AWS adds partial container downloads to SageMaker training jobs

AWS has added support for SOCI-indexed container images to SageMaker AI training and processing jobs in all regions where SageMaker AI is available. AWS says the feature starts training containers in seconds and cuts GPU idle time during image downloads by up to 90%, without affecting training performance. The index lets SageMaker fetch files needed to start a container, then stream the rest while code runs. Users can select an indexed AWS image or add an index to their own image in a private Amazon ECR repository without rebuilding it. Existing job definitions need no changes. If the index cannot be used, SageMaker falls back to downloading the full image.