#LLaDA-V

LLaDA-V introduces a purely diffusion-based approach to multimodal large language modeling, achieving impressive results in visual instruction tuning and reasoning across diverse tasks.