-
Notifications
You must be signed in to change notification settings - Fork 395
Open
Description
It seems to me that there are assumptions of using CUDA? I'm very new to python, but would it make sense to do something like this
if torch.cuda.is_available(): torch.cuda.set_device(local_rank) elif torch.mps.is_available(): torch.mps.set_device(local_rank) else: torch.cpu.set_device(local_rank)
in places where currently
torch.cuda.set_device(local_rank)
is used, or have I misunderstood something?
Metadata
Metadata
Assignees
Labels
No labels