
Amazon SageMaker Pipelines' new Selective Execution feature enables ML developers to run specific pipeline steps instead of entire workflows, saving time and compute resources while reusing outputs from reference runs.


<div><p>This article introduces Selective Execution, a new Amazon SageMaker Pipelines feature enabling users to run specific pipeline steps instead of entire workflows.</p><ul><li>Run individual or contiguous pipeline steps while reusing outputs from non-selected steps</li><li>Requires reference run with Successful, Failed, or Stopped status</li><li>Saves time and compute resources by eliminating unnecessary step execution</li><li>Supports runtime parameter modification for selected steps</li><li>Use cases: single step execution, multi-step training/evaluation, failed step recovery, pipeline branch testing</li><li>Requires SageMaker Python SDK version 2.162.0 or higher</li></ul><p>Selective Execution streamlines ML workflows by allowing developers to focus on specific pipeline sections, reducing iteration time and infrastructure costs.</p></div>


Related articles