ART: Attention Run-time Termination for Efficient Large Language Model Decoding

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Qiu, Chen, Li, Guozhong, Kalnis, Panos
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866914619115700224
author Qiu, Chen
Li, Guozhong
Kalnis, Panos
author_facet Qiu, Chen
Li, Guozhong
Kalnis, Panos
contents Long-context decoding in Large Language Models (LLMs) is severely constrained by the memory bandwidth required to fetch the extensive Key-Value (KV) cache. Most existing KV management methods rely on key-only pruning before decoding, despite the evidence that attention outputs depend jointly on keys and values, as incorporating values in their methods incurs prohibitive additional overhead. In this paper, we propose Attention Run-time Termination (ART), a lightweight run-time mechanism that tracks accumulated attention outputs during kernel execution and terminates subsequent KV block accesses once further contributions become negligible. This design makes ART orthogonal to existing key-based KV cache management methods, enabling seamless integration with them. Experiments on LongBench benchmarks show that ART achieves 20% higher generation throughput in large batch size than state-of-the-art baseline while maintaining comparable accuracy.
format Preprint
id arxiv_https___arxiv_org_abs_2606_00024
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle ART: Attention Run-time Termination for Efficient Large Language Model Decoding
Qiu, Chen
Li, Guozhong
Kalnis, Panos
Computation and Language
Long-context decoding in Large Language Models (LLMs) is severely constrained by the memory bandwidth required to fetch the extensive Key-Value (KV) cache. Most existing KV management methods rely on key-only pruning before decoding, despite the evidence that attention outputs depend jointly on keys and values, as incorporating values in their methods incurs prohibitive additional overhead. In this paper, we propose Attention Run-time Termination (ART), a lightweight run-time mechanism that tracks accumulated attention outputs during kernel execution and terminates subsequent KV block accesses once further contributions become negligible. This design makes ART orthogonal to existing key-based KV cache management methods, enabling seamless integration with them. Experiments on LongBench benchmarks show that ART achieves 20% higher generation throughput in large batch size than state-of-the-art baseline while maintaining comparable accuracy.
title ART: Attention Run-time Termination for Efficient Large Language Model Decoding
topic Computation and Language
url https://arxiv.org/abs/2606.00024