How to Track Model Lineage and Artifacts in Vertex AI?
You have a custom job that runs on Vertex AI on a weekly basis. The job is implemented using a proprietary ML workflow that produces the datasets, models, and custom artifacts, and sends them to a Cloud Storage bucket. Many different versions of the datasets and models were created. Due to compliance requirements, your company needs to track which model was used for making a particular prediction, and needs access to the artifacts for each model. How should you configure your workflows to meet these requirements?
Community Votes
100% of anonymous learners picked answer A. Votes are pick records left by other test-takers — they are not the verified answer.
Community Insight
This question tests the distinction between high-level model management and low-level metadata tracking, with the common trap being the assumption that Model Registry or autologging inherently satisfies strict compliance lineage requirements.
The Vertex AI Metadata API enables granular tracking of ML workflow lineage by programmatically linking contexts, executions, artifacts, and events. Community consensus strongly endorses Option A as the definitive solution for compliance-driven artifact auditing in custom Vertex AI jobs.
Option D is frequently selected incorrectly because while Vertex AI Model Registry excels at versioning and deploying models, it does not natively capture the fine-grained execution events and custom artifact relationships needed for comprehensive compliance auditing.
Community Discussion (3 comments)
Comments & Corrections
No comments yet — spotted an error or have a note? Share it below.
Expert Analysis
Why the Answer Is Correct
The Vertex AI Metadata API provides low-level primitives such as contexts, executions, artifacts, and events specifically engineered for tracking machine learning workflow lineage. By programmatically instantiating these entities within your custom job, you can explicitly map datasets, trained models, and proprietary artifacts to specific inference requests. This programmatic approach guarantees an auditable trail that directly satisfies rigorous compliance mandates requiring precise model-to-prediction attribution.Why the Other Options Are Wrong
Option B relies on autologging, which automates metric capture but often fails to record custom or third-party artifacts outside its predefined schema. Option C introduces TensorFlow Extended, a framework-specific orchestration layer that adds unnecessary complexity and incompatibility with non-TensorFlow custom jobs. Option D utilizes Model Registry, which focuses primarily on model versioning, comparison, and deployment pipelines rather than capturing granular execution history and cross-entity relationships.Community Comment Notes
Contributors consistently identify the compliance and tracking keywords as direct triggers for selecting the Metadata API [1][2]. Multiple users emphasize that autologging lacks the structural flexibility to ingest custom workflow outputs, explaining why Option A reliably outperforms Option B [3]. The unanimous voting pattern reflects strong alignment with Google Cloud's documented best practices for ML lineage management.Official Reference
Exam Strategy
When exam prompts emphasize compliance, auditability, or custom artifact lineage, immediately prioritize the Metadata API over managed registries or automated logging features. Familiarize yourself with the four core metadata entities—contexts, executions, artifacts, and events—and practice writing pseudo-code to link them programmatically.
Related Analysis
Practice All PMLE Questions
Access 65 questions with complete answers and detailed explanations.
View Full PMLE Practice Test →