DataTerminology.comNEW
Also known as: Model Catalog
A system for versioning, approving, and tracking machine learning models.
DESCRIPTION
A model registry centralizes model artifacts, versions, stages (e.g., Staging, Production), and lineage. It supports approval workflows, governance, and auditability for promoting models. Registries integrate with training pipelines, feature stores, and serving platforms. They provide metadata, metrics, and deployment history. Popular implementations include MLflow and cloud-native registries from major vendors.
ALTERNATIVE DEFINITIONS
MLflow
The MLflow Model Registry is a centralized model store, set of APIs, and UI to collaboratively manage the lifecycle of ML models.
Emphasizes collaboration and lifecycle management with APIs and UI.
AWS SageMaker
The SageMaker Model Registry helps you catalog, organize, and manage model versions, approvals, and deployments across your ML workflow.
Highlights cataloging, approvals, and deployment integration.
Google Cloud
Vertex AI Model Registry is a centralized repository to manage, version, and deploy models with integrated governance and lineage.
Emphasizes governance and lineage integration in Vertex AI.
RELATED TERMS
MLflow Model Versioning Model Governance Model Serving Experiment Tracking Model Lifecycle Model Cards Approval Workflow Metadata Management
COMPETING TERMS
KNOWLEDGE AREAS
MLOps AI Governance Data Management Metadata Management
TARGET AUDIENCE
MLOps Engineer Data Scientist AI Engineer Data Governance Manager Data Steward
EXTERNAL SOURCES
- Wikipedia - MLflow
- MLflow - Official MLflow Model Registry docs
- Databricks - Databricks model registry guide
- Google Cloud - Vertex AI model registry overview
TOP QUESTIONS
Why is a Model Registry essential in MLOps?
It enforces governance and traceability, enabling safe promotions and audits of model changes in production.
Category: Relevancy
What metadata should a Model Registry track?
Versions, stages, metrics, lineage, artifacts, owners, and approvals ensure full lifecycle visibility.
Category: Requirements
What are common mistakes with Model Registry usage?
Missing approval workflows, weak naming conventions, and unlinked datasets can break audit trails.
Category: Mistakes
What risks exist without a Model Registry?
Uncontrolled model changes, lost provenance, and compliance issues increase operational and legal risk.
Category: Risk
How is a Model Registry integrated with pipelines?
Training pipelines register models, attach metrics, and transition stages via APIs under governance policies.
Category: Management
What alternatives compete with a Model Registry?
Model Hubs focus on sharing; artifact registries store binaries but lack model-specific lifecycle features.
Category: Alternatives
What is a hot topic around Model Registry today?
Model cards, lineage graphs, and automated SLO-based approvals drive better governance.
Category: Topic
How does a Model Registry support rollback?
It preserves prior model versions and stages, enabling quick promotion of a stable model.
Category: Example
Where should teams start implementing a Model Registry?
Pilot with MLflow, define stage gates, capture lineage, and integrate approvals with CI/CD.
Category: Requirements