Part1 2020 : 02_Production - found at least two devices, cuda:0 and cpu

Part 1 (2020)
9

From what I can tell, this is the problem:
In data/load.py iter

# fix issue 2899. If the process start method isn't fork, the data will be copied to cuda in learner 
one_batch.
if self.device is not None and multiprocessing.get_start_method().lower() == "fork":
    b = to_device(b, self.device)

the start_method is spawn for Windows 10
and one_batch does not get called before the error happens, I imagine because it is running transforms on the object and the transforms expect the data to already be on the GPU.

Changing it to this makes it work:

    if self.device is not None:
        b = to_device(b, self.device)