v2.99.0

Features

• heterogeneous cluster set up in distribution config
• support heterogeneous cluster for training
• include fields to work with inference recommender

Bug Fixes and Other Changes

• Moving the newly added field instance_group to the end of method
• image_uri does not need to be specified with instance_groups
• Loosen version of attrs dependency
• Add PipelineVariable annotation in estimatory, processing, tuner, transformer base classes
• model table link

Documentation Changes

• documentation for heterogeneous cluster

v2.98.0

Features

• Adding deepar image

Documentation Changes

• edit to clarify how to use inference.py

v2.97.0

Deprecations and Removals

• remove support for python 3.6

Features

• update prebuilt models documentation

Bug Fixes and Other Changes

• Skipping test_candidate_estimator_default_rerun_and_deploy
• Update model name from 'compiled.pt' to 'model.pth' for neo
• update pytest, skip hf integ temp
• Add override_pipeline_parameter_var decorator to give grace period to update invalid pipeline var args

v2.96.0

Features

• Add helper method to generate pipeline adjacency list

Bug Fixes and Other Changes

• changing trcomp integ tests to be able to run in all regions

v2.95.0

Features

• Adding Training Compiler support for TensorFlow estimator starting TF 2.9
• Add support for TF 2.9 training

Bug Fixes and Other Changes

• integs fallback from p3 to p2 instance
• bucket exists check for session.default_bucket
• make instance type fields as optional

Documentation Changes

• improvements on the docstring of ModelStep
• Add XGBoostProcessor

v2.94.0

Features

• AutoGluon 0.4.2 image_uris support

v2.93.1

Bug Fixes and Other Changes

• add input parameterization tests for workflow job steps
• add parameterized tests to transformer

v2.93.0

Features

• MxNet 1.9 support

Bug Fixes and Other Changes

• bump importlib-metadata version upperbound to support TF2.9
• fix pipeline doc code example where process.run only accepts argument
• Fix Tensorflow default model_dir generation when output_path is pipeline variable
• Support transformer data parameterization

v2.92.2

Bug Fixes and Other Changes

• turn off Pipeline Parameter inheritance from python primitives
• Add more validations for pipeline step new interfaces
• Changed method description per AWS request

v2.92.1

Bug Fixes and Other Changes

• pin protobuf to < 4.0 to fix breaking change