AWS sagemaker documentation change
Summary
Updated documentation to reflect SageMaker Python SDK v3 terminology (ModelTrainer instead of estimator) and added new prerequisite requiring EFS for TensorBoard functionality in domains created after June 1, 2026
Security assessment
The changes are focused on SDK version updates and functional requirements (EFS enablement). There is no mention of security vulnerabilities, patches, or security incidents. The EFS requirement is for functionality, not security hardening.
Diff
diff --git a/sagemaker/latest/dg/debugger-htb-prepare-training-job.md b/sagemaker/latest/dg/debugger-htb-prepare-training-job.md index 35c37b1ae..da1dbb15a 100644 --- a//sagemaker/latest/dg/debugger-htb-prepare-training-job.md +++ b//sagemaker/latest/dg/debugger-htb-prepare-training-job.md @@ -11 +11 @@ PrerequisitesStep 1: Modify your training script with open-source TensorBoard he -A typical training job for machine learning in SageMaker AI consists of two main steps: preparing a training script and configuring a SageMaker AI estimator object of the SageMaker AI Python SDK. In this section, you'll learn about the required changes to collect TensorBoard-compatible data from SageMaker training jobs. +A typical training job for machine learning in SageMaker AI consists of two main steps: preparing a training script and configuring a SageMaker AI ModelTrainer (formerly estimator in PySDK v2) object of the SageMaker AI Python SDK. In this section, you'll learn about the required changes to collect TensorBoard-compatible data from SageMaker training jobs. @@ -22,0 +23,2 @@ For instructions on setting up a domain, see [Onboard to Amazon SageMaker AI dom + * Amazon EFS must be enabled on your domain. TensorBoard requires an Amazon EFS file system to function. For domains created through quick setup after June 1, 2026, EFS is not created by default during domain creation. To enable EFS after domain creation, see [Amazon EFS creation and auto-mounting in Amazon SageMaker Studio](https://docs.aws.amazon.com/sagemaker/latest/dg/studio-updated-automount.html). + @@ -65 +67 @@ For Transformers with TensorFlow, use the `tf.keras.tensorboard.callback`, and p -You can also use a different container local output path. However, in Step 2: Create a SageMaker training estimator object with the TensorBoard output configuration, you must map the paths correctly for SageMaker AI to successfully search the local path and save the TensorBoard data to the S3 output bucket. +You can also use a different container local output path. However, in Step 2: Create a SageMaker training ModelTrainer object with the TensorBoard output configuration, you must map the paths correctly for SageMaker AI to successfully search the local path and save the TensorBoard data to the S3 output bucket. @@ -72 +74,9 @@ You can also use a different container local output path. However, in Step 2: Cr -## Step 2: Create a SageMaker training estimator object with the TensorBoard output configuration +## Step 2: Create a SageMaker training ModelTrainer object with the TensorBoard output configuration + +SageMaker Python SDK v3 + + +Use the `sagemaker.debugger.TensorBoardOutputConfig` while configuring a SageMaker AI ModelTrainer. This configuration API maps the S3 bucket you specify for saving TensorBoard data with the local path in the training container (`/opt/ml/output/tensorboard`). Pass the object of the module to the `tensorboard_output_config` parameter of the ModelTrainer class. The following code snippet shows an example of preparing a TensorFlow ModelTrainer with the TensorBoard output configuration parameter. + +SageMaker Python SDK v2 (Legacy) + @@ -85,0 +96,38 @@ This example assumes that you use the SageMaker Python SDK. If you use the low-l +SageMaker Python SDK v3 + + + + import os + from sagemaker.train import ModelTrainer + from sagemaker.train.configs import SourceCode + from sagemaker.core.debugger import TensorBoardOutputConfig + from sagemaker.train.configs import Compute + + # Set variables for training job information, + # such as s3_out_bucket and other unique tags. + ... + + LOG_DIR="/opt/ml/output/tensorboard" + + output_path = os.path.join( + "s3_output_bucket", "sagemaker-output", "date_str", "your-training_job_name" + ) + + **tensorboard_output_config = TensorBoardOutputConfig( + s3_output_path=os.path.join(output_path, 'tensorboard'), + container_local_output_path=LOG_DIR + )** + + model_trainer = ModelTrainer( + source_code=SourceCode(entry_script="train.py", source_dir="src"), + role=role, + training_image=image_uri, + compute=Compute(instance_count=1, instance_type="ml.c5.xlarge"), + base_job_name="your-training_job_name", + **tensorboard_output_config=tensorboard_output_config,** + hyperparameters=hyperparameters + ) + +SageMaker Python SDK v2 (Legacy) + +