Amazon SageMaker
Developer Guide

The AWS Documentation website is getting a new look!
Try it now and let us know what you think. Switch to the new look >>

You can return to the original look by selecting English in the language selector above.

Environmental Variables used by Amazon SageMaker Containers to Define Entry Points

When creating a Dockerfile, you must define an entry point that specifies the location of the code to run when the container starts. Amazon SageMaker Containers does this by setting an ENV environment variable. The environment variable that you need to set depends on the job you want to do:

  • To run a script,specify the SAGEMAKER_PROGRAM ENV variable.

  • To train an algorithm, specify the SAGEMAKER_TRAINING_MODULE ENV variable.

  • To host a model, specify the SAGEMAKER_SERVING_MODULE ENV variable.

You can use the Amazon SageMaker containers SDK package to set environment variables.


Train scripts similar to those you would use outside Amazon SageMaker using Amazon SageMaker Script Mode. It supports Python and Shell scripts: Amazon SageMaker uses the Python interpreter for any script with the .py suffix.Amazon SageMaker uses the Shell interpreter to execute any other script.

When running a program to specify the entry point for Script Mode, set the SAGEMAKER_PROGRAM environmental variable. The script must be located in the /opt/ml/code folder.

For example, the container used in the example in Get Started: Use Amazon SageMaker Containers to Run a Python Script sets this ENV as follows.


The Amazon SageMaker PyTorch container sets the ENV variable as follows.


In the example, is the program that implements the training algorithm and handles loading the model for inferences. For more information, see the Extending our PyTorch containers notebook.


When training an algorithm, specify the location of the module that contains the training logic by setting the SAGEMAKER_TRAINING_MODULE environment variable. An Amazon SageMaker container invokes this module when the container starts training. For example, you set this environment variable in MXNet as follows.


For TensorFlow, set this environmant variable as follows.


The code that implements this logic is in Amazon SageMaker Containers.


To locate the module that contains the hosting logic when deploying a model, set the SAGEMAKER_SERVING_MODULE environmental variable. An Amazon SageMaker container invokes this module when it starts hosting.

ENV SAGEMAKER_SERVING_MODULE sagemaker_mxnet_container.serving:main

The code that implements this logic is in the: Amazon SageMaker Containers.