Amazon SageMaker AI now supports instance preference lists for training and processing jobs
Amazon SageMaker AI introduces instance preference lists for training and processing jobs, simplifying compute capacity acquisition and enabling jobs to start sooner by allowing prioritized instance type selections.
Today, Amazon SageMaker AI introduced instance priority lists for training and processing jobs, simplifying compute capacity allocation for customer workloads. Previously, users could specify only one instance type per job submission, leading to unpredictable wait times during peak gpu demand and requiring complex retry logic. With this feature, users simply provide a prioritized list of acceptable instance types and sizes; SageMaker then runs the job on the first available configuration. This may accelerate job start times. To use it, specify instance type and count priorities when submitting training or processing jobs-for example, listing ml.g6.48xlarge (2 instances) or ml.g5.48xlarge (4 instances). SageMaker processes the list and launches the job with the first available capacity. Jobs can also be configured to procure capacity from on-demand sources or reserved SageMaker flexible training plans within the same submission. The feature streamlines compute acquisition during high-demand periods, reducing manual retries using the same apis already employed for SageMaker training and processing jobs. Instance priority lists are now available across all AWS regions where SageMaker operates, accessible via the SageMaker cli, apis, sdks, and console ui. For more information, see the documentation or launch blog.
Why it matters
This update benefits users who need to efficiently manage compute resources for AI training, fine-tuning, and data processing workloads.