作者位次 #4/7ACL 2026Poster· Natural Language Processing
Stability Implies Redundancy: Delta Attention Selective Halting for Efficient Long-Context Prefilling
Yujie Chen · Tailai Chen · Yifeng Gao · Zoe Wanying He · Yijue Xu · Shaobo Wang · Linfeng Zhang
摘要
Prefilling computational costs pose a significant bottleneck for Large Language Models (LLMs) and Large Multimodal Models (LMMs) in long-context settings. While token pruning reduces sequence length, prior methods rely on heuristics that break compatibility with hardware-efficient kernels like Fl...