作者位次 #2/7ACL 2026Poster· Natural Language Processing
SCVQ : Sparse-Compensated Vector Quantization for Large Language Models
Zixuan Zhou · Yujun Diao · Zicheng Kong · Dehua Ma · Zhenbo Xu · Pei Pei Li · Zhaofeng He
摘要
Large Language Models (LLMs) are primarily constrained by memory and bandwidth bottlenecks during deployment. Although Vector Quantization (VQ) has emerged as a promising solution, existing methods incur inference overhead due to massive codebook storage and intensive index lookups. Moreover, the...