Research

Paper

AI LLM February 24, 2026

CHESS: Context-aware Hierarchical Efficient Semantic Selection for Long-Context LLM Inference

Authors

Chao Fei, Guozhong Li, Chenxi Liu, Panos Kalnis

Abstract

Long-context LLMs demand accurate inference at low latency, yet decoding becomes primarily constrained by KV cache as context grows. Prior pruning methods are largely context-agnostic: their token selection ignores step-wise relevance and local semantics, which undermines quality. Moreover, their irregular accesses and selection overheads yield only limited wall-clock speedups. To address this, we propose \textbf{CHESS}, an \textit{algorithm-system co-design} KV-cache management system. Algorithmically, CHESS introduces a context-aware, hierarchical selection policy that dynamically reconstructs a coherent context for the current decoding. System-wise, coarse granularity selection eliminates expensive data movement, fully realizing practical acceleration from theoretical sparsity. Extensive evaluations demonstrate that CHESS surpasses Full-KV quality using only \textbf{1\%} of the KV cache, delivers low-latency stable inference with up to \textbf{4.56$\times$} higher throughput, and consistently outperforms other strong baselines. Code is available at \href{https://anonymous.4open.science/r/CHESS-9958/}{https://anonymous.4open.science/r/CHESS/}.

Metadata

arXiv ID: 2602.20732
Provider: ARXIV
Primary Category: cs.AI
Published: 2026-02-24
Fetched: 2026-02-25 06:05

Related papers

Raw Data (Debug)
{
  "raw_xml": "<entry>\n    <id>http://arxiv.org/abs/2602.20732v1</id>\n    <title>CHESS: Context-aware Hierarchical Efficient Semantic Selection for Long-Context LLM Inference</title>\n    <updated>2026-02-24T09:54:59Z</updated>\n    <link href='https://arxiv.org/abs/2602.20732v1' rel='alternate' type='text/html'/>\n    <link href='https://arxiv.org/pdf/2602.20732v1' rel='related' title='pdf' type='application/pdf'/>\n    <summary>Long-context LLMs demand accurate inference at low latency, yet decoding becomes primarily constrained by KV cache as context grows. Prior pruning methods are largely context-agnostic: their token selection ignores step-wise relevance and local semantics, which undermines quality. Moreover, their irregular accesses and selection overheads yield only limited wall-clock speedups. To address this, we propose \\textbf{CHESS}, an \\textit{algorithm-system co-design} KV-cache management system. Algorithmically, CHESS introduces a context-aware, hierarchical selection policy that dynamically reconstructs a coherent context for the current decoding. System-wise, coarse granularity selection eliminates expensive data movement, fully realizing practical acceleration from theoretical sparsity. Extensive evaluations demonstrate that CHESS surpasses Full-KV quality using only \\textbf{1\\%} of the KV cache, delivers low-latency stable inference with up to \\textbf{4.56$\\times$} higher throughput, and consistently outperforms other strong baselines. Code is available at \\href{https://anonymous.4open.science/r/CHESS-9958/}{https://anonymous.4open.science/r/CHESS/}.</summary>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.AI'/>\n    <published>2026-02-24T09:54:59Z</published>\n    <arxiv:primary_category term='cs.AI'/>\n    <author>\n      <name>Chao Fei</name>\n    </author>\n    <author>\n      <name>Guozhong Li</name>\n    </author>\n    <author>\n      <name>Chenxi Liu</name>\n    </author>\n    <author>\n      <name>Panos Kalnis</name>\n    </author>\n  </entry>"
}