[Python] Roboflow device management. In which GPU is the model loaded?

I am experimenting with the SDK of Roboflow, I haven't found documentation on device management. I have three questions:

1. Is the model loaded to GPU by default?
2. What happens when multiple GPUs are available?
3. And how do I define in which GPU I want the model allocated?

EDIT:

Roboflow does not expose the device directly as Pytorch would:

rf = Roboflow(api_key="YOUR API KEY")
project = rf.workspace().project("license-plate-recognition-rxg4e")
model = project.version("4").model
model = model.to("cuda:4")
> AttributeError: 'ObjectDetectionModel' object has no attribute 'to'

Continue reading...