After writing a Kubeflow pipeline in Python, which artifact must you generate before Vertex AI can run it?
Show answer and explanation
Answer: D
Kubeflow Pipelines definitions are compiled into a YAML or JSON pipeline specification that describes the component graph, inputs, and container images. Vertex AI reads that compiled package to create runs. The Python source must be compiled first, a saved model is a trained artifact rather than a pipeline definition, and a manifest without a graph cannot describe execution order.
Why the other options are wrong
- A. A saved Keras model is the output of training, not the definition of a pipeline. Vertex AI needs the workflow specification to orchestrate steps.
- B. A container image manifest identifies images but does not describe step order, inputs or outputs. Pipeline execution requires a graph definition.
- C. Raw Python source cannot be executed directly by the pipeline service. It must be compiled into the declarative specification first.