Beyond Speedup -- Utilizing KV Cache for Sampling and Reasoning

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Xing, Zeyu, Li, Xing, Zhen, Hui-Ling, Yuan, Mingxuan, Pan, Sinno Jialin
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911404139741184
author Xing, Zeyu
Li, Xing
Zhen, Hui-Ling
Yuan, Mingxuan
Pan, Sinno Jialin
author_facet Xing, Zeyu
Li, Xing
Zhen, Hui-Ling
Yuan, Mingxuan
Pan, Sinno Jialin
contents KV caches, typically used only to speed up autoregressive decoding, encode contextual information that can be reused for downstream tasks at no extra cost. We propose treating the KV cache as a lightweight representation, eliminating the need to recompute or store full hidden states. Despite being weaker than dedicated embeddings, KV-derived representations are shown to be sufficient for two key applications: \textbf{(i) Chain-of-Embedding}, where they achieve competitive or superior performance on Llama-3.1-8B-Instruct and Qwen2-7B-Instruct; and \textbf{(ii) Fast/Slow Thinking Switching}, where they enable adaptive reasoning on Qwen3-8B and DeepSeek-R1-Distil-Qwen-14B, reducing token generation by up to $5.7\times$ with minimal accuracy loss. Our findings establish KV caches as a free, effective substrate for sampling and reasoning, opening new directions for representation reuse in LLM inference. Code: https://github.com/cmd2001/ICLR2026_KV-Embedding.
format Preprint
id arxiv_https___arxiv_org_abs_2601_20326
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Beyond Speedup -- Utilizing KV Cache for Sampling and Reasoning
Xing, Zeyu
Li, Xing
Zhen, Hui-Ling
Yuan, Mingxuan
Pan, Sinno Jialin
Computation and Language
Artificial Intelligence
Machine Learning
KV caches, typically used only to speed up autoregressive decoding, encode contextual information that can be reused for downstream tasks at no extra cost. We propose treating the KV cache as a lightweight representation, eliminating the need to recompute or store full hidden states. Despite being weaker than dedicated embeddings, KV-derived representations are shown to be sufficient for two key applications: \textbf{(i) Chain-of-Embedding}, where they achieve competitive or superior performance on Llama-3.1-8B-Instruct and Qwen2-7B-Instruct; and \textbf{(ii) Fast/Slow Thinking Switching}, where they enable adaptive reasoning on Qwen3-8B and DeepSeek-R1-Distil-Qwen-14B, reducing token generation by up to $5.7\times$ with minimal accuracy loss. Our findings establish KV caches as a free, effective substrate for sampling and reasoning, opening new directions for representation reuse in LLM inference. Code: https://github.com/cmd2001/ICLR2026_KV-Embedding.
title Beyond Speedup -- Utilizing KV Cache for Sampling and Reasoning
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2601.20326