This custom image uses the official TensorFlow 2.3 image from DockerHub as a custom image in SageMaker Studio and demostrates how to bundle a file along with the custom image and access it in SageMaker Studio.
Build the Docker image and push to Amazon ECR.
# Modify these as required. The Docker registry endpoint can be tuned based on your current region from https://docs.aws.amazon.com/general/latest/gr/ecr.html#ecr-docker-endpoints
REGION=<aws-region>
ACCOUNT_ID=<account-id>
# Build the image
IMAGE_NAME=custom-tf23
aws --region ${REGION} ecr get-login-password | docker login --username AWS --password-stdin ${ACCOUNT_ID}.dkr.ecr.${REGION}.amazonaws.com/smstudio-custom
docker build . -t ${IMAGE_NAME} -t ${ACCOUNT_ID}.dkr.ecr.${REGION}.amazonaws.com/smstudio-custom:${IMAGE_NAME}
docker push ${ACCOUNT_ID}.dkr.ecr.${REGION}.amazonaws.com/smstudio-custom:${IMAGE_NAME}
Create a SageMaker Image with the image in ECR.
# Role in your account to be used for the SageMaker Image
ROLE_ARN=<role-arn>
aws --region ${REGION} sagemaker create-image \
--image-name ${IMAGE_NAME} \
--role-arn ${ROLE_ARN}
aws --region ${REGION} sagemaker create-image-version \
--image-name ${IMAGE_NAME} \
--base-image "${ACCOUNT_ID}.dkr.ecr.${REGION}.amazonaws.com/smstudio-custom:${IMAGE_NAME}"
# Verify the image-version is created successfully. Do NOT proceed if image-version is in CREATE_FAILED state or in any other state apart from CREATED.
aws --region ${REGION} sagemaker describe-image-version --image-name ${IMAGE_NAME}
Create a AppImageConfig for this image
aws --region ${REGION} sagemaker create-app-image-config --cli-input-json file://app-image-config-input.json
Create a Domain, providing the SageMaker Image and AppImageConfig in the Domain creation. Replace the placeholders for VPC ID, Subnet IDs, and Execution Role in create-domain-input.json
aws --region ${REGION} sagemaker create-domain --cli-input-json file://create-domain-input.json
If you have an existing Domain, you can also use the update-domain
aws --region ${REGION} sagemaker update-domain --cli-input-json file://update-domain-input.json