SageMaker Deploy

The mental model for choosing between real-time, serverless, asynchronous, and multi-model/multi-container SageMaker endpoints - and which SDK config field marks each one.

Deploy a trained model to a real-time endpoint via SDK - register the model, configure and create the endpoint, wait for it to come up, and invoke it for predictions.

Provisioning a real-time SageMaker endpoint and scaling its instance count to traffic with Application Auto Scaling, via boto3 and AWS SDK v3.

Queue-backed inference for large payloads or slow predictions - AsyncInferenceConfig on the endpoint, InvokeEndpointAsync to submit, and polling S3 or SNS for the result.

A checklist for deploying SageMaker endpoints from the SDK - endpoint config and sizing, cost controls, auto scaling, choosing serverless/async/multi-model, safe rollouts, and monitoring - for boto3 and AWS SDK v3.