AWS sagemaker documentation change

Service: sagemaker · 2026-06-28 · Documentation low

File: sagemaker/latest/dg/neo-compilation-preparing-model.md

Summary

Added SageMaker Python SDK v3 code examples for model training with ModelTrainer and Neo compilation with ModelBuilder.optimize(), alongside existing v2 examples

Security assessment

The changes solely add new code examples demonstrating SDK usage patterns for model training and compilation. No security vulnerabilities, security configurations, or security features are mentioned or addressed in the added content.

Diff

diff --git a/sagemaker/latest/dg/neo-compilation-preparing-model.md b/sagemaker/latest/dg/neo-compilation-preparing-model.md
index c49de52f8..63e06d25b 100644
--- a//sagemaker/latest/dg/neo-compilation-preparing-model.md
+++ b//sagemaker/latest/dg/neo-compilation-preparing-model.md
@@ -283,0 +284,26 @@ For example, you can use a `sagemaker.TensorFlow` to define a TensorFlow estimat
+SageMaker Python SDK v3
+    
+    
+    
+    from sagemaker.train import ModelTrainer
+    from sagemaker.train.configs import Compute, SourceCode
+    
+    model_trainer = ModelTrainer(
+        training_image='insert appropriate TensorFlow training image URI',
+        role=role,  #param role can be arn of a sagemaker execution role
+        compute=Compute(instance_type='ml.c4.xlarge', instance_count=2),
+        hyperparameters={
+            'training_steps': '1000',
+            'evaluation_steps': '100',
+        },
+        source_code=SourceCode(entry_script='mnist.py'),
+    )
+
+Then train the model with the `.train` method:
+    
+    
+    model_trainer.train(inputs)
+
+SageMaker Python SDK v2 (Legacy)
+    
+    
@@ -302,0 +329,30 @@ Before finally compiling model with the build in `compile_model` method:
+SageMaker Python SDK v3
+    
+    
+    
+    # Neo compilation in V3 uses ModelBuilder.optimize() with compilation_config
+    from sagemaker.serve import ModelBuilder
+    
+    # Specify output path of the compiled model
+    output_path = '/'.join(model_trainer.output_path.split('/')[:-1])
+    
+    model_builder = ModelBuilder(
+        model=model_trainer,
+        role_arn=role,
+        instance_type='ml.c5.xlarge',
+    )
+    optimized_model = model_builder.optimize(
+        instance_type='ml.c5.xlarge',
+        output_path=output_path,
+        compilation_config={
+            'OverrideEnvironment': {
+                'OPTION_TARGET_INSTANCE_FAMILY': 'ml_c5',
+                'OPTION_FRAMEWORK': 'tensorflow',
+                'OPTION_FRAMEWORK_VERSION': '1.15.3',
+            },
+        },
+    )
+
+SageMaker Python SDK v2 (Legacy)
+    
+    
@@ -314,0 +371,59 @@ You can also use the `sagemaker.estimator.Estimator` Class to initialize an esti
+SageMaker Python SDK v3
+    
+    
+    
+    from sagemaker.core import image_uris
+    from sagemaker.core.helper.session_helper import Session
+    from sagemaker.train import ModelTrainer
+    from sagemaker.train.configs import Compute, InputData
+    
+    sagemaker_session = Session()
+    aws_region = sagemaker_session.boto_region_name
+    
+    # Specify built-in algorithm training image
+    training_image = image_uris.retrieve(framework='image-classification', 
+                              region=aws_region, image_scope='training')
+    
+    # Create ModelTrainer object for training
+    model_trainer = ModelTrainer(training_image=training_image,
+                                              role=role,  #param role can be arn of a sagemaker execution role
+                                              compute=Compute(
+                                                  instance_count=1,
+                                                  instance_type='ml.p3.8xlarge',
+                                                  volume_size_in_gb=50,
+                                              ),
+                                              output_data_config={'s3_output_path': s3_training_output_location},
+                                              base_job_name='image-classification-training'
+                                              )
+                                              
+    # Setup the input data_channels to be used later for training.                                          
+    train_data = InputData(channel_name='train', data_source=s3_training_data_location)
+    validation_data = InputData(channel_name='validation', data_source=s3_validation_data_location)
+    data_channels = [train_data, validation_data]
+    
+    # Train model
+    model_trainer.train(inputs=data_channels)
+    
+    # Neo compilation in V3 uses ModelBuilder.optimize() with compilation_config
+    from sagemaker.serve import ModelBuilder
+    
+    model_builder = ModelBuilder(
+        model=model_trainer,
+        role_arn=role,
+        instance_type='ml.c5.xlarge',
+    )
+    optimized_model = model_builder.optimize(
+        instance_type='ml.c5.xlarge',
+        output_path=s3_compilation_output_location,
+        compilation_config={
+            'OverrideEnvironment': {
+                'OPTION_TARGET_INSTANCE_FAMILY': 'ml_c5',
+                'OPTION_FRAMEWORK': 'mxnet',
+                'OPTION_FRAMEWORK_VERSION': '1.7',
+            },
+        },
+    )
+
+SageMaker Python SDK v2 (Legacy)
+    
+