DM-Align：基于样本引导分布匹配的视频生成联合对齐与蒸馏
原标题：Joint Alignment and Distillation for Video Generation via Sample-Guided Distribution Matching
AI 摘要
arXiv 论文提出 DM-Align，一种基于分布匹配的统一单阶段优化框架，用于视频生成模型的蒸馏与人类偏好对齐。该方法通过样本引导的分布匹配梯度，同时优化蒸馏质量和偏好对齐，避免了 RL 在蒸馏前的高计算成本和蒸馏后的模型崩溃。实验表明，DM-Align 在多个基础视频模型上优于独立变体和两阶段流水线。
正文节选
Joint Alignment and Distillation for Video Generation via Sample-Guided Distribution Matching Abstract Aligning video generative models to human preferences heavily relies on Reinforcement Learning (RL), which suffers from extensive computational overhead. Existing workflows typically treat RL and distillation as disconnected stages: applying RL before distillation incurs prohibitive computational costs, whereas applying RL after distillation frequently leads to model collapse. To overcome the