v2.257.1

Bug Fixes

• Fix test failures with pytest and setuptools compatibility (#5574)

Dependency Updates

• Relax protobuf version constraint to <7.0 (#5573)

Enhancements

• Add telemetry for Feature Store (#5557)
• Add VERL (Versatile Reinforcement Learning) support (#5498)

v3.5.0

Features

• Feature Store v3: New version of Feature Store functionality
• Batch job listing by share identifier: Added support for listing Batch jobs filtered by share identifier
• Stop condition for model customization trainers: Added stopping condition support to model customization trainers
• EMRStep smart output: Enhanced EMR step output handling with smart output capabilities
• Transform AMI version support: Added support for specifying AMI version in SageMaker transform jobs

Enhancements

• Inference pipeline notebook example: Added example notebook demonstrating inference pipeline usage
• Migration documentation: Added migration documentation

Bug Fixes

• Model Customization bugs: Fixed multiple issues in Model Customization functionality
• Default stopping condition removal: Removed default stopping condition for MC trainer to prevent conflicts
• Instance groups parameter handling: Fixed issue where default instance_type/instance_count were incorrectly applied when instance_groups was set
• JumpStart alt config resolution: Resolved alternative configuration resolution for JumpStart models
• Inference processor naming: Updated inference processor identifier from 'inf2' to 'neuronx'
• HuggingFace Neuronx PyTorch version: Corrected the PyTorch version for HuggingFace Neuronx
• License additions: Added license to sagemaker-mlops and sagemaker-serve packages

v3.4.1

Fixes

• Pipelines: Correct Tag class usage in pipeline creation (#5526)
• ModelTrainer: Support PipelineVariables in hyperparameters (#5519)
• HyperparameterTuner: Include ModelTrainer internal channels (#5516)
• Experiments: Don't apply default experiment config for pipelines in non-Eureka GA
  regions (#5500)

Features

• JumpStart: Added ISO regions support (#5505)
• JumpStart: Added version 1.4 and 1.5 (#5538)

Chores

• Added unit and integration tests for JumpStart search functionality (#5544)
• Removed java-kotlin from CodeQL workflow (#5517)

v2.257.0

v2.257.0 (2026-02-03)

Features

• Update image URIs for DJL 0.36.0 release

v3.4.0

Features

• feat: add emr-serverless step for SageMaker Pipelines

Bug fixes and Other Changes

• Add Nova recipe training support in ModelTrainer
• Add Partner-app Auth provider
• Add sagemaker dependency for remote function by default V3

v2.256.1

Bug fixes and Other Changes

• Bug fixes remote function

v3.3.1

Bug fixes and Other Changes

• ProcessingJob fix - Remove tags in Processor while Job creation
• Telemetry Updates
• sagemaker-mlops bug fix - Correct source code 'dependencies' parameter to 'requirements'
• aws_batch bug fix - remove experiment config parameter as it Estimator is deprecated.

v2.256.0

Features

• Image for Numpy 2.0 support with XGBoost

Bug fixes and Other Changes

• Bug fix for Triton Model server for inference
• Removal of hmac key parameter for remote function
• Bug fixes for input validation for local mode and resource management for iterators

v3.3.0

Features

• AWS_Batch: queueing of training jobs with ModelTrainer

Bug fixes and Other Changes

• Fixes for model registry with ModelBuilder

v3.2.0

Features

• Evaluator handshake with trainer
• Datasets Format validation

Bug fixes and Other Changes

• Add xgboost 3.0-5 to release
• Fix get_child_process_ids parsing issue