conversion to non-scalar type torch::jit::load("model.pt")

[caxelos]

🐛 CMake error: conversion from ‘torch::jit::script::Module’ to non-scalar type ‘std::shared_ptrtorch::jit::script::Module .

I am trying to use the torch::jit::load() function,in order to embed my pretrained model
to a c++ project.I am doing exactly what is shown here:https://pytorch.org/tutorials/advanced/cpp_export.html#a-minimal-c-application.I have searched,but I couldn't find a solution here.

#include <iostream>
#include <memory>
#include "libtorch/include/torch/script.h" // One-stop header.

int main(int argc, const char* argv[]) {
  if (argc != 2) {
    std::cerr << "usage: example-app <path-to-exported-script-module>\n";
    return -1;
  }

  // Deserialize the ScriptModule from a file using torch::jit::load().
  // Here is the error!!
  std::shared_ptr<torch::jit::script::Module> module = torch::jit::load("model.pt");

  assert(module != nullptr);
  std::cout << "ok\n";
}

The compilation is done with the command

sudo cmake -DCMAKE_PREFIX_PATH=../libtorch .. && make

Environment

• PyTorch version: 1.1.0

• Is debug build: No

• CUDA used to build PyTorch: 9.0.176

• OS: Ubuntu 16.04.5 LTS

• GCC version: (Ubuntu 5.4.0-6ubuntu1~16.04.11) 5.4.0 20160609

• CMake version: version 3.5.1

• Python version: 3.6

• Is CUDA available: No

• CUDA runtime version: 7.5.17

• GPU models and configuration: Could not collect

• Nvidia driver version: Could not collect

• cuDNN version: Could not collect

Versions of relevant libraries:

• [pip3] numpy==1.15.4
• [pip3] numpydoc==0.8.0
• [pip3] torch==1.1.0
• [pip3] torchsummary==1.5.1
• [pip3] torchvision==0.3.0
• [conda] blas 1.0 mkl
• [conda] mkl 2019.1 144
• [conda] mkl-service 1.1.2 py36he904b0f_5
• [conda] mkl_fft 1.0.10 py36ha843d7b_0
• [conda] mkl_random 1.0.2 py36hd81dba3_0
• [conda] torch 1.1.0
• [conda] torchsummary 1.5.1
• [conda] torchvision 0.3.0

[zdevito]

It looks like you are using a nightly build. We recently changed the output type of load. This should work:

 torch::jit::script::Module module = torch::jit::load("model.pt");

Tutorials/documentation are still for the 1.1 release. It will be updated before we release 1.2.

[caxelos]

@zdevito Thanks,it worked.

• But now there is a problem with the assert() function.It throws the following error :error: no match for ‘operator!=’ (operand types are ‘torch::jit::script::Module’ and ‘std::nullptr_t’)
• How can I fix that bug in the pointer comparison?

[zdevito]

That assert can just be removed. Module isn't a pointer anymore, it behaves like torch::Tensor, and load will never return an invalid module.

[hamidkhb]

@zdevito I did what you suggested but now I'm stuck here:

at::Tensor output = model[0].forward(inputs).toTensor();

I get Illegal instruction

void test(vector<module_type>& model){
    //pseudo input
    vector<torch::jit::IValue> inputs;
    inputs.push_back(torch::ones({1, 3, 224, 224}));
    Tensor output = model[0].forward(inputs).toTensor();
    cout << output << endl;
}

int main(int argc, const char *argv[]) {

    if (argc == 2){
        cout << argv[1] << endl;


    }
    else {
        cerr << "no path of model is given" << endl;
        return -1;
    }
    // test
    module_type module = torch::jit::load(argv[1]);
    vector<module_type> modul;
    modul.push_back(module);
    test(modul);

}