I am wondering why this 'residual' tensor behave like it used deepcopy? in pytorch

Question

I\'m running a standard transformer model on MT task.

I found something strange while I was looking at the code.

In the code below, that "residual" tenso