# TraceKV: Policy-Preserving Low-Rank KV Cache Compression for Reasoning Vision-Language Models

**Authors:** Ke Li, Zishan Shao, Yiran Chen  
**Published:** 11 Nov 2026  
**Links:** [arXiv](https://arxiv.org/abs/2611.15569) · [PDF](https://arxiv.org/pdf/2611.15569) · [Code](https://github.com/AK-DREAM/TraceKV) · [This page](https://omen-bit.github.io/ai-paper-daily/paper/2611.15569.html)

## Key numbers

- **5.64×**: achieving near-lossless reasoning accuracy with 5.64× KV cache compression when combined with quantization
- **1.83×**: Optimized low-rank attention kernels further deliver up to 1.83× end-to-end decoding speedup

## Abstract

Reasoning vision-language models (VLMs) impose substantial KV cache demands from both large multimodal contexts and growing reasoning traces. Low-rank KV compression has emerged as a promising paradigm that stores low-dimensional latents of KV states and can be combined with quantization. However, existing low-rank methods often induce growing predictive policy drift from the dense model when compressing reasoning KV states, resulting in reasoning failures. We identify two key causes of this degradation: calibration-state mismatch between static contexts and on-policy reasoning states, and unequal error sensitivity across KV states. To address these challenges, we propose TraceKV, a training-free low-rank KV compression framework that unifies on-policy calibration, sensitivity-aware SVD, and global rank allocation under a single policy-preserving principle, substantially reducing policy-KL divergence during reasoning. Experiments across multiple VLMs and visual reasoning benchmarks demonstrate superior accuracy--memory trade-offs, achieving near-lossless reasoning accuracy with 5.64× KV cache compression when combined with quantization. Optimized low-rank attention kernels further deliver up to 1.83× end-to-end decoding speedup. Code will be available at https://github.com/AK-DREAM/TraceKV.

---

Notes written by AI from the paper, numbers copied from the abstract. Check the paper before you rely on them. From AI Papers Daily.