作者位次 #2/10ACL 2026Poster· Natural Language Processing
APB - V : Accelerating Long-Video Understanding via Sequence-Parallelism-aware Approximate Attention
Yuxiang Huang · Mingye Li · Xu Han · Chaojun Xiao · Weilin Zhao · Ao Sun · Ziqi Yuan · Hao Zhou · Fandong Meng · Zhiyuan Liu
摘要
The efficiency of long-video inference remains a critical bottleneck, mainly due to the dense computation in the prefill stage of Large Multimodal Models (LMMs). Existing methods either compress visual embeddings or apply sparse attention on a single GPU, yielding limited acceleration or degraded...