Deploy an ML Model
Last updated
Last updated
There are multiple ways to deploy a model in Azure Machine Learning. One of the simplest ways is to use the designer to automate the deployment process. Use the following steps to deploy a model as a real-time endpoint:
Run your completed training pipeline at least once.
After the job completes, at the top of the canvas, select Create inference pipeline > Real-time inference pipeline.
The designer converts the training pipeline into a real-time inference pipeline. A similar conversion also occurs in Studio (classic).
In the designer, the conversion step also registers the trained model to your Azure Machine Learning workspace.
Your new pipeline looks like this.
Select Submit to run the real-time inference pipeline, and verify that it runs successfully.
After you verify the inference pipeline, select Deploy.
Enter a name for your endpoint and a compute type.
In the dialog box that appears, you can select from any existing Azure Kubernetes Service (AKS) Compute clusters to deploy your model to. If you don't have an AKS cluster, use the following steps to create one.
Select Compute in the left navigation to go to the Compute page.
On the navigation ribbon, select Kubernetes Clusters > + New > AksCompute
Select the Region and specifications for the compute instance.
Click Create.
After completing the above steps, go back to the inference pipeline and complete the Deploy dialogue box as shown in Step 6. Creation of the endpoint may take a while. The endpoint will be Transitioning while it is being created. Wait until the Deployment state of the endpoint changes from Transitioning to Healthy.
After deployment completes, you can see more details and test your endpoint:
Go to the Endpoints tab.
Select your endpoint.
Paste the URI from the above step in the browser to see the contents of swagger.json. The contents of swagger.json describe the REST API endpoints of the deployed model or web service. It defines how a client can interact with the deployed model. Copy these contents.
Open the Postman app and click on Import. Paste the contents of swagger.json from Step 4 into the text box.
The above step should import the endpoint into the Postman app workspace. Select the model name and set the variable initial and current value of baseURL under the Variables tab to the REST endpoint URL of the endpoint. You can obtain this from the Consume tab of the endpoint in Machine Learning Studio. Note that baseURL should be set to only the IP address part of the REST endpoint.
Next, select "Run ML Service" in the Postman app. You will need to expand all folder hierarchies in the left-side navigation panel to locate the "Run ML Service" option. Once selected, navigate to the Authorisation tab and select Bearer Token from the Type dropdown
Paste the Primary Key obtained from the Azure ML endpoint's Consume tab into the Token field.
Click on the Body tab. Select raw and ensure the format is set to JSON.
Enter the sample input data for your model and click the Send button. Postman will send the request to the Azure ML model endpoint. If the request is successful, you should receive a 200 OK response with a JSON response body containing the model's predictions.
Select the Details tab and copy the Swagger URI, which looks like as below:
Download and install . This app will allow you to test deployed model by sending sample inputs to the model and receive results from the model.
