New Method For LLM Quantization