Performance of repeated torch::jit::load() calls

Consider:

void loop() {
  torch::jit::script::Module module;
  while (true) {
    module = torch::jit::load(x);
    ...
  }
}

If I know that on each load() after the first, that the module architecture will remain unchanged, and that only the weights will change, is it possible to make those load() calls faster somehow?