Pytorch: the value of loss function is always 0

See more:

PyTorch

I have little experience in training neural networks. When I try to train the network function, I find the loss function is always 0. Then I make the following attempt:

>>> from torch import tensor                                                                                            >>> import torch.nn as nn                                                                                               >>> t1=tensor([[0.23],[0.33]])                                                                                          >>> t2=tensor([[1.0],[1.0]])                                                                                      >>> criterion = nn.CrossEntropyLoss()                                                                                   >>> criterion(t1,t2)                                                                                                    tensor(-0.)                                                                                                             >>> criterion(t1,t2).item()                                                                                             -0.0                       

>>> criterion = nn.CrossEntropyLoss().cuda()                                                                            >>> t1=t1.cuda()                                                                                                        >>> t2=t2.cuda()                                                                                                        >>> criterion(t1,t2)                                                                                                    tensor(-0., device='cuda:0')                                                                                            >>> criterion(t1,t2).item()                                                                                             -0.0 

Could you tell me why the output of criterion is always 0?

What I have tried:

t1 t2 is what the output of the NN and my label like.