By using AWS re:Post, you agree to the Terms of Use

How to export tresained models to ECR as container image


I want to train and build the model in Sagemaker studio and then be able to export the model as a container image to ECR, so I can use the model in external platform by sharing the ECR image to another account where I Can create container with the image from ECR

1 Answer
Accepted Answer

The models you train in SageMaker are stored in S3 as .tar.gz files that you can use to deploy to an endpoint, or even test locally (extracting the model file from the tar file). If you are using a built-in algorithm, you can share the .tar.gz file to the second account and deploy the model in the second account, since built-in algorithm containers can be accessed from any AWS account.

If you are using a custom training image (docs here), you can push this image to ECR and allow a second account to pull the image and then use the image with the model that you have trained. However, note that Studio at this time does not support building Docker images out of the box. You can use SageMaker Notebook Instances instead.

I would recommend keeping the model (.tar.gz) and the image (Docker) separate, since you can easily retrain and deploy the newer versions of models without updating the image every single time.

answered 13 days ago
  • so escentially, If I need to host the model on a container outside of sagemaker, then I take the .tar.gz file and build a container image, correct ?

You are not logged in. Log in to post an answer.

A good answer clearly answers the question and provides constructive feedback and encourages professional growth in the question asker.

Guidelines for Answering Questions