NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
PyPI · #4409 most downloaded on PyPI
Open source library for training and deploying models on Amazon SageMaker.
Last release 10 days ago
24 Sep 2026
Ships on a steady schedule
a new release about every 2 weeks
Nearly every release is documented
notes for 31 of 32 stable releases
Nothing withdrawn
no release was ever pulled
10 months old
32 releases · first in 2025
One column per month.
update using_sklearn.rst parameter name
fix(train): enforce S3 ownership on ai_registry default bucket
add encryption option to "record_set"
### Features * Support for TFS preprocessing
prevent false positive PR test results
add document embedding support to Object2Vec algorithm
bug-fix: Handle StopIteration in CloudWatch Logs retrieval
bug-fix: Workflow: Revert appending Airflow retry id to default job name
__all__ from __init__.py filesSM_HPS environment variable in MXNet READMEinclude_cls_metadata default to False for everything except Frameworksfeature: Add 0.10.1 coach version
enhancement: Local Mode: add explicit pull for serving
feature: Estimator: add script mode and Python 3 support for TensorFlow
get_resolved_recipe() (#2034)is_multimodal utils function for multimodal data auto-detection (#2033)base_model_name param (#2085)model_package_group validation when HyperPod compute is provided in CPTTrainer (#2068).hyperparameters.* through recipe resolver (#2078)get_resolved_recipe() includes all overrides and displays as nested dictionary (#2084)feature: add support for sagemaker-tensorflow-serving container
fix: Address MTRL Eval Hyperparameters issue
feature: Estimator: add input mode to training channels
distributions for customizing distributed training with the new training script formatfeature: add support for TensorFlow 1.11.0
bug-fix: setting health check timeout limit on local mode to 30s
doc-fix: add deprecation warning for current MXNet training script format
feature: add support for TensorFlow 1.10.0
feature: Estimators: add support for MXNet 1.2.1
wait_timeout parameter to train() for SFT, DPO, RLAIF, RLVR, and BaseTrainerSAGEMAKER_HUB_NAME environment variable to override the HUB_NAME constantOutputDataConfig from ModelTrainer so kms_key_id, compression_type, and other fields are preservedNone hyperparameters in to_dict instead of converting them to the string "None"us-west-2 to Nova supported regionsbug-fix: removing PCA from tuner
bug-fix: get_execution_role no longer fails if user can't call get_role
create_model()ScriptProcessor and FrameworkProcessor, enabling SDK usage tracking for processing jobs via the telemetry attribution module (new PROCESSING feature enum added to telemetry constants)accept_eula handling in ModelBuilder's LoRA deployment path — previously hardcoded to True, now respects the user-provided value and raises a ValueError if not explicitly set to Truelambda_function.lambda_handler instead of deriving it from the source filename, which caused invocation failures when the source file had a non-default namefeature: Transformer: add support for batch transform jobs
feature: Add Chainer 4.1.0 support
feature: Add Support for PyTorch Framework
HyperparameterTuner to not include estimator metadata in jobbug-fix: Local Mode: Fix for non Framework containers
bug-fix: Remove __all__ and add noqa in __init__
EstimatorBase.attach()Removing experiment_config parameter for aws_batch as it is no longer needed with the removal of Estimator
* feature: Add chainer
feature: Add Support for Local Mode
feature: Estimators: add requirements.txt support for TensorFlow
requirements.txt support for TensorFlowmodel_package_group_name param to model_package_group in finetuning interfacesdataset param for benchmark evaluatorfeature: Estimators: add support for TensorFlow-1.5.0
sagemaker_timestamp when creating endpoint names in integration testspredictor.predict() in the JSON serializer to accept dictionariesNothing published for this version
Your coding agent can read these notes before it upgrades. Set up the MCP server →