解决多卡并行时tensor在不同的cuda上

解决RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:
https://blog.csdn.net/aminghhhh/article/details/130839821