自适应匹配蒸馏优化少步生成

Optimizing Few-Step Generation with Adaptive Matching Distillation

February 7, 2026

作者: Lichen Bai, Zikai Zhou, Shitong Shao, Wenliang Zhong, Shuo Yang, Shuo Chen, Bojun Chen, Zeke Xie

cs.AI

摘要

分布匹配蒸馏（DMD）是一种高效的加速范式，但其在"禁区"（即真实教师提供不可靠指导而虚拟教师排斥力不足的区域）中的稳定性常受影响。本研究提出统一优化框架，将现有技术重新阐释为规避这些污染区域的隐式策略。基于此洞见，我们引入自适应匹配蒸馏（AMD），该自校正机制利用奖励代理显式检测并逃离禁区。AMD通过结构信号分解动态优先处理校正梯度，并采用排斥势场锐化技术构建陡峭能量壁垒以防止失败模式坍缩。在图像与视频生成任务（如SDXL、Wan2.1）和严格基准测试（如VBench、GenEval）上的大量实验表明，AMD显著提升了样本保真度与训练鲁棒性。例如在SDXL模型上，AMD将HPSv2分数从30.64提升至31.25，超越现有最优基线。这些发现证实，在禁区内显式修正优化轨迹对于突破少步生成模型的性能瓶颈至关重要。

English

Distribution Matching Distillation (DMD) is a powerful acceleration paradigm, yet its stability is often compromised in Forbidden Zone, regions where the real teacher provides unreliable guidance while the fake teacher exerts insufficient repulsive force. In this work, we propose a unified optimization framework that reinterprets prior art as implicit strategies to avoid these corrupted regions. Based on this insight, we introduce Adaptive Matching Distillation (AMD), a self-correcting mechanism that utilizes reward proxies to explicitly detect and escape Forbidden Zones. AMD dynamically prioritizes corrective gradients via structural signal decomposition and introduces Repulsive Landscape Sharpening to enforce steep energy barriers against failure mode collapse. Extensive experiments across image and video generation tasks (e.g., SDXL, Wan2.1) and rigorous benchmarks (e.g., VBench, GenEval) demonstrate that AMD significantly enhances sample fidelity and training robustness. For instance, AMD improves the HPSv2 score on SDXL from 30.64 to 31.25, outperforming state-of-the-art baselines. These findings validate that explicitly rectifying optimization trajectories within Forbidden Zones is essential for pushing the performance ceiling of few-step generative models.