NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
PyPI · #1039 most downloaded on PyPI
Open source library for training and deploying models on Amazon SageMaker.
Last release 4 days ago
18 Sep 2026
Ships on a steady schedule
a new release about every 2 weeks
Nearly every release is documented
notes for 58 of the last 60 stable releases
3 versions withdrawn
withdrawn after publishing
9 years old
691 releases · first in 2017
add EIA support for TFS 1.15.0 and 2.0.0
use valid encryption key arg for S3 downloads
One column per quarter.
remove .strip() from batch transform
use .format instead of os.path.join for Processing S3 paths.
correct local mode behavior for CN regions
fix documentation to provide working example.
support cn-north-1 and cn-northwest-1
### Features * inferentia support
Allow setting S3 endpoint URL for Local Session
Fix local _SageMakerContainer detached mode (aws#1374)
Add docs for debugger job support in operator
add us-gov-west-1 to neo supported regions
Check that session is a LocalSession when using local mode
make repack_model only removes py file when new entry_point provided
handle empty inputs/outputs in ProcessingJob.from_processing_name()
skip pytorch ei test in unsupported regions
Update Kubernetes Operator default tag
correct Estimator code_location default S3 path
change default compile model max run to 15 mins
fix PR builds to run on changes to their own buildspecs
### Bug Fixes and Other Changes * upgrade framework versions
use sagemaker_session when initializing Constraints and Statistics
### Bug Fixes and Other Changes * enable Neo integ tests
remove TF framework mode notebooks from PR build
don't use os.path.join for S3 path when repacking TFS model
allow download_folder to download file even if bucket is more restricted
enable network isolation for amazon estimators
fix HyperparameterTuner.attach for Marketplace algorithms
add documentation guidelines to CONTRIBUTING.md
remove NEO_ALLOWED_TARGET_INSTANCE_FAMILY
remove labels from issue templates
account for EI and version-based ECR repo naming in serving_image_uri()
disable Debugger defaults in unsupported regions
update py2 warning message since python 2 is deprecated
add additional information to Transformer class transform function doc string
Append serving to model framework name for PyTorch, MXNet, and TensorFlow
Use serving_image_uri for Airflow
Remove version number from default version comment
ignore private Automatic Model Tuning hyperparameter when attaching AlgorithmEstimator
honor 'wait' flag when updating endpoint
fix PyTorchModel deployment crash on Windows
allow disabling debugger_hook_config
create ProcessingJob from ARN and from name
use name_from_base in auto_ml.py but unique_name_from_base in tests.
allow setting the default bucket in Session
Revert "feature: allow setting the default bucket in Session (#1168)"
update PR template with items about tests, regional endpoints, and API docs
modify schedule cleanup to abide by latest validations
create auto ml job for tests that based on existing job
add support for TF 1.15.0, PyTorch 1.3.1 and MXNet 1.6rc0.
multiclass_classifierDocumentation for Amazon Sagemaker Operators
move sagemaker config loading to LocalSession since it is only used for local code support.
add pyyaml dependencies to the required list.
Remove local mode dependencies from required.
separating sagemaker dependencies into more use case specific installable components.
remove red from possible colors when streaming logs
clarify that source_dir can be an S3 URI
clarify how to use parameter servers with distributed MXNet training
use regional endpoint for STS in builds and tests
### Bug fixes and other changes * exclude regions for P2 tests
Your coding agent can read these notes before it upgrades. Set up the MCP server →