Which Vertex AI services track and compare model versions across projects?

Why the Answer Is Correct

Option D is correct because it includes the three services that work together for end-to-end ML workflow tracking and comparison. Vertex AI Pipelines orchestrates the training steps, Vertex AI Experiments records parameters and metrics for each run, and Vertex AI ML Metadata stores detailed lineage information—allowing you to find, compare, and audit model versions across projects. This combination directly addresses the need to track and compare model performance.

The suggested answer aligns with the official Google Cloud MLOps pattern. Pipelines provide reproducible workflows, Experiments offer a simple way to log and compare runs, and Metadata captures artifacts and lineage across the entire lifecycle. Together, they form the backbone for 'find, track, and compare' requirements, especially complex multi-project environments.

Why the Other Options Are Wrong

Option A is wrong because Dataplex is a data management and governance service, not an ML experiment tracker. Vertex AI Feature Store manages feature definitions and serving, but it does not log or compare model versions. TensorBoard is a visualization tool, but without Experiments and Metadata, it cannot provide the structured comparison of versions required by the question.

Option B is also incorrect because it includes Feature Store instead of ML Metadata. Feature Store is for feature reuse and consistency, not for storing model run lineage. While Pipelines and Experiments are essential, the absence of ML Metadata means you lose the critical link between runs, artifacts, and parameters, which is necessary for detailed 'find and track' capabilities.

Option C is the most competitive, but it replaces Pipelines with Dataplex. Dataplex is useful for centralizing data across projects, but it does not orchestrate ML workflows. The question's focus on model version tracking and comparison demands Pipelines for workflow automation and reproducibility. Without Pipelines, the combination remains a data-focused solution rather than an ML workflow solution.

Community Comment Notes

Community members with the most upvoted comments favored D. One commenter (likes=5) explained that neither Dataplex nor Feature Store is directly tied to model versioning and comparison, and that TensorBoard alone is insufficient. Another commenter (likes=2) noted, 'Vertex AI Metadata is for spans across multiple Google Cloud projects data used by the model,' reinforcing that Metadata handles the cross-project lineage requirement.

Some test-takers argued for C, pointing to Dataplex's ability to centralize different projects and ML Metadata's tracking. However, the higher-vote comments correctly identified that Pipelines is a core orchestration component in the ML workflow. The debate hinges on whether 'ML workflow' implies Pipelines—most experienced exam takers agree it does. One commenter (likes=1) incorrectly dismissed Metadata as unnecessary, but Metadata is essential for artifact and parameter lineage, especially when tracking across projects.