Large Language Models (LLMs) demand substantial computational resources for inference. GPUs offer high memory bandwidth