Trustworthy Large Vision-Language Model