Improving Multimodal Large Language Models Using Continual Learning