identity:
        client_id: <YOUR MANAGED IDENTITY> ## You may change the way of authentication such as service principal. In such a case, you need to modify the method `credentialManagedID` in ./src/utils.py
    ml:
        workspace_name: <YOUR AZURE ML WORKSPACE NAME>
        compute:
            instance_type: "Standard_DS2_v2"
            instance_count: 1

API:
    endpoint_name: <ENDPOINT NAME>
    deployment_name: <DEPLOYMENT NAME>

How to use

• Please follow the instruction in how-to-use.ipynb, which refers to the script.

Basic features

• Used manated identity for Entra ID(Azure AD) authentication in consuming populated Endpoints.