How to handle multiple inputs and outputs?

What is the best practice for adding multiple inputs and outputs for torch::jit::script::Module ?

I use below method:

// model
torch::jit::script::Module net_;

// input
std::vector<torch::jit::IValue> torch_inputs;
torch::Tensor a,b,c;
torch_inputs.push_back(a);
torch_inputs.push_back(b);
torch_inputs.push_back(c);

// 
std::vector<at::Tensor> torch_outputs = 
   net_.forward(torch_inputs).toTensorVector();

// output
torch::Tensor o1 = torch_outputs[0];
torch::Tensor o2 = torch_outputs[1];
torch::Tensor o3 = torch_outputs[2];

My question is whether it is the best practice to do this, and how to be compatible if the input situation is not like this?

1 Like