fix usage of gpu

[alishahali1382]

when calling model.to(device), a new instance of model is created and loaded on the device and returns the new model. But by calling this and using the previous model, all operations are performed on the previous device (here it is cpu).
therefore GPU is not used in current code.

[amirhossein-razlighi]

Please look at the example below:

model = nn.Linear(1, 1)
print(model.weight.device)
> cpu
model.cuda()
print(model.weight.device)
> cuda:0

x = torch.randn(1)
print(x.device)
> cpu
x.cuda() # ERROR!!!
print(x.device)
> cpu

So, as you can see, despite tensors, the nn.Module instances' weights are moved to the cuda device inplace! So, no bugs here as self.model is an instance of nn.Module and not a tensor.
Please refer to this discussion on PyTorch's forum for further info.