Submitted by Songtao Wei 6 LEAD: Length-Efficient Adaptive and Dynamic Reasoning for Large Language Models University of Texas at Dallas 3 1