How does consumption work?

You can track your quota plan by navigating the subscription view.

  • Compute time: GPU hours for computing. Time spent training a model.

  • Predictions limit:

    • Only individual predicted results served by the deployed model is counted
    • Predictions count on CPU hours, not the GPU hours
    • Inspections in the evaluation view are not counted
    • Note that testing from a public Web App does not count against the prediction limit
  • Storage: Disk space for storage

  • Number of member: Number of seats in the organization

  • Active deployment: Number of active deployments

  • Concurrent experiments: Number of parallel experiments you can run. Note that, you can have an unlimited amount of experiments in each plan. The restriction only applies to how many of them you can train at the same time.