Batch Deployment
Explore how to create batch endpoints for scoring large datasets, deploy MLflow and registered models, and configure non-MLflow models with custom scoring scripts. Understand using token-based authentication and managing deployment resources and retries for effective batch processing in Azure Machine Learning.
We'll cover the following...
Creating a batch endpoint
We will create batch deployments in this lesson. We have learned that batch deployments enable us to score batches of data records at once. The syntax of batch endpoint creation is the same, except that we use the command batch-enpoint . We will create an endpoint using the token mode of authentication this time. Therefore, we set auth_mode to aad_token in the YAML file. Token-based authentication for the active directory (aad_token) ensures that web API requests are accompanied by a valid access token.
Let's create a batch endpoint—batch-endpt—using the job below. Use the endpoint name in the $BatchEndpoint variable and run the wizard.
$schema: https://azuremlschemas.azureedge.net/latest/batchEndpoint.schema.json name: batch-endpt description: my sample batch endpoint auth_mode: aad_token
We can visualize this endpoint in Azure Machine Learning studio, as shown below:
Deployment of an MLflow model
Now, we can deploy the models to the specific endpoint. Let's start with an MLflow model. This ...