How do I create a Dockerfile for BYOC training which allows me to pass in entry_point and source_dir arguments to SM Estimator?

Question

I am following this tutorial: https://github.com/aws/amazon-sagemaker-examples/blob/main/advanced_functionality/tensorflow_bring_your_own/container/Dockerfile

by SM to create a docker image. According to this tutorial, the training scripts are copied into the docker container meaning that you have to rebuild the container if you make any changes.

My question is: How do I create a Dockerfile such that I don't have to copy my script when I am building the image? Instead, I want to pass in the entry_point and source_dir arguments to the SM Estimator which copies those files upon calling the .fit method.

Answer

Hi there,

It seems like this should be possible through the sagemaker script mode, which allows you to pass in the entry_point argument through the SM estimator as can be seen [here](https://aws.amazon.com/blogs/machine-learning/bring-your-own-model-with-amazon-sagemaker-script-mode/)

Hopefully this resource will help provide some insight into your problem.
____

Re: so I kept looking and I believe [this](https://github.com/aws/amazon-sagemaker-examples/blob/main/advanced_functionality/custom-training-containers/script-mode-container-2/notebook/script-mode-container-2.ipynb) is a more appropriate resource to address your problem. Hopefully this one is what you are looking for.

Regards NN

How do I create a Dockerfile for BYOC training which allows me to pass in entry_point and source_dir arguments to SM Estimator?

Relevant content