Lab 5 - Deploy to Managed Endpoint

This lab may not work in an Azure Pass Subscription.

For this lab to work, the Microsoft.PolicyInsights needs to be registered in your subscription. Follow this tutorial

# Delete the compute cluster (only needed if you have an Azue subscription)
az ml compute delete -n gpu-cluster

In this lab you are going to deploy the model in managed endpoint.

Get the scoring script

# Create a directory
mkdir deploy
cd deploy

# Download the scoring script
wget https://raw.githubusercontent.com/GlobalAICommunity/back-together-2021/main/workshop-assets/amls/score.py

Stay in the same directory

Create the scoring environment

Create an empty yml file for the environment configuration.

code conda.yml

Add the content below to the conda.yml file.

name: project_environment
channels:
  - conda-forge
  - defaults
dependencies:
  - python=3.6.2
  - pip:
    - azureml-defaults
    - torch
    - torchvision
    - pillow==5.4.1

Create the endpoint configuration

Create an empty yml file for the deployment configuration.

code deployment.yml

Add the content below to the file

$schema: https://azuremlschemas.azureedge.net/latest/managedOnlineEndpoint.schema.json
name: <your-name>
type: online
auth_mode: key
traffic:
  version1: 100
deployments:
  - name: version1    
    app_insights_enabled: true
    model: azureml:SimpsonsClassification-pytorch:1
    code_configuration:
      code: 
        local_path: ./
      scoring_script: score.py
    environment: 
      name: simpsons-inference
      version: 1           
      path: .
      conda_file: file:conda.yml
      docker:
        image: mcr.microsoft.com/azureml/openmpi3.1.2-ubuntu18.04:20210727.v1
    instance_type: Standard_F4s_v2
    scale_settings:
      scale_type: Manual
      instance_count: 1
      min_instances: 1
      max_instances: 1

Replace: <your-name> with your unique name, like: henks-endpoint

Use the command below to create a managed endpoint

az ml endpoint create --file deployment.yml

Test the endpoint

Create a new file for the request json.

code request.json

Add the content below to the file

{
    "image":"https://raw.githubusercontent.com/hnky/dataset-lego-figures/master/_test/Bart.jpg"
}

Use the command below to invoke the endpoint!

az ml endpoint invoke -n simpsons-endpoint --request-file request.json

As a result you show see that it has seen Bar Simpson on the image.

Recap