作者位次 #4/7ACL 2026Poster· Natural Language Processing
Chain-of-Thought Compression Should Not Be Blind: V -Skip for Efficient Multimodal Reasoning via Dual-Path Anchoring
Dongxu Zhang · Yiding Sun · Cheng Tan · Wenbiao Yan · Ning Yang · Jihua Zhu · Haijun Zhang
摘要
While Chain-of-Thought (CoT) reasoning significantly enhances the performance of Multimodal Large Language Models (MLLMs), its autoregressive nature incurs prohibitive latency constraints. Current efforts to mitigate this via token compression often fail by blindly applying text-centric metrics t...