Expected object of scalar type Long but got scalar type Float for argument #2 'target'

2

Change:

labels = labels.to(device)

to

labels = labels.to(device=device, dtype=torch.int64)

CrossEntropyLoss expects targets to be class indices, and hence Long Tensors, but you gave it a Float Tensor.

9 Likes