Kubeflow Pipelines currently lacks native support for packaging and pushing trained models to OCI-compliant registries using the ModelKit format

Create a reusable Kubeflow Pipeline component that packages model artifacts as ModelKits and pushes them to any OCI registry.

Features

1. Metadata Extraction

• Pull training parameters from Kubeflow ML Metadata
• Include pipeline run ID, experiment name as Metadata
• Map Kubeflow metrics to ModelKit metadata fields

2. Authentication

• Support Kubernetes secret mounting for registry credentials
• Handle IRSA/Workload Identity for cloud deployments