HuggingFace Daily Papers·· 1 天前
Nemotron-Labs-Diffusion-Uni：跨范式在线策略蒸馏的弹性自精炼统一扩散 LLM
Nemotron-Labs-Diffusion-Uni: Elastic Self-Refining Unified Diffusion LLM with Cross-Paradigm On-Policy Distillation
AI 导读
研究提出统一离散扩散大语言模型 NLD-Uni，用于视觉理解与生成任务。它引入跨范式在线策略蒸馏（OPD）框架，把自回归 VLM 的视觉理解能力和连续流模型的视觉生成能力迁移到统一 DLLM；并采用渐进式自精炼扩散过程，训练中分多步逐步破坏每个 token，推理时可对 token 做中间提交以实现渐进自精炼。
正文 · 原文
Abstract:We propose Nemotron-Labs-Diffusion-Uni (NLD-Uni), a unified discrete diffusion large language model (DLLM) for visual understanding and generation tasks. Compared with prior unified diffusion models, NLD-Uni introduces two major innovations. First, we propose a unified cross-paradigm on-policy distillation (OPD) framework that successfully transfers the visual understanding ability of autoregressive VLMs and visual generation ability of continuous flow models, to a unified DLLM. Second, NLD-Uni incorporates a novel progressive self-refining diffusion process that gradually corrupts each token across multiple steps during training, as opposed to adopting the conventional masked diffusion formulation that directly corrupts a clean image token to an uninformative mask token. At inference, this design allows NLD-Uni to make intermediate commit to a token, enabling progressive self-refinement for better visual quality. Through extensive experiments, we show that NLD-Uni achieves state-of-the-art performance among unified DLLMs on diverse tasks including visual question-answering, multi-modal reasoning, object grounding, text-to-image generation, and image editing.
| Subjects: | Computer Vision and Pattern Recognition (cs.CV) |
| Cite as: | arXiv:2610.36466 [cs.CV] |
| (or arXiv:2610.36466v1 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2610.36466 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Shufan Li [view email]
[v1]
Tue, 27 Oct 2026 01:21:26 UTC (15,003 KB)
来源：HuggingFace Daily Papers · arxiv.org