Load pre-trained torch model but not get the same result

1

I use load_lua to load torch model successfully. But the result seems to be worse than what produced by the original torch model. Are there any reasons or solutions?