C3 AI Documentation Home

C3 AI Studio Model Ops

The Model Ops section of C3 AI Studio is where you can review and manage your machine learning models and model routes.

In the C3 Agentic AI Platform, a machine learning project organizes related assets, such as models and features, for a business use case—for example predicting user churn. The project's MlModel.Router determines which model route handles prediction requests based on the subject. A MlModel.Route specifies exactly which model serves which subjects. Model routes and routers are part of C3 AI's Model Deployment Framework (MDF). To learn more about the MDF, see Model Deployment Overview.

On the Model Ops page, you can inspect model routes and change the status of deployed models. You can also make your models public or load public models created by other users into your application. The following diagram shows a navigation overview:

flowchart TD start(*Model Ops*) appModels[Application Models] mRegistry[Model Registry] mDetails[Model Details] start --> Deployments start --> Models Deployments --> |Model Health|mDetails mDetails --> Training mDetails --> Monitoring Models --> appModels Models --> mRegistry

The Model Ops page has two subpages: Deployments and Models. Each subpage has a side-panel for filtering lists by several values, including:

  • Project: The name of a machine learning application. See MlProject for more information.
  • Subject: A filter string, for example 'id==Some_Value', that identifies a Subject Type to use with a model route.
  • Model ID: A model's unique identifier
  • Model training date range: The earliest or latest training date of a model.
  • Model Route Status: The status of a model route. For the full range of possible values, see MlModel.Route.Status.Label.

Deployments

In the C3 Agentic AI Platform, a deployed model is available for making data predictions or inferences. You can access deployed models through model routes. The Deployments tab shows a table of such model routes in an application. Each route displays the following information:

  • Model: A unique identifier for the model.
  • Subject Filter: A filter condition that identifies a subject Type and the asset that the model operates on.
  • Created: The date the model route was created.
  • Project: The parent machine learning project for the ML route.
  • Model Health: A value representing the model's performance against acquired ground truth (if this model includes model health reporting). Select the value to view information about the model's training, or configure model performance monitoring. For more information on model monitoring, see MlModel.Monitor.
  • Output: The output handler for this model route. An output handler processes the output of a model to a store (for example, a database) or forwards it to an external system such as a Kafka Queue. See MlModel.OutputHandler to learn about output handlers.
  • Status: The deployment status of this model route. Possible values include:
    • CHAMPION
    • CHALLENGER
    • CANDIDATE
    • RETIRED

Use the ellipsis button on each listed route to set its model status as CHALLENGER, CHAMPION, or RETIRED. See MlModel.Route.Status.Label for a description of route status values.

Models

Models contains information about your application models. The page is split into two subpages, Application Models and Model Registry.

Application Models

Application Models lists machine learning models in your application. Each list item shows the following information:

  • Id: The model's unique identifier. This value links to a page with more information about the model.
  • Training Status: Trained models show the date of their training and identify the user who triggered the training. Untrained models show the text "Not Trained".
  • Deployments: Shows the number of routes and subjects linked to this model. A single model artifact can be associated with more than one route.
  • Project: The name of the machine learning project that contains this model.
  • Feature sets: The feature sets the model operates on. A feature set is a group of measurable properties from an asset that can be used as a machine learning input for prediction or inference. To learn more about features and feature sets, visit C3 AI Feature Store Overview.

Hover over each Application Model list item to display the following controls:

  • Deploy: Make the model available for prediction and inference through a model route.
  • Register: Make the model available for use outside your application or organization. For more information, see the Model Registry section of this document.

Model Registry

Any user can register a model for use in any organization or application. You can find such public models in the Model Registry and load them into your application.

The Model Registry list shows the following information:

  • Model path: A partial model resource locator within the C3 Agentic AI Platform. This value links to a page with more information about the model.
  • Description: A text description of the model, if the model creator provided one.
  • Versions: How many versions the model has.
  • Updated: When the model or its information was last changed.
  • Tags: Any content tags supplied by the model creator.

Hover over a list item to access the Load Latest Version control. Select this control to open a modal where you can specify how your application loads the model. For more information about the Model Registry, see Overview of Model Registry.

See also

Was this page helpful?