Create a custom SageMaker AI image - Amazon SageMaker AI
Create a SageMaker AI image from the consoleCreate a SageMaker AI image from the AWS CLI

Create a custom SageMaker AI image

Important

Custom IAM policies that allow Amazon SageMaker Studio or Amazon SageMaker Studio Classic to create Amazon SageMaker resources must also grant permissions to add tags to those resources. The permission to add tags to resources is required because Studio and Studio Classic automatically tag any resources they create. If an IAM policy allows Studio and Studio Classic to create resources but does not allow tagging, "AccessDenied" errors can occur when trying to create resources. For more information, see Provide permissions for tagging SageMaker AI resources.

AWS managed policies for Amazon SageMaker AI that give permissions to create SageMaker resources already include permissions to add tags while creating those resources.

Important

As of November 30, 2023, the previous Amazon SageMaker Studio experience is now named Amazon SageMaker Studio Classic. The following section is specific to using the Studio Classic application. For information about using the updated Studio experience, see Amazon SageMaker Studio.

This topic describes how you can create a custom SageMaker AI image using the SageMaker AI console or AWS CLI.

When you create an image from the console, SageMaker AI also creates an initial image version. The image version represents a container image in Amazon Elastic Container Registry (ECR). The container image must satisfy the requirements to be used in Amazon SageMaker Studio Classic. For more information, see Custom SageMaker AI image specifications. For information on testing your image locally and resolving common issues, see the SageMaker Studio Classic Custom Image Samples repo.

After you have created your custom SageMaker AI image, you must attach it to your domain or shared space to use it with Studio Classic. For more information, see Attach a custom SageMaker AI image.

Create a SageMaker AI image from the console

The following section demonstrates how to create a custom SageMaker AI image from the SageMaker AI console.

To create an image

  1. Open the Amazon SageMaker AI console at https://console.aws.amazon.com/sagemaker/.

  2. On the left navigation pane, choose Admin configurations.

  3. Under Admin configurations, choose Images.

  4. On the Custom images page, choose Create image.

  5. For Image source, enter the registry path to the container image in Amazon ECR. The path is in the following format:

    acct-id.dkr.ecr.region.amazonaws.com/repo-name[:tag] or [@digest]

  6. Choose Next.

  7. Under Image properties, enter the following:

    • Image name – The name must be unique to your account in the current AWS Region.

    • (Optional) Display name – The name displayed in the Studio Classic user interface. When not provided, Image name is displayed.

    • (Optional) Description – A description of the image.

    • IAM role – The role must have the AmazonSageMakerFullAccess policy attached. Use the dropdown menu to choose one of the following options:

      • Create a new role – Specify any additional Amazon Simple Storage Service (Amazon S3) buckets that you want users of your notebooks to have access to. If you don't want to allow access to additional buckets, choose None.

        SageMaker AI attaches the AmazonSageMakerFullAccess policy to the role. The role allows users of your notebooks access to the S3 buckets listed next to the checkmarks.

      • Enter a custom IAM role ARN – Enter the Amazon Resource Name (ARN) of your IAM role.

      • Use existing role – Choose one of your existing roles from the list.

    • (Optional) Image tags – Choose Add new tag. You can add up to 50 tags. Tags are searchable using the Studio Classic user interface, the SageMaker AI console, or the SageMaker AI Search API.

  8. Choose Submit.

The new image is displayed in the Custom images list and briefly highlighted. After the image has been successfully created, you can choose the image name to view its properties or choose Create version to create another version.

To create another image version

  1. Choose Create version on the same row as the image.

  2. For Image source, enter the registry path to the Amazon ECR container image. The container image shouldn't be the same image as used in a previous version of the SageMaker AI image.

Create a SageMaker AI image from the AWS CLI

You perform the following steps to create a SageMaker AI image from the container image using the AWS CLI.

  • Create an Image.

  • Create an ImageVersion.

  • Create a configuration file.

  • Create an AppImageConfig.

To create the SageMaker AI image entities

  1. Create a SageMaker AI image.

    aws sagemaker create-image \
    --image-name custom-image \
    --role-arn arn:aws:iam::<acct-id>:role/service-role/<execution-role>

    The response should look similar to the following.

    {
    "ImageArn": "arn:aws:sagemaker:us-east-2:acct-id:image/custom-image"
}

  2. Create a SageMaker AI image version from the container image.

    aws sagemaker create-image-version \
    --image-name custom-image \
    --base-image <acct-id>.dkr.ecr.<region>.amazonaws.com/smstudio-custom:custom-image

    The response should look similar to the following.

    {
    "ImageVersionArn": "arn:aws:sagemaker:us-east-2:acct-id:image-version/custom-image/1"
}

  3. Check that the image version was successfully created.

    aws sagemaker describe-image-version \
    --image-name custom-image \
    --version-number 1

    The response should look similar to the following.

    {
    "ImageVersionArn": "arn:aws:sagemaker:us-east-2:acct-id:image-version/custom-image/1",
    "ImageVersionStatus": "CREATED"
}
    Note

    If the response is "ImageVersionStatus": "CREATED_FAILED", the response also includes the failure reason. A permissions issue is a common cause of failure. You also can check your Amazon CloudWatch logs if you experience a failure when starting or running the KernelGateway app for a custom image. The name of the log group is /aws/sagemaker/studio. The name of the log stream is $domainID/$userProfileName/KernelGateway/$appName.

  4. Create a configuration file, named app-image-config-input.json. The Name value of KernelSpecs must match the name of the kernelSpec available in the Image associated with this AppImageConfig. This value is case sensitive. You can find the available kernelSpecs in an image by running jupyter-kernelspec list from a shell inside the container. MountPath is the path within the image to mount your Amazon Elastic File System (Amazon EFS) home directory. It needs to be different from the path you use inside the container because that path will be overridden when your Amazon EFS home directory is mounted.

    Note

    The following DefaultUID and DefaultGID combinations are the only accepted values:

    • DefaultUID: 1000 and DefaultGID: 100

    • DefaultUID: 0 and DefaultGID: 0 

    {
    "AppImageConfigName": "custom-image-config",
    "KernelGatewayImageConfig": {
        "KernelSpecs": [
            {
                "Name": "python3",
                "DisplayName": "Python 3 (ipykernel)"
            }
        ],
        "FileSystemConfig": {
            "MountPath": "/home/sagemaker-user",
            "DefaultUid": 1000,
            "DefaultGid": 100
        }
    }
}

  5. Create the AppImageConfig using the file created in the previous step.

    aws sagemaker create-app-image-config \
    --cli-input-json file://app-image-config-input.json

    The response should look similar to the following.

    {
    "AppImageConfigArn": "arn:aws:sagemaker:us-east-2:acct-id:app-image-config/custom-image-config"
}