An LMM bridges the vision-centric and vision-language tasks