Model Registry on the C3 Agentic AI Platform Overview
The C3 AI Model Registry enables sharing of models across environments and applications. Two of the primary use cases are:
Promoting one or more models for a given application across the DEV -> QA -> PROD environments for the deployment of the application. This allows you to decouple the model iteration lifecycle from the application iteration lifecycle. For example, you could retrain a PROD model in DEV, promote to QA for testing and validation, and promote to PROD to compete with or replace the current model, all without needing to do an application upgrade.
Sharing models between different applications within an organization. This can be particularly helpful for models that accomplish a specific task. For example, suppose there are two business units within an organization developing different applications. However, both applications require an NLP model to tokenize documents from clients. In this case, the NLP model and any improvements can be shared between applications.
The primary capabilities of the C3 AI Model Registry are:
- Model registration + versioning
- Model loading
- Search and discovery
- Model approval
C3 AI Model Registry Service
The C3 AI Model Registry Service is a microservice deployed in a C3 AI environment that can be accessed by authorized C3 AI applications.
See Set up Model Registry Service for details on initially setting up a model registry service on the C3 Agentic AI Platform.
C3 AI Model Registry
The C3 AI Model Registry is a client interface for interacting with the Model Registry Service, providing APIs for loading and registering supported model formats (such as loadMlPipe and loadMlModel) from your application to the C3 AI Model Registry.
The C3 AI Model Registry also provides APIs for searching for Model Registry entries with list and listVersions.
See Use the C3 AI Model Registry for more information.
Model Registry entry
An entry in the C3 AI Model Registry is created when a supported model format is registered. For example, ModelRegistry.registerMlPipe(mlPipe=pipeline, ...) creates a ModelRegistry.Entry. Model Registry entries are versioned, so a ModelRegistry.Entry is uniquely identified by the uri and ver fields.
C3 AI Model Approval Service
The C3 AI Model Approval Service is a microservice deployed in a C3 AI environment that can be accessed by authorized C3 AI applications.
See Set up the C3 AI Model Approval Service for details on initially setting up a model approval service on the C3 Agentic AI Platform.
C3 AI Model Approval
The C3 AI Model Approval is the client interface for interacting with the C3 AI Model Approval Service, providing APIs for changing the approval status for entries in the C3 AI Model Registry.
See Approve Models in the C3 AI Model Registry for more information.
C3 AI Model Approval record
A Record is created in the C3 AI Model Approval Service when you update the status of a C3 AI Model Registry entry. For example, ModelApproval.update(modelUri = entry.uri, modelVersion = entry.ver, scope = "Approval Scope", status = "APPROVED") creates a ModelApprovalService.Record.