Model Foundry

Model Foundry lets you use foundation models to prelabel and enrich your unstructured data. Model Foundry creates model runs that predict annotations based on your input.

You can apply the predictions to your data, evaluate model run performance, and compare model runs using different models, prompts, and settings.

🚧

This article describes features currently in preview. There might be differences between the app and the docs.

Here, we introduce Model Foundry and show how to use it.

Before using Model Foundry, you should:

• Connect your cloud data to Labelbox via IAM delegated access.
• Import your data to Labelbox.
• Create a project in Annotate.
• Configure your labeling workflow.
• Review pricing details.

Model Foundry workflow

Here's how to use Model Foundry to create and start a model run:

1. Use Catalog to select data.
2. From the Model Gallery, choose a model.
3. Configure the model run settings and parameters.
4. Generate preview predictions to verify your choices.
5. Submit the model run and wait for it to finish.

For details, see each step in turn.

Common tasks

Here are some other things you can do with model runs:

• Use the Notification Center to view model run progress.
• Use Catalog to view predictions from a model run.
• Use Model to view model run details.
• Use predictions to prelabel data
• Export predictions for other uses

Additional Resources

• Model Foundry billing details
• Manage Model Foundry access to specific models