RuntimeError: Input type (CUDAFloatTensor) and weight type (CPUFloatTensor) should be the same

vision
1

I am training a GAN model, i want to train it 200 epochs, but when it comes to the second epoch, an error comes:
RuntimeError: Input type (CUDAFloatTensor) and weight type (CPUFloatTensor) should be the same

I have put the data and model into gpu yet, and the first epoch training is fine, where did this error come?

2

Could you please provide more context behind what your model looks like? You’re right that it’s strange that after one good epoch of training the model crashes with that assertion.

3

My model is based on cyclegan model,its about 600 lines, and part of it is here:https://gist.github.com/yingdiLee/de3c8c13a06109249aa78838e584a4e7. If it is not clear enough, I can provide all of the source code, thank you for your help !

4

I have the same problem …

RuntimeError: Input type (CUDAFloatTensor) and weight type (CPUFloatTensor) should be the same

1 Like
5

it seems like after first epoch of training the parameters of model changes from torch.cuda.FloatTensor to torch.FloatTensor, I do not know if it is from my definition of model or somewhere else, and it makes me crazy…

6

Ah. got my problem solve. missing lines

if use_cuda:
    net.cuda()
    net = torch.nn.DataParallel(net, device_ids=range(torch.cuda.device_count()))
    cudnn.benchmark = True
2 Likes