Multimodal discrete diffusion language models (dLLMs) generate responses through iterative mask prediction and offer a promising alternative to autoregressiv…
机构:清华
来源:arXiv 2611.05066 | AI4Papers 论文推荐平台