1 paper
Ling-Xiang Zhang, Tao Hu, Da-Wei Zhou
Multimodal Large Language Models (MLLMs) achieve strong performance through instruction tuning, yet real-world applications often require them to continually acquire new capabiliti…