torch.max
torch.max(input, dim, keepdim=False, out=None) -> (Tensor, LongTensor)
It just ignores the unneeded Tensor above.