Quantization is one of the techniques for accelerating the execution of large language models (LLMs). However, different quantized values exhibit different computational and power characteristics. In this master's thesis, your task is to analyze the impact of different weight values on the latency and power of a decoder-only LLM model. Based on these insights, you will then quantize the model using values that minimize computational latency and power consumption, ultimately achieving a faster and more energy-efficient decoder-only LLM model.