Ayan Sengupta

Vision-language models can significantly enhance language models through knowledge distillation, even without direct textual understanding, challenging conventional KD paradigms.