Repository navigation
v3.30.0 (2026-11-16)
Packages: sagemaker 3.30.0, sagemaker-core 2.30.0, sagemaker-train 1.30.0, sagemaker-serve 1.30.0, sagemaker-mlops 1.30.0.
New Features
- feature: add DataPreparer for data generation, quality evaluation, and transformation jobs (#6585)
- feat: support managed pipelines in Pipeline.start and Pipeline.get_all (#6581)
- feat(workflow): Add IfExpression and Omit for step arguments (#6567)
- feature: Add pipeline version alias support to Pipeline (#6574)
- feature: Add HubContentStep and Session.import_hub_content (#6565)
- feat: update SDK to use latest LMI v30 image for sdk v3.x (#6483)
- feat: Add
dataset_file_formatparameter to DataSet.create(). Defaults to "Custom" - feat: Add
document_schema_versionparameter to DataSet.create(). Defaults to "5.0.0" - change: update image_uri_configs 11-06-2026, 11-09-2026, 11-13-2026 and 11-16-2026
- change: Daily Sync with Botocore v1.44.2 (#6583)
Bug Fixes
- fix: preserve AgentCore context in RFT handlers (#6571)
- fix: validate RFT rewards before reporting rollout success (#6572)
- fix: Python 3.14 tuner tags (#6554) and tuner source staging bucket (#6555) (#6578)
- fix: apply ModelTrainer output KMS key to tuner source upload (#6555) (#6559)
- fix: expose local training result from ModelTrainer.train (#6556) (#6560)
- fix: create AI Registry boto3 clients lazily to avoid import-time region errors (#6558)
- fix: training-sessions - drop the static model list (#6568)
- fix: use the shifted input form in training session loss input examples (#6584)
- fix: use a distinct source URI for the FP pipeline version context (#6540)
- fix: Update DataSet.create() default document_schema_version from "2.0.0" to "5.0.0"
- fix: Produce v5-compatible hub content document instead of legacy v2 fields rejected by the v5 schema