A base image is the foundation for a container image: Docker starts with it, then adds application-specific layers and configuration. In a Dockerfile, the FROM instruction selects the base image. section 3.2.2
Choosing a suitable base image reduces manual dependency installation and version conflicts. In Kubeflow, the same base image can support notebooks, pipeline components, and deployment containers, keeping development close to production. section 7.1 section 8.1.4
Python application
A Dockerfile can begin with FROM python:3.10-slim-buster, then define /app as the working directory, install requirements, copy application files, and configure an entrypoint. section 3.2.2
Listing 8.12 Defining the KFP component for YOLO model validation
@dsl.component( base_image="ultralytics/ultralytics:8.0.194-cpu", packages_to_install=["minio", "tqdm"] ) def validate_model( data_yaml: Input[Artifact], #1 model: Input[Model], #2 validation_dataset: Input[Dataset], #3 metrics: Output[Metrics] #4 ): # ...
Inspect a KFP validation component that uses the Ultralytics base image and adds only the remaining packages through packages_to_install. section 8.1.4
