Running TAO Toolkit on an AKS

Introduction

The Code used in this repo will deploy the Nvidia TAO Toolkit on an Azure AKS Cluster. The original reference can be found here: https://docs.nvidia.com/tao/tao-toolkit/text/tao_toolkit_api/api_setup.html#tao-toolkit-api-aks-setup

Prerequisites

To get a successful deployment you'll need following:

• Azure account and appropriate permissions
  • The original code is using Service Principal in various occassion. I kept the original approach and therefore you must have a Service Principal ready with the appropriate permissions.
• GitHub Secrets
  • You may use the code to deploy the components manually however I recommend the automated way with GitHub Action. Following Secrets must be therefore created to make this working:
    1. AZURE_CLIENT_ID
    2. AZURE_CLIENT_SECRET
    3. AZURE_TENANT_ID
    4. AZURE_SUBSCRIPTION_ID
    5. AZURE_RESOURCE_GROUP
    6. AZURE_STORAGE_ACCOUNT
    7. AZURE_CONTAINER
    8. NGC_API_KEY
    9. NGC_EMAIL
    10. SSH_PUBLIC_KEY

Getting Started

To get started with the deployment you first need to clone the repo:

• git clone https://github.com/yourusername/nvidia-quickstart-api-azure-aks.git
• cd nvidia-quickstart-api-azure-aks

Update the file terraform/terraform.tfvars with the appropiate values.

• NOTE: The variable Name is used to deploy the unique AKS cluster along with the terraform state file. If you plan to deploy a second instance then please update the name variable with a new unique name.

Check-in the code to get the workflow action initiated and verify the status on the action.

• NOTE: You can change the deployment yaml file to run the workflow only on a PR and not on every push.