Applies to SUSE AI Factory with NVIDIA

5 How do I manage AI blueprints? #

SUSE AI Factory with NVIDIA helps you manage custom and preconfigured AI blueprints on your Kubernetes cluster.

Important

You must first configure all the necessary secrets before installing AI applications and blueprints. Refer to Post-deployment steps for detailed procedures.

Screenshot showing a list of AI blueprints

Figure 5.1: List of AI blueprints #

5.1 How do I create custom blueprints? #

From the main left panel, select SUSE AI Factory › Blueprints › Create.
On the Basic Info screen, enter the name, version, and optional description of the blueprint.
Figure 5.2: Basic information about a blueprint #
On the Select Apps screen, search for an application to add to the template, select its name, and specify the version to install. Repeat this step for all applications you want to include in the blueprint.
Figure 5.3: Selecting applications of a blueprint #
On the Configuration screen, review and optionally update the Helm deployment chart of the blueprint.
Figure 5.4: Custom applications configuration #
On the Review & Create screen, verify that the entered information is correct and confirm with Create.
Figure 5.5: Review blueprint configuration #

5.2 How do I modify blueprints? #

From the main left panel, select SUSE AI Factory › Blueprints.
From the list of existing templates, select the blueprint to modify, open its three-dot action menu, and select Edit.
Figure 5.6: Modify blueprints #
On the Basic Info screen, update the version number and description.
On the Select Apps screen, update the list of applications in the blueprint by adding new applications or removing existing ones. Confirm with Next.
On the Configuration screen, review and optionally update the Helm charts of the blueprint applications.
On the Review & Create screen, verify that the entered information is correct and confirm with Save as New Version.

5.3 How do I install NVIDIA blueprints? #

From the main left panel, select SUSE AI Factory › Blueprints.
From the list of existing blueprints, select the NVIDIA blueprint to install, specify the deployment version, and select the Install button.
On the Basic Info screen, customize the default name and target namespace for your workload.
On the Target screen, specify the deployment type and the target cluster for your workload. You can also review the estimated hardware requirements for the workload.
On the Review screen, verify that the configuration is correct:
- Default storage class. Ensure that a default storage class is set on the target cluster. Applications such as Milvus and Qdrant require persistent storage. If no default storage class is available, pods remain in the Pending state.
- Ingress settings. If your blueprint includes applications that use Ingress, ensure the host field is set correctly for each application. Leaving the host field empty causes the Ingress service to be created without a valid host name.
- Required application fields. Ensure all required application fields are filled in. Missing values can prevent workloads from starting correctly.
  Tip
  If the workload stays in the Pending state for an extended period after installation, a configuration value is missing or incorrect. To identify the issue, check the pod status directly:
  > kubectl get pods -n <WORKLOAD_NAMESPACE> > kubectl describe pod <POD_NAME> -n <WORKLOAD_NAMESPACE>
  The output shows the exact reason why the pod is stuck, such as a missing PVC, an unresolvable image or a failed volume mount.
Start the deployment by selecting Install.
Navigate to SUSE AI Factory › Workloads to monitor the startup progress and verify deployment status.

5.4 How do I clone a blueprint? #

From the main left panel, select SUSE AI Factory › Blueprints.
From the list of existing blueprints, select the template you want to clone, specify the source version, open its three-dot action menu, and select Copy.
Figure 5.7: Clone blueprints #
On the Basic Info screen, you can modify the suggested name, version and description.
On the Select Apps screen, update the list of applications in the blueprint by adding new applications or removing existing ones. Confirm with Next.
On the Configuration screen, review and optionally update the Helm deployment chart of the blueprint.
On the Review & Create screen, verify that the entered information is correct and confirm with Create.

5.5 How do I deprecate blueprints? #

From the main left panel, select SUSE AI Factory › Blueprints.
From the list of existing blueprints, select the template to deprecate.
Specify the version to deprecate, open its three-dot action menu, and select Deprecate.
Figure 5.8: Deprecate blueprints #
Confirm with Deprecate.

5.6 How do I restore deprecated blueprints? #

From the main left panel, select SUSE AI Factory › Blueprints.
A deprecated blueprint is hidden from the list of blueprints. To make it visible again, check the Show deprecated box.
Select the blueprint you want to restore and specify the deprecated version.
Open the three-dot action menu for the blueprint, and select Undeprecate.
Confirm with Undeprecate.

5.7 How do I delete blueprints? #

Note: Bundled blueprints cannot be deleted

Bundled blueprints shipped with SUSE AI Factory with NVIDIA are protected and cannot be deleted. The Delete action is not shown for them, and any deletion request is rejected. To customize a bundled blueprint, clone it as described in Section 5.4, “How do I clone a blueprint?” and edit the copy.

From the main left panel, select SUSE AI Factory › Blueprints.
From the list of existing templates, select the blueprint to delete and specify the target version.
Open the three-dot action menu for the blueprint, select Delete, and confirm the deletion.
Figure 5.9: Delete blueprints #

5.8 How do I manage environment-specific configurations with blueprints? #

Consider the following use-case: You want to deploy a single LiteLLM blueprint to both a development cluster and a production cluster, but with different configurations. Because of resource constraints on development nodes, you must configure different LLM models, Ingress settings, or GPU allocations. You can achieve this by managing separate blueprints named blueprint-litellm-dev and blueprint-litellm-prod.

To configure and manage environment-specific deployments, perform the following steps:

Use a prebuilt blueprint or create a custom blueprint as described in Section 5.1, “How do I create custom blueprints?”. Because this template serves the production cluster, name the blueprint blueprint-litellm-prod.
Create a new version of the blueprint with the desired configurations as described in Section 5.2, “How do I modify blueprints?”.
Clone the blueprint as described in Section 5.4, “How do I clone a blueprint?” and name it blueprint-litellm-dev.
Deploy the blueprints as described in Section 5.3, “How do I install NVIDIA blueprints?”. Specify the correct version for each deployment.
1. Deploy blueprint-litellm-prod to the production cluster.
2. Deploy blueprint-litellm-dev to the development cluster.

Tip

To apply a new blueprint version during future maintenance window upgrades, select SUSE AI Factory › Workloads and select the corresponding Upgrade button.