Libtorch error when loading model

I have got an error when using libtorch load model.
The error message is this:
what(): tensor_proto.has_device() && !tensor_proto.device().empty() ASSERT FAILED at /pytorch/torch/csrc/jit/import.cpp:124, please report a bug to PyTorch. (loadTensor at /pytorch/torch/csrc/jit/import.cpp:124)
frame #0: std::function<std::string ()>::operator()() const + 0x11 (0x7f8a585a6fe1 in /home/dell/software/libtorch/lib/
frame #1: c10::Error::Error(c10::SourceLocation, std::string const&) + 0x2a (0x7f8a585a6dfa in /home/dell/software/libtorch/lib/
frame #2: <unknown function> + 0x667539 (0x7f8a6148f539 in /home/dell/software/libtorch/lib/
frame #3: torch::jit::load(std::istream&, c10::optional<c10::Device>) + 0xe5 (0x7f8a614912f5 in /home/dell/software/libtorch/lib/
frame #4: torch::jit::load(std::string const&, c10::optional<c10::Device>) + 0x6c (0x7f8a6149141c in /home/dell/software/libtorch/lib/
frame #5: main + 0xb9 (0x4027ac in ./test)
frame #6: __libc_start_main + 0xf5 (0x7f8a57cc2f45 in /lib/x86_64-linux-gnu/
frame #7: ./test() [0x402589]

Aborted (core dumped)

My os is ubuntu 14.04 and the newest libtorch, I have just try libtorch on cu9.0 and cpu version.