Quantization for LLM: A HW/SW Co-Design Perspective