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 2 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3 versions withdrawn
withdrawn after publishing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
add xgboost framework version 1.2-2
RegisterModel step and custom dependency support
One column per quarter.
### Testing and Release Infrastructure * disable smdataparallel tests
Add HF transformer version 4.5.0
support multiprocess feature group ingest
add experiment_config for clarify processing job
enable smdataparallel custom mpi options support
boto3 client param updated for feature store
Add support for accelerator in Clarify
Add environment variable support for SageMaker training job
disable profiler in some release tests
Enable Profiler in China Regions
added documentation for Hugging Face Estimator
add HuggingFace framework estimator
Allow users to send custom attributes to the model endpoint
move pandas to required dependency from specific use cases
return all failed row indices in feature_group.ingest
add support for TensorFlow 2.4.1 for training, inference and data parallel
support creating endpoints with model images from private registries
Change Estimator.logs() to use latest_training_job.name
disable profiler by default for regions not support it
Add Framework Version support for PyTorch compilation (Neo)
Use the output path to store the Clarify config file
Add tests for visualizer to improve test coverage
Add other Neo supported regions for Inferentia inference images
test_tag/test_tags method assert fix in association tests
add integration test for listing artifacts by type
Remove pytest fixture and fix test_tag/s method
SM DDP release notes/changelog files
create profiler specific unsupported regions
add support for Std:Join for pipelines
add artifact, action, context to virsualizer
increase time allowed for trial components to index
update predict_fn implementation for PyTorch EIA 1.5.1
### Bug Fixes and Other Changes * remove captureWarninig setting
improve optional dependency error message
remove shell=True in subprocess.check_output
Allow online store only FeatureGroups
Add support for actions in debugger rules.
Support local mode for Amazon SageMaker Processing jobs
remove D205 to enable PEP257 Docstring Conventions
add dataset definition support for processing jobs
add tensorflow 1.15.4 and 2.3.1 as valid versions
all de/serializers support content type
bump minor version for re:Invent 2020 features
use eia python version fixture in integration tests
use ECR-hosted image for ubuntu:16.04
fix failures for multiple spark run() invocations
create default bucket only if needed
ensure 1p algos are compatible with forward-port
clarify non-breaking changes after v1 forward port
update image uri for neo tensorflow
add kms_key optional arg to Pipeline.deploy()
### Bug Fixes and Other Changes * refactor _create_model_request
refactor out batch transform job input generation
include more notebook tests, logger to warn
forward-port v1 names as deprecated aliases
add a condition to retrieve correct image URI for xgboost
add xgboost framework version 1.2-1
### Features * upgrade Neo MxNet to 1.7
Your coding agent can read these notes before it upgrades. Set up the MCP server →