NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
PyPI · #882 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 59 of the last 60 stable releases
3 versions withdrawn
withdrawn after publishing
9 years old
692 releases · first in 2017
### Bug fixes and other changes * exclude regions for P2 tests
add support for me-south-1 region
validation args now use default framework_version for TensorFlow
One column per quarter.
### Features * Add support for PyTorch 1.2.0
use default bucket for checkpoint_s3_uri integ test
add kwargs to create_model for 1p to work with kms
paginating describe log streams
### Documentation changes * model local mode
update tfs documentation for requirements.txt
### Bug fixes and other changes * update using_mxnet.rst
Revert "fix issue-987 error by adding instance_type in endpoint_name (#1058)"
preserve EnableNetworkIsolation setting in attach
### Bug fixes and other changes * re-enable airflow_config tests
lazy import of tensorflow module
add estimator preparation to airflow configuration
enable sklearn for network isolation mode
use new ECR images in us-iso-east-1 for TF and MXNet
expose kms_key parameter for deploying from training and hyperparameter tuning jobs
add support to TF 1.14 serving with elastic accelerator.
pass enable_network_isolation when creating TF and SKLearn models
expose vpc_config_override in transformer() methods
pass enable_network_isolation in Estimator.create_model
copy dependencies into new folder when repacking model
Estimator.fit like logs for transformer
update: disable efs fsx integ tests in non-pdx regions
clean up resources created by file system set up when setup fails
skip EFS tests until they are confirmed fixed.
change AMI ids in tests to be dynamic based on regions
skip efs tests in non us-west-2 regions
add logic to use asimov image for TF 1.14 py2
support training inputs from EFS and FSx
Add support for Managed Spot Training and Checkpoint support
eliminate dependency on mnist dataset website
add XGBoost Estimator as new framework
Refactor Using PyTorch topic for consistency
fix integration test failures masked by timeout bug
rework CONTRIBUTING.md to include a development workflow
prevent integration test's timeout functions from hiding failures
allow Airflow enabled estimators to use absolute path entry_point
update sklearn document to include 3p dependency installation
allow serving image to be specified when calling MXNet.deploy
waiting for training tags to propagate in the test
removing unnecessary tests cases
enable line-too-long Pylint check
deal with credentials for Git support for GitHub
allow custom model name during deploy
enable logging-format-interpolation pylint check
correct code per len-as-condition Pylint check
support Endpoint_type for TF transform
remove unnecessary failure case tests
allow only one integration test run per time
add git_config and git_clone, validate method
network isolation mode in training
update Sagemaker Neo regions and instance families
prevent race condition in vpc tests
### Bug fixes and other changes * Update setup.py
Add DataProcessing Fields for Batch Transform
add wait argument to estimator deploy
emit estimator transformer tags to model
use unique job name in hyperparameter tuning test
Your coding agent can read these notes before it upgrades. Set up the MCP server →