Search arXiv⌕ Search
Home/Archive/arXiv

arXiv · 2609.03235

SGD-KV: Summarization Guided KV Cache Compression

Zeyu Liu·Woomin Song·Xuandi Fu·Sai Muralidhar Jayanthi·Vivek Govindan·Aram Galstyan·Sravan Babu Bodapati·Srikanth Ronanki
View original source ↗Read source document ↗Export metadata ↓

Abstract

Large language models (LLMs) face severe memory bottlenecks in long-context inference due to the linearly growing size of key-value (KV) caches. Existing KV cache compression techniques typically rely on simple heuristics, overlooking the distinct functional roles of different attention heads. We present SGD-KV (Summarization-Guided KV Cache Compression), a head-aware framework that leverages a novel chunk-summarization diagnostic task to systematically identify and prioritize attention heads specialized in hierarchical information aggregation. Experiments on Qwen2.5-7B-1M and Qwen3-32B across diverse long-context benchmarks demonstrate that SGD-KV achieves state-of-the-art performance with contexts up to 1M tokens, while reducing KV cache memory usage by up to 75%. Our findings show that strategically allocating the KV cache budget based on the summarization score distribution of attention heads yields a superior efficiency-accuracy trade-off for long-context inference.

Explore related subjects

cs.CL

Keep this discovery

BibTeXRIS

Zeyu Liu, Woomin Song, Xuandi Fu, Sai Muralidhar Jayanthi, Vivek Govindan, Aram Galstyan, Sravan Babu Bodapati, Srikanth Ronanki. 2026-09-03. SGD-KV: Summarization Guided KV Cache Compression. https://arxiv.org/abs/2609.03235

Cite the original work for its findings. Save a collection to share your selection of sources.

Discover connections

cs.CL subject

Connections use source metadata and explicit phrase matches, not verified experimental comparisons.

KEEP EXPLORING

Related papers

arXiv2026

CyrillicQA: The Influence of Phonetically Encoded Secret Language on LLM Performance

Due to the selection of their training data, large language models (LLMs) perform best on standard-language inputs from languages using the Latin alphabet with large speaker populations, while disadvantaging other language varieties. Nevertheless, they can also be a versatile tool for preserving precisely such endangered languages. But do they also possess the necessary creativity and capacity for abstraction to decode phonetically encoded language the same way humans do?

cs.CL↗
arXiv2024

Density Matrices for Metaphor Understanding

In physics, density matrices are used to represent mixed states, i.e. probabilistic mixtures of pure states. This concept has previously been used to model lexical ambiguity. In this paper, we consider metaphor as a type of lexical ambiguity, and examine whether metaphorical meaning can be effectively modelled using mixtures of word senses. We find that modelling metaphor is significantly more difficult than other kinds of lexical ambiguity, but that our best-performing density matrix method outperforms simple baselines as well as some neural language models.

cs.CL↗
arXiv2026

Informational Antilocality and the Locality Bias in LLMs

We consider the ability of transformer-based language models (LLMs) to learn what we call k-antilocal languages, i.e., languages that have no mutual information across any span of $k$ contiguous symbols. We construct such languages with increasing $k$, finding that LLMs trained on them achieve comparable cross-entropy loss regardless of antilocality, but converge more slowly on more antilocal languages. Our findings support the idea that non-local dependencies are more difficult to learn, but the evidence for this bias comes from learning speed rather than learning success.

cs.CL↗