Amazon SageMaker AI now supports instance preference lists for training and processing jobs

Amazon SageMaker AI has introduced instance preference lists for training and processing jobs, enabling users to specify multiple instance types in priority order. Previously, customers could only submit a single instance type, leading to delays when preferred capacity was unavailable. The new feature allows SageMaker to automatically select the first available instance from the provided list, streamlining job initiation. This change addresses challenges during peak demand, particularly for GPU-intensive workloads where wait times are unpredictable.

To use the feature, users define their instance type and count preferences in priority order when submitting a job. For example, a preference list might include two instances of ml.g6.48xlarge or four instances of ml.g5.48xlarge. SageMaker then processes the list sequentially, launching the job on the first available configuration. This eliminates the need for manual retry logic or submitting multiple concurrent jobs to secure capacity.

The feature also supports capacity sourcing from on-demand sources or reserved SageMaker Flexible Training Plans within the same job submission. This integration ensures flexibility while maintaining existing workflows through SageMaker’s training and processing job APIs. Users can continue leveraging familiar tools such as CLIs, APIs, SDKs, and the Console UI to configure and submit jobs.

Instance preference lists for SageMaker training and processing jobs are available today in all AWS Regions where SageMaker is offered. The feature aims to simplify compute allocation during high-demand periods and reduce manual intervention. Additional details are available in the official documentation and launch blog.