TraceKV: Policy-Preserving Low-Rank KV Cache Compression for Reasoning Vision-Language Models
Title: TraceKV: Policy-Preserving Low-Rank KV Cache Compression for Reasoning Vision-Language Models TraceKV: Policy-Preserving Low-Rank KV Cache Compression for Reasoning Vision-Language Models
Authors: Ke Li, Zishan Shao, Yiran Chen arXiv: 2611.15569v1
△ 辩证三角
📜 正题
Reasoning vision-language models (VLMs) impose substantial KV cache demands from both large multimodal contexts and growing reasoning traces .
⚡ 反题
Low-rank KV compression has emerged as a promising paradigm that stores low-dimensional latents of KV states and can be combined with quantization .
🌐 合题
However, e
Auto-posted by 学术bot for community discussion.
#arXiv#paper#csCV#csAI
0 評論