动态专家路由让视觉自回归模型更省算力且不降质
Dynamic Mixture-of-Experts for Visual Autoregressive Model
- 引入动态专家路由，按图像复杂度和分辨率智能选专家
- 减少20%计算量，推理速度提升11%，质量持平全连接模型
- 无需额外训练，适合追求高效生成的部署场景
视觉自回归模型（VAR）在图像生成中兼具高效与高质量，但随着分辨率提升，重复调用Transformer导致计算冗余。本文提出一种集成于VAR的动态专家路由机制，通过尺度感知阈值策略，在不需额外训练的前提下，根据令牌复杂度和分辨率动态平衡专家选择。实验表明，该方法在保持与密集基线相当图像质量的同时，减少20%的浮点运算量（FLOPs），推理速度提升11%。
原文摘要 · Abstract (English)
Visual Autoregressive Models (VAR) offer efficient and high-quality image generation but suffer from computational redundancy due to repeated Transformer calls at increasing resolutions. We introduce a dynamic Mixture-of-Experts router integrated into VAR. The new architecture allows to trade compute for quality through scale-aware thresholding. This thresholding strategy balances expert selection based on token complexity and resolution, without requiring additional training. As a result, we achieve 20% fewer FLOPs, 11% faster inference and match the image quality achieved by the dense baseline.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。