D-LLM: A Token Adaptive Computing Resource Allocation Strategy for Large Language Models | Read Paper on Bytez