Head Forcing:利用注意力头异质性实现长时自回归视频生成
TLDR
Head Forcing 通过异构注意力头定制 KV 缓存策略,无需训练即可实现分钟级视频生成。
评分理由
The paper identifies functional heterogeneity in attention heads and proposes a training-free, head-specific caching and memory scheme to mitigate error accumulation in long video generation. Strengths include a novel analysis and practical, plug-in approach; weaknesses are that abstract-level evidence lacks detailed benchmarks and the connection to broader world models is unstated.
Read-first 评分解释
综合优先阅读分 18.2,由主题、引用、图谱、方法、可复现性和近期性等信号加权得到。 原始总分保留为 6。
研究版图角色
前沿论文
排序敏感性
稳定性:volatile;排名波动范围:31。