Help Center/ ModelArts/ Image Management/ Image Registration and Management
Updated on 2026-07-27 GMT+08:00

Image Registration and Management

ModelArts allows you to register ModelArts unified images or other images uploaded to SWR. A registered image can be used for model training, real-time inference, and notebook instances. This section describes how to register an image and view, edit, use, and delete a registered image.

Prerequisites

You have obtained the ModelArts unified image address or the SWR address of another image.

Registering an Image

  1. Log in to the ModelArts console. In the navigation pane on the left, perform the following operations:
    • New console: Choose Asset Management > Images. In the Registered Images tab, click Register.
    • Old console: Choose Asset Management > Image Management. In the Registered Images tab, click Register.
  2. On the Register Image page, set the following parameters and click Register Now.
    Figure 1 Registering an Image

    Table 1 Parameters

    Parameter

    Description

    SWR Source

    Use one of the following methods to configure the image source:

    • Method 1: Enter the SWR address of the ModelArts unified image or other images obtained in Prerequisites.

      Click Check Available Images to go to the My Images page of the SWR console. Click the image name. On the Image Tags tab page, obtain the SWR image address from the Pull Command column. Example image path:

      <swr-domain-name>/<namespace>/<repository>:<tag>
    • Method 2: Click , select an image from either SWR or SWR Enterprise Edition tab, and click OK.

      For details about SWR images, see What Is SWR?

    Tags

    Specify tags for the image. You can add a maximum of 20 tags.

    Version Description

    Enter a description for the image version. It can contain up to 256 characters and cannot include the following special characters: &<>"'/

    Architecture

    Select ARM or x86_64 based on your requirements. The architecture must match that of the image source.

    You can run the docker run --rm {image_name} uname -m command to view the architecture. Replace image_name with the actual image name. Example:
    docker run --rm swr.cn-north-4.myhuaweicloud.com/dev-custom/notebook-test-cc:v5 uname -m

    As shown in the following figure, the architecture of the image is x86.

    Figure 2 Viewing the image architecture

    Type

    Select NPU, CPU, or GPU based on your requirements. The type must match that of the image source.

    You can check whether the CUDA or CANN package is installed by viewing the image address. The CUDA package is for GPU, and the CANN package is for NPU.

    For example, the image address is swr.cn-north-4.myhuaweicloud.com/atelier/pytorch_cuda:pytorch_1.12.1-cuda_10.2-py_3.9.11-ubuntu_20.04-x86_64-20260425154604-d9a1f9f. From cuda_10.2, it can be seen that CUDA is installed, so the type is GPU.

    Specifications

    Select the target specifications based on the image. If the type is NPU, the specification's model must match the image model.

    Get the specification from the image address. For example, if the address is swr.cn-north-4.myhuaweicloud.com/atelier/pytorch_ascend:pytorch_2.7.1-cann_8.5.2-py_3.12-euler_2.10.11-aarch64-snt9b-20260417112518-aabfd52, the image specification is snt9b.

    Image Description

    Provide detailed information for the uploaded custom image. Only .md files with a size no more than 32 KB can be uploaded. It can contain a maximum of 30,000 characters.

    Click Add to upload a .md file. You can modify an uploaded .md file or write one directly. Example:

    # Primary functions
    ...
    # Version differences
    ...
    # Usage instructions
    ...

    The image details will be displayed in the Overview tab.

    You can view the registered image in the Registered Images tab.

Using a Registered Image

After an image is registered, you can use it for model training, real-time inference, and notebook instances.

  • Model training:
    • New console: When creating a custom training job, select a custom image in the Training Configuration section and click . In the Custom Images tab of the Image panel, select a registered image as required. For details, see Creating a Custom Training Job.
      Figure 3 Training Configuration - Custom Images

    • Old console: When creating a training job, set Boot Mode to Custom image in the Environment Settings area and select the registered image. For details, see Creating a Production Training Job.
  • Real-time inference: When deploying a real-time service, select Custom Images in the Unit Settings section and click . In the Registered Images tab of the Image panel, select a registered image as required. For details, see Inference Deployment.
    Figure 4 Unit Settings - Custom Images

  • Notebook: When creating a notebook instance, select Custom Images in the Environment section and click . In the Registered Images tab of the Image panel, select a registered image as required. For details, see Creating a Notebook Instance.
    Figure 5 Environment - Custom Images

Adding an Image Version

In the Registered Images tab, use either of the following methods to add an image version:
  • Method 1:
    1. In the Operation column of the target image, click Add Version.
    2. In the Add Image Version panel, click , select the target version, and click OK. Configure the version description, architecture, type, and specifications as required, and click OK.
    3. In the Registered Images tab, click the image name. In the Versions tab, view the new version and its information.
  • Method 2:
    1. Click the name of the target image. In the upper right corner of the page, click Add Version.
    2. In the Add Image Version panel, click , select the target version, and click OK. Configure the version description, architecture, type, and specifications as required, and click OK.
    3. In the Versions tab, view the new version and its information.

Editing a Registered Image

  1. In the Registered Images tab, use either of the following methods to edit a registered image:
    For details about the parameters, see Table 1.
    • Method 1:
      1. In the Operation column of the target image, click Edit.
      2. On the Edit Image panel, configure the image description and tags as required, and click OK.
    • Method 2:
      1. Click the name of the target image. In the upper right corner of the page, click Edit.
      2. On the Edit Image panel, configure the image description and tags as required, and click OK.
    Figure 6 Editing an image
  2. View the edited image description and tags.
    • Viewing the edited image details: In the Registered Images tab, click the name of the target image. You can then view the image details in the Overview tab.
    • Viewing the edited tags: in the Registered Images tab, locate the Tags column and hover your mouse over to view the updated tags.

Creating a Notebook Instance Using a Registered Image

  1. In the Registered Images tab, click the target image name and choose Versions.
  2. In the version list, locate the target image version and click Create Notebook in the Operation column.
  3. On the Create Notebook page, configure the required information, select the agreement, and click Next.

    For details about the parameters, see Creating a Notebook Instance.

Adding Overview Information

As shown in the following figure, if no image description were configured during registration, the Overview tab will display No overview information available. You can follow the steps below or edit a registered image (Editing a Registered Image) to add overview information.

Figure 7 Overview
  1. In the Registered Images tab, click the target image name. In the Overview tab, click to enter the image description.
  2. In the Image Description area, click Add to upload the .md file.

    For details about the .md file and examples, see Table 1.

Synchronizing an Image

Synchronize the image status from SWR.

  1. In the Registered Images tab, click the name of the target image.
  2. In the Operation column of the version list, click Sync.

    If the message Synchronization status command set successfully is displayed, the image synchronization is successful.

    Figure 8 Synchronizing an image

Deleting an Image Version

Deleted versions cannot be restored. After an image is deleted, all instances using this image version will fail to be restarted. Exercise caution when performing this operation.

  1. In the Registered Images tab, click the name of the target image.
  2. In the Operation column of the version list, click Delete for the target version.
  3. In the Delete Version dialog box, enter DELETE and click OK.

    You can delete the associated SWR image version only when the image version is registered using the image saving function of a notebook instance.

    Figure 9 Deleting the associated SWR image version

Deleting a Registered Image

Deleted images cannot be restored. After an image is deleted, all instances using this image will fail to be restarted. Exercise caution when performing this operation. This will unbind the SWR image. The original image in SWR will remain intact.

Use either of the following methods to delete a registered image:

  • Method 1:
    1. In the Registered Images tab, click Delete in the Operation column.
    2. In the Delete Image dialog box, enter DELETE and click OK.
  • Method 2:
    1. In the Registered Images tab, click the name of the target image.
    2. In the upper right corner of the page, click Delete. In the Delete Image dialog box, enter DELETE and click OK.