trainer.on_gpu == True for gpus = 0 or gpus=[]

[mpariente]

Describe the bug
trainer.on_gpu == True for gpus = 0 or gpus=[] because of this line.
When resuming from a checkpoint, the model is transfered to gpu at this line, even if gpus=0

Expected behavior
gpus=0 should make training happen on CPU.

I can make a PR for that, the fix is pretty simple, I think we can just replace
self.on_gpu = gpus is not None and torch.cuda.is_available()
by
self.on_gpu = True if (gpus and torch.cuda.is_available()) else False

Am I missing something?

[williamFalcon]

Thanks for bringing this up. To use no gpus just set

gpus=None

Or don't pass any arguments to trainer about gpu

If you think that gpus=0 or gpus=[] should also run on no GPUs feel free to submit a PR