一作ACL 2026Poster· Natural Language Processing
Sparrow: Text-Anchored Window Attention with Visual-Semantic Glimpsing for Speculative Decoding in Video LLM s
Libo Zhang · Zhaoning Zhang · Hongwanyang · Peng Qiao · Dongsheng Li
摘要
Although speculative decoding is widely used to accelerate Vision-Language Models (VLMs) inference, it faces severe performance collapse when applied to Video Large Language Models (Vid-LLMs). The draft model typically falls into the trap of attention dilution and negative visual gain due to key-...