Knowledge-Aware Vision-Language Models