Access LibTorch CUDA Tensor

Is it possible to get the underlying CUDA Tensor when using LibTorch? How?