Optimize_for_mobile get a larger model file

I followed the mobile tutorial for a NMT model. After optimize_for_mobile and _save_for_lite_interpreter, I get a larger model file which is 127MB include encoder and decoder. But my pytorch model is an fp16 model with only 97MB before.
Is this situation correct? Are there any steps I’m missing?

cc @kimishpatel. I don’t think we support fb16.