How to set default cuda device in libtorch?

There is a function in pytorch to set the default cuda device. But I cannot find something similar in libtorch. Can someone help me out here?

I’m unsure if you can set the default device globally, but you could use the CUDAGuard() inside your method instead:

CUDAGuard device_guard(device);

which makes sure to use the specified device id.
(Thanks @tom for clarification :slight_smile: )

Thanks for the tip :slight_smile: