AWS sagemaker documentation change

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

File: sagemaker/latest/dg/train-warm-pools-how-to-use.md

Summary

Added SageMaker Python SDK v3 examples for creating warm pools and reusing warm pools, while retaining v2 legacy examples. Updated terminology from 'estimator' to 'training job'.

Security assessment

Changes are purely about SDK version updates and code examples for warm pool functionality. No security vulnerabilities, incidents, or security features are mentioned or addressed.

Diff

diff --git a/sagemaker/latest/dg/train-warm-pools-how-to-use.md b/sagemaker/latest/dg/train-warm-pools-how-to-use.md
index bebd71d0a..ac2940d39 100644
--- a//sagemaker/latest/dg/train-warm-pools-how-to-use.md
+++ b//sagemaker/latest/dg/train-warm-pools-how-to-use.md
@@ -47 +47,53 @@ This feature is available in the SageMaker AI [Python SDK v2.110.0](https://pypi
-To create a warm pool, use the SageMaker Python SDK to create an estimator with a `keep_alive_period_in_seconds` value greater than 0 and call `fit()`. When the training job completes, a warm pool is retained. For more information on training scripts and estimators, see [Train a Model with the SageMaker Python SDK](https://sagemaker.readthedocs.io/en/stable/overview.html#train-a-model-with-the-sagemaker-python-sdk). If your script does not create a warm pool, see [Warm pool creation](./train-warm-pools.html#train-warm-pools-creation) for possible explanations.
+To create a warm pool, use the SageMaker Python SDK to create a training job with a `keep_alive_period_in_seconds` value greater than 0 and start training. When the training job completes, a warm pool is retained. If your script does not create a warm pool, see [Warm pool creation](./train-warm-pools.html#train-warm-pools-creation) for possible explanations.
+
+SageMaker Python SDK v3
+    
+
+Use the SageMaker Python SDK to create a `ModelTrainer` with a `keep_alive_period_in_seconds` value greater than 0 in the `Compute` configuration and call `train()`. For more information on training with the SageMaker Python SDK, see [Train a Model with the SageMaker Python SDK](https://sagemaker.readthedocs.io/en/stable/).
+    
+    
+    from sagemaker.train import ModelTrainer
+    from sagemaker.train.configs import Compute, SourceCode
+    from sagemaker.core.helper.session_helper import Session, get_execution_role
+    from sagemaker.core import image_uris
+    from sagemaker.train.configs import InputData
+    
+    # Creates a SageMaker AI session and gets execution role
+    session = Session()
+    role = get_execution_role()
+    
+    # Retrieve the TensorFlow training image
+    training_image = image_uris.retrieve(
+        framework="tensorflow", region=session.boto_region_name,
+        version="2.2", py_version="py37",
+        instance_type="ml.g4dn.xlarge", image_scope="training"
+    )
+    
+    # Creates an example ModelTrainer with warm pool enabled
+    model_trainer = ModelTrainer(
+        training_image=training_image,
+        source_code=SourceCode(source_dir='code', entry_script='my-training-script.py'),
+        role=role,
+        compute=Compute(
+            instance_type='ml.g4dn.xlarge',
+            instance_count=1,
+            volume_size_in_gb=250,
+            keep_alive_period_in_seconds=1800,
+        ),
+        hyperparameters={
+            "batch-size": "512",
+            "epochs": "1",
+            "learning-rate": "1e-3",
+            "beta_1": "0.9",
+            "beta_2": "0.999",
+        },
+    )
+    
+    # Starts a SageMaker training job and waits until completion
+    train_data = InputData(channel_name="training", data_source='s3://my_bucket/my_training_data/')
+    model_trainer.train(input_data_config=[train_data])
+
+SageMaker Python SDK v2 (Legacy)
+    
+
+Use the SageMaker Python SDK to create an estimator with a `keep_alive_period_in_seconds` value greater than 0 and call `fit()`. For more information on training scripts and estimators, see [Train a Model with the SageMaker Python SDK](https://sagemaker.readthedocs.io/en/stable/overview.html#train-a-model-with-the-sagemaker-python-sdk).
@@ -85,0 +138,31 @@ Next, create a second matching training job. In this example, we create `my-trai
+SageMaker Python SDK v3
+    
+    
+    
+    # Creates a second ModelTrainer that reuses the warm pool
+    model_trainer_2 = ModelTrainer(
+        training_image=training_image,
+        source_code=SourceCode(source_dir='code', entry_script='my-training-script.py'),
+        role=role,
+        compute=Compute(
+            instance_type='ml.g4dn.xlarge',
+            instance_count=1,
+            volume_size_in_gb=250,
+            keep_alive_period_in_seconds=1800,
+        ),
+        hyperparameters={
+            "batch-size": "512",
+            "epochs": "2",
+            "learning-rate": "1e-3",
+            "beta_1": "0.9",
+            "beta_2": "0.999",
+        },
+    )
+    
+    # Starts a SageMaker training job and waits until completion
+    train_data = InputData(channel_name="training", data_source='s3://my_bucket/my_training_data/')
+    model_trainer_2.train(input_data_config=[train_data])
+
+SageMaker Python SDK v2 (Legacy)
+    
+