When Neural Code Completion Models Size up the Situation: Attaining Cheaper and Faster Completion through Dynamic Model Inference | IEEE Conference Publication | IEEE Xplore