ValGrind shows potential memory leak

Hi, recently l noticed some potential memory leak in libtorch_cpu.
I am using Libtorch_cpu 1.8.1 in linux. Here is the c++ code that I used for testing.

#include <iostream>

#include <torch/torch.h>
#include <torch/script.h>

bool f()
{
    torch::NoGradGuard no_grad;
    std::cout << "--------------------------------------------------------------------" << std::endl;
    bool det = true;
    try {
        std::cout << "Engine test" << std::endl;
        //torch::jit::script::Module model = torch::jit::load("/home/ion3/Downloads/retinaface-mnet-complete.pt");
        torch::jit::script::Module model = torch::jit::load("/home/ion3/Downloads/arcface-r101.pt");
        model.eval();
        model.to(torch::kCPU);

        for (int i = 0; i < 100; ++i) {
            std::cout << i << std::endl;
            //torch::Tensor input_tensor = torch::rand({ 1, 3, 480, 640}, torch::kFloat32).to(torch::Device(torch::kCPU));
            torch::Tensor input_tensor = torch::rand({ 1, 3, 112, 112}, torch::kFloat32).to(torch::Device(torch::kCPU));
            //auto outputs = model.forward({ input_tensor }).toTuple();
            auto outputs = model.forward({ input_tensor });
        }
    }
    catch (std::exception& e) {
        std::cout << e.what() << std::endl;
        std::cout << "--------------------------------------------------------------------\n";
        return false;
    }
    std::cout << "--------------------------------------------------------------------\n";
    return det;
}


int main(int, char**) {
    bool res = f();
    return res;
}

I am not sure if this is libtorch bug or not because valgrind shows

==39105== LEAK SUMMARY:
==39105==    definitely lost: 0 bytes in 0 blocks
==39105==    indirectly lost: 0 bytes in 0 blocks
==39105==      possibly lost: 70,976 bytes in 5 blocks
==39105==    still reachable: 776,523 bytes in 12,086 blocks
==39105==         suppressed: 0 bytes in 0 blocks
==39105== 
==39105== ERROR SUMMARY: 3 errors from 3 contexts (suppressed: 0 from 0)

Any opinion regarding this problem?

PS:
here is the link for the full valgrind report. I have tested with 2 different model.
https://drive.google.com/file/d/1qHvIgSFTMzPLZH9Zxbyi4PpIAS57mDWw/view?usp=sharing