Platform for AI (PAI) supports custom images. You can add images from Alibaba Cloud Container Registry to a PAI workspace and use the images in different modules of PAI. This topic describes how to add and view custom images.
Permission requirements
Alibaba Cloud account: An Alibaba Cloud account has the permissions to perform all operations that are described in this topic.
RAM user: A Resource Access Management (RAM) user must obtain the required permissions to perform the operations that are described in this topic. To grant the RAM user the required permissions, add the RAM user as a workspace member and assign the corresponding role. For information about the permissions of each role, go to the Roles and Permissions page. For information about how to add a workspace member, see Manage members of the workspace.
Add a custom image
Open a workspace in the PAI console and choose AI Computing Asset Management > Images in the left-side navigation pane. On the Image Management page, click the User-defined image tab.
Click Add Image.
In the Add Image panel, configure the parameters. The following table describes the parameters.
Parameter
Description
Image Name
The name of the image that you want to add. The name is displayed in the image list.
Image Type
Valid values:
Personal Mirror: provides basic container images for individual developers.
Enterprise Mirror: allows enterprises to manage and distribute Open Container Initiative (OCI) artifacts, such as container images, Helm charts, and Kubernetes operators, in a secure and efficient manner.
Enterprise Edition Mirror Instance
If you set the Image Type parameter to Enterprise Mirror, you must configure this parameter.
Select the Container Registry Enterprise Edition instance to which the custom image is pushed.
Image Namespace
Select a namespace. You can manage a collection of repositories in the namespace.
Image Repository
Select the image repository that you created. For information about how to create an image repository, see Main features of a repository and Create a repository and build images.
Image Version
Select an image version.
Use a custom domain name
This parameter is available only if you set the Image Type parameter to Enterprise Edition Image (Container Registry).
After you specify a custom domain name, you can use the domain name to access the Enterprise Edition instance over HTTP.
Visible range
The visibility of the custom image. Valid values:
Only visible to oneself: The image is visible only to you and the administrator of the current workspace.
Publicly visible within the workspace: The image is visible to all members of the current workspace.
CPU/GPU
You can select CPU or GPU to ensure that the image can run in a specific environment.
Description
The description of the image. Enter a description based on your business requirements to facilitate image management.
Click OK.
View a custom image
Open a workspace in the PAI console and choose AI Computing Asset Management > Images in the left-side navigation pane. On the Image Management page, click the User-defined image tab.
Find the image that you want to view in the image list and view the following information:
Image ID: Move the pointer over the name of the image to view the image ID.
Click the icon in the Image Address column to copy the image address of the image.
Public endpoint: Click View ACR images in the Operation column. On the Details page, you can view the public endpoint to access the image. Copy the endpoint to use the image for model training.
References
For information about the official images provided by PAI, see PAI public images.
After you configure images, resources, code builds, and datasets, you can submit a Deep Learning Container (DLC) job. For more information, see Before you begin.
You can configure custom metrics and APIs that regularly report custom metrics when you create a custom image. For more information, see Configure a custom monitoring and scaling metric.