Thought-Aware KV Cache Compaction for Reasoning via Adaptive Attention Matching
Reasoning language models generate lengthy chain-of-thought sequences, making the KV cache grow linearly and become a memory bottleneck. Existing compression methods treat reasoning as flat token sequences and apply uniform compression, missing the hierarchical importance of reasoning steps. This paper introduces Thought-Aware Attention Matching (TAM), which combines thought segmentation, adaptive budget allocation, and pivotal token protection. TAM is theoretically optimal under a convex error model, and experiments with Qwen3-4B on AIME 2024 and MATH-500 show improved accuracy over uniform compaction at the same memory footprint, with periodic compaction cutting peak memory to approximately 3.1–3.2 GB (a 65% reduction).
-->
[Submitted on 1 Jun 2026]
Title:Thought-Aware KV Cache Compaction for Reasoning via Adaptive Attention Matching
View a PDF of the paper titled Thought-Aware KV Cache Compaction for Reasoning via Adaptive Attention Matching, by Yang Liu and 5 other authors
View PDF HTML (experimental)
Abstract:Reasoning language models generate lengthy chain-of-thought (CoT) sequences whose key-value (KV) cache grows linearly and becomes a memory bottleneck during decoding. Existing compaction methods treat reasoning trajectories as flat token sequences and apply uniform compression, ignoring the hierarchical structure of CoT reasoning where different steps vary drastically in importance. We propose \textbf{Thought-Aware Attention Matching (TAM)}, which exploits this structure through three mechanisms: (i)~thought segmentation that decomposes the trajectory into reasoning blocks, (ii)~adaptive budget allocation that assigns compression budget based on each segment's importance and size, and (iii)~pivotal token protection that preserves high-attention reasoning anchors. We prove that the allocation rule is optimal under a convex error model and that cumulative error under sequential compaction remains bounded. Experiments on AIME 2024 and MATH-500 with Qwen3-4B show that TAM improves accuracy over uniform compaction at the same memory footprint, with periodic compaction bounding peak memory to 3.1--3.2\,GB (a 65\% reduction) while maintaining competitive accuracy.
Comments: 16 pages, 5 figures
Subjects:
Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
Cite as: arXiv:2608.12331 [cs.CL]
(or arXiv:2608.12331v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2608.12331
arXiv-issued DOI via DataCite
Submission history
From: Yang Liu Aron [view email] [v1] Mon, 1 Jun 2026 23:47:00 UTC (853 KB)
Full-text links:
Access Paper:
View a PDF of the paper titled Thought-Aware KV Cache Compaction for Reasoning via Adaptive Attention Matching, by Yang Liu and 5 other authors
View PDF
HTML (experimental)
TeX Source
view license
Current browse context:
cs.CL
new | recent | 2026-08
Change to browse by:
cs cs.AI
References & Citations
NASA ADS
Google Scholar
Semantic Scholar
Loading...
Data provided by:
Bibliographic Tools
Bibliographic and Citation Tools
Bibliographic Explorer Toggle
Bibliographic Explorer (What is the Explorer?)
Connected Papers Toggle
Connected Papers (What is Connected Papers?)
Litmaps Toggle
Litmaps (What is Litmaps?)
scite.ai Toggle
scite Smart Citations (What are Smart Citations?)
Code, Data, Media
Code, Data and Media Associated with this Article
alphaXiv Toggle
alphaXiv (What is alphaXiv?)
Links to Code Toggle
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub Toggle
DagsHub (What is DagsHub?)
GotitPub Toggle
Gotit.pub (What is GotitPub?)
Huggingface Toggle
Hugging Face (What is Huggingface?)
ScienceCast Toggle
ScienceCast (What is ScienceCast?)
Demos
Demos
Replicate Toggle
Replicate (What is Replicate?)
Spaces Toggle
Hugging Face Spaces (What is Spaces?)
Spaces Toggle
TXYZ.AI (What is TXYZ.AI?)
Related Papers
Recommenders and Search Tools
Link to Influence Flower
Influence Flower (What are Influence Flowers?)
Core recommender toggle
CORE Recommender (What is CORE?)
Author
Venue
Institution
Topic
About arXivLabs
arXivLabs: experimental projects with community collaborators
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Which authors of this paper are endorsers? | Disable MathJax (What is MathJax?)