is there weight initialization process in model.half()?

Question

I am dealing with bert-base pretrained model finetuning for korean.
however, I wanted speed acceleration, so i changed model to fp16 by model.half()
by inspectin