Due to unknown reason, repeat calling of forward method of some models might not produce identical results. Therefore, I need to load the model repeatedly, to ensure consistency. By creating a memory buffer, I hope it will speed up the process a little bit.

let mut file = std::fs::File::open(model_path).unwrap();
let mut buffer: Vec = Vec::new();
file.read_to_end(&mut buffer).unwrap();
let mut model = tch::CModule::load_buffer_on_device(&buffer, device).unwrap();