New devices are installed with a small model that runs on the CPU. To serve a larger model, or to use the GPU or NPU, you can deploy a model from the Fleet Manager.
Deploy
Section titled “Deploy”- In the Fleet Manager, open Fleet → Models and choose a model from the catalog.
- Choose how to place it:
- Whole model: every chosen device serves its own full copy. Use this when the model fits in one device’s memory.
- Sharded: the model is split into stages across an ordered list of devices that work together as one pipeline. Use this when the model is too large for one device.
- Choose the devices and start the deploy.
Each device downloads its part and switches over. Fleet → Activity shows the progress of each device and whether the deploy succeeded.
To deploy to a single device, expand its row in Fleet → Devices.
Free up disk space
Section titled “Free up disk space”Models a device no longer serves stay on its disk. To remove them, expand the device’s row in Fleet → Devices, select the idle models and delete them. A device never deletes the model it’s serving.
Current limitations
Section titled “Current limitations”When the Fleet Manager is connected to your console account, it can’t yet replace the model a device is already serving, edit or move an existing deployment, or place models automatically.
Contact your Cascadia representative if you need one of these.

