SageMaker Hyperpod Flexible Training Plans now supports instant start times and multiple offers

As of February 14, 2025, SageMaker Flexible Training Plans now supports instant start times that allow customers to book a plan starting as soon as the next 30 minutes.
Amazon SageMaker‘s Flexible Training Plan (FTP) makes it easy for customers to access GPU capacity to run ML workloads. Customers who use Flexible Training Plans can plan their ML development cycles with confidence in knowing they’ll have the GPUs they need on a specific date for the amount of time they reserve. There are no long-term commitments, so customers get capacity assurance while only paying for the amount of GPU time necessary to complete their workloads.

With the ability to start a reservation within 30 minutes (subject to availability), Flexible Training Plan accelerates compute resource procurement for customers running machine learning workloads. The system first attempts to find a single, continuous block of reserved capacity that precisely matches a customer’s requirement. If a continuous block isn’t available, SageMaker automatically splits the total duration across two time segments and attempts to fulfill the request using two separate reserved capacity blocks. Additionally, with this release, Flexible Training Plan will return up to three distinct options, providing flexibility in compute resource procurement.

You can create a Training Plan using either the SageMaker AI console or programmatic methods. The SageMaker AI console offers a visual, graphical interface with a comprehensive view of your options, while programmatic creation can be done using the AWS CLI or SageMaker SDKs to interact directly with the training plans API. You can get started with the API experience here.