Model registry

As explained in LLMs in Production

A model registry is a system for storing, versioning, annotating, and controlling access to trained machine-learning models.

A model registry is a shared home for trained models: it stores model artifacts, links them to metadata, and keeps replacements from becoming an operational guessing game. A basic object store or shared filesystem can work, but production use benefits from versioning, controlled promotion, and access rules. section 3.4.3

Choose a registry by checking the operational details that matter after training:

• Link models to experiment-tracker metadata, especially when models come from open-source projects or external purchases. section 3.4.3
• Version models so teams can replace obsolete models and run A/B tests or shadow tests. section 3.4.3
• Restrict access and limit promotion or demotion authority to teams that understand the models and their training rationale. section 3.4.3
• Check storage limits before adopting a registry for LLMs; a 10 GB cap may be unusable, while on-premise Ceph requires substantial capacity. section 3.4.3

Figure 3.7 A high level view of an MLOps infrastructure with LLMs in mind. This diagram attempts to cover the full picture and the complexity of the many tools involved to make ML models work in production.

Places the model registry among experiment trackers, feature stores, data systems, deployment services, and other MLOps infrastructure. section 3.4

How LLMs in Production covers it

How deep it goes

Explains storage, metadata links, versioning, permissions, and LLM size constraints, but stops short of implementation details.

Examples

Uses object stores, shared filesystems, open-source models, purchased models, 10 GB limits, and Ceph deployments.

What it assumes

Assumes familiarity with trained models and the surrounding MLOps infrastructure, while treating registry fundamentals as straightforward.

What you can do after

Operate/configure a registry by selecting storage, metadata integration, versioning, permissions, and capacity appropriate for deployed models.

How it connects

Links model registries to experiment tracking, object storage, A/B testing, shadow testing, and access control.

liveBookmanning.com

Read LLMs in Production on liveBook Get the book at manning.com

Also explained in

• 

  liveBookmanning.com
  Machine Learning Platform EngineeringA model registry is a centralized system for storing, versioning, and promoting machine learning models across deployment stages.

Compare how 2 books explain model registry →

Related concepts

• MLflowpart of · MLflow is an open-source platform for managing the machine-learning lifecycle, including experiment tracking, model and artifact management, and model deployment.
• metadatauses · Metadata is information that describes, identifies, or provides context about other data, resources, or artifacts.
• experiment trackinguses · Experiment tracking is the systematic recording and organization of experiment configurations, metrics, artifacts, and results to support inspection, comparison, or reproducibility.
• model managementimplements · Model management is the operational practice of organizing, configuring, tracking, and controlling machine-learning models and their artifacts across their lifecycle.
• contractuses · An API contract is a specification of the operations, data, and behavioral expectations that govern interactions between software components.

This explainer was generated by Manning's knowledge graph from the text of LLMs in Production. The section links point to the parts of the book it draws on.