Saved in:
Bibliographic Details
Main Authors: Szilvasy, Gergely, Faysse, Manuel, Lomeli, Maria, Douze, Matthijs, Mazaré, Pierre-Emmanuel, Cabannes, Loïc, Yih, Wen-tau, Jégou, Hervé
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2605.14037
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914564903272448
author Szilvasy, Gergely
Faysse, Manuel
Lomeli, Maria
Douze, Matthijs
Mazaré, Pierre-Emmanuel
Cabannes, Loïc
Yih, Wen-tau
Jégou, Hervé
author_facet Szilvasy, Gergely
Faysse, Manuel
Lomeli, Maria
Douze, Matthijs
Mazaré, Pierre-Emmanuel
Cabannes, Loïc
Yih, Wen-tau
Jégou, Hervé
contents Under modern test-time compute and agentic paradigms, language models process ever-longer sequences. Efficient text generation with transformer architectures is increasingly constrained by the Key-Value cache memory footprint and bandwidth. To address this limitation, we introduce Self-Pruned Key-Value Attention (SP-KV), a mechanism designed to predict future KV utility in order to reduce the size of the long-term KV cache. This strategy operates at a fine granularity: a lightweight utility predictor scores each key-value pair, and while recent KVs are always available via a local window, older pairs are written in the cache and used in global attention only if their predicted utility surpasses a given threshold. The LLM and the utility predictor are trained jointly end-to-end exclusively through next-token prediction loss, and are adapted from pretrained LLM checkpoints. Rather than enforcing a fixed compression ratio, SP-KV performs dynamic sparsification: the mechanism adapts to the input and typically reduces the KV cache size by a factor of $3$ to $10\times$, longer sequences often being more compressible. This leads to vast improvements in memory usage and decoding speed, with little to no degradation of validation loss nor performance on a broad set of downstream tasks. Beyond serving as an effective KV-cache reduction mechanism, our method reveals structured layer- and head-specific sparsity patterns that we can use to guide the design of hybrid local-global attention architectures.
format Preprint
id arxiv_https___arxiv_org_abs_2605_14037
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Self-Pruned Key-Value Attention: Learning When to Write by Predicting Future Utility
Szilvasy, Gergely
Faysse, Manuel
Lomeli, Maria
Douze, Matthijs
Mazaré, Pierre-Emmanuel
Cabannes, Loïc
Yih, Wen-tau
Jégou, Hervé
Machine Learning
Computation and Language
I.2.6; I.2.7
Under modern test-time compute and agentic paradigms, language models process ever-longer sequences. Efficient text generation with transformer architectures is increasingly constrained by the Key-Value cache memory footprint and bandwidth. To address this limitation, we introduce Self-Pruned Key-Value Attention (SP-KV), a mechanism designed to predict future KV utility in order to reduce the size of the long-term KV cache. This strategy operates at a fine granularity: a lightweight utility predictor scores each key-value pair, and while recent KVs are always available via a local window, older pairs are written in the cache and used in global attention only if their predicted utility surpasses a given threshold. The LLM and the utility predictor are trained jointly end-to-end exclusively through next-token prediction loss, and are adapted from pretrained LLM checkpoints. Rather than enforcing a fixed compression ratio, SP-KV performs dynamic sparsification: the mechanism adapts to the input and typically reduces the KV cache size by a factor of $3$ to $10\times$, longer sequences often being more compressible. This leads to vast improvements in memory usage and decoding speed, with little to no degradation of validation loss nor performance on a broad set of downstream tasks. Beyond serving as an effective KV-cache reduction mechanism, our method reveals structured layer- and head-specific sparsity patterns that we can use to guide the design of hybrid local-global attention architectures.
title Self-Pruned Key-Value Attention: Learning When to Write by Predicting Future Utility
topic Machine Learning
Computation and Language
I.2.6; I.2.7
url https://arxiv.org/abs/2605.14037