Amazon SageMaker AI Announces Instance Preference Lists for Training Jobs

Amazon Web Services has announced new instance preference lists for Amazon SageMaker AI Training and Processing Jobs, allowing users to specify a ranked list of up to five acceptable instance types.

Capacity Provisioning and Manual Processes

Accessing the right GPU resources, especially during periods of high demand, is among the greatest challenges when training AI models at scale.

If operations were tied to a single specific GPU configuration, users previously had to manually try alternatives or wait.

Automatic Ranking and Evaluation

With the new feature, users can specify a ranked list of up to five acceptable instance types when creating a training or processing job.

Amazon SageMaker AI automatically evaluates this list in order of priority and initiates the process on the first instance with available capacity.

Key Features and Integration

The system evaluates multiple instance types through a single API call, eliminating manual retry loops and complex monitoring scripts.

It allows reserved capacity from Flexible Training Plans (FTP) to be tied to specific preferences while seamlessly falling back to on-demand capacity for others.