Do Language Models Use Their Depth Efficiently? | Read Paper on Bytez