Skip to content
AI News HubLIVE
Original source2 min read

VarRate: Training-Free Variable-Rate KV Cache Compression for Long-Context LLMs

Summary

VarRate is a training-free KV cache codec that allocates a variable low-rank budget to each token based on query salience instead of evicting tokens. At a 20% budget on LongBench, it stays within 0.8 points of the uncompressed model on Llama-3.1-8B and Qwen2.5-7B, and matches KVzip accuracy with about one-eighth the prefill overhead.

SourcearXiv Computational LinguisticsAuthor: Shahrzad Esmat, Dhawal Shah, Ali Jannesari
VarRate: Training-Free Variable-Rate KV Cache Compression for Long-Context LLMs
Report an error

The correction channel is not available yet. You can copy the article reference below for later.

Correction instructions
Read article

[Submitted on 16 Jul 2026]

Title:VarRate: Training-Free Variable-Rate KV Cache Compression for Long-Context LLMs

View a PDF of the paper titled VarRate: Training-Free Variable-Rate KV Cache Compression for Long-Context LLMs, by Shahrzad Esmat and 2 other authors

View PDF HTML (experimental)

Abstract:The key-value (KV) cache is the main memory bottleneck in long-context large language model (LLM) inference. Two leading training-free families are both structurally limited: token-selection methods (SnapKV, Ada-KV) score importance from an observation window and evict low-scoring tokens, but eviction is irreversible -- so when the importance signal degrades under query-agnostic reuse, accuracy collapses by 11-15 points; uniform low-rank coding keeps every token but spends equal rank everywhere, wasting budget. We observe that both failures share one cure: rank should be allocated, not evicted. We present VarRate, a training-free KV codec that assigns each token a variable low-rank budget by its query salience, keeping every token at a nonzero rank. Comparable adaptive-rank codecs reach this allocation only through training; VarRate requires none. Because no token is dropped, it degrades by only 3.5-5.5 points where query-aware selection collapses. At a matched 20% budget on LongBench (16 tasks), VarRate stays within 0.8 points of the uncompressed model on both Llama-3.1-8B and Qwen2.5-7B. Averaged over the two, it is the strongest matched-memory compressor. It significantly beats its uniform-rank ablation on both models. Against KVzip, a method purpose-built for query-agnostic reuse, it is accuracy-equivalent in three of four settings and within a point overall, at about one-eighth the prefill overhead.

Comments: 20 pages, 8 figures, 24 tables. Includes appendix with additional experiments and analyses

Subjects:

Computation and Language (cs.CL); Machine Learning (cs.LG)

Cite as: arXiv:2607.15498 [cs.CL]

(or arXiv:2607.15498v1 [cs.CL] for this version)

https://doi.org/10.48550/arXiv.2607.15498

arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Shahrzad Esmat [view email] [v1] Thu, 16 Jul 2026 23:03:14 UTC (416 KB)

Full-text links:

Access Paper:

View a PDF of the paper titled VarRate: Training-Free Variable-Rate KV Cache Compression for Long-Context LLMs, by Shahrzad Esmat and 2 other authors

View PDF

HTML (experimental)

TeX Source

view license

Current browse context:

cs.CL

new | recent | 2026-07

Change to browse by:

cs cs.LG

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?)

Key points and analysis

Article intelligence

EngineersAdvanced

Key points

  • Allocates rank per token rather than evicting, avoiding irreversible information loss
  • Within 0.8 points of uncompressed models at 20% budget on LongBench
  • Degrades only 3.5-5.5 points under query-agnostic reuse where selection methods collapse by 11-15 points
  • About 1/8 the prefill overhead of KVzip at equivalent accuracy

Highlights and analysis are generated automatically and may contain errors. Check the original source.