WKVQuant: Quantizing Weight and Key/Value Cache for Large Language Models Gains More