Skip to main content
Image management covers the container images used by development environments, training jobs, and inference services. An image packages the operating system, framework, Python and CUDA versions, libraries, and startup tools a program needs, defining the runtime environment available to a workload. Platform images fall into two categories: preset images and custom images. The platform maintains preset images, which you can select and use directly. You can build and register custom images from existing images, Dockerfiles, third-party images, or images in existing registries to meet your requirements for dependencies, framework versions, or business code.

​
Use Cases

  • Choose a suitable runtime environment for a development environment, training job, or inference service.
  • Install your own dependencies on a platform image to build a custom image that your team can reuse.
  • Import or synchronize external images to the platform image registry.
  • Review image usage in training jobs, development environments, and inference services.

​
Image Types

​
Images and Clusters

Images are organized by region and cluster. The same image may have a separate replica and a different image URL in each cluster. Before creating a development environment, training job, or inference service, confirm that an available image exists in the target region and cluster. The platform provides each tenant with an image registry project for storing replicas of preset images, custom images, and imported images. Registry storage is limited, so check current usage before building or importing an image. You can distribute custom images to multiple clusters by configuring image registry sync policies. After synchronization completes, workloads in the target clusters can select and pull the images. Whether a custom image is visible usually depends on its owner, shared project groups, region, and target cluster. If you cannot find an image when creating a workload, check that the selected region, cluster, project group visibility scope, and filters match the image details.

​
Custom Images Page

Go to Images > Custom, then select the tab that matches your task:

​
Usage Process

  1. Select a preset image or create a custom image.
  2. To use the image across clusters, configure or verify its sync policy.
  3. Wait for the image build or synchronization to complete.
  4. Select the image when creating a development environment, training job, or inference service.
  5. Use usage statistics and registry events to identify images to remove or troubleshoot image pull issues.