Latent-Condensed Transformer for Efficient Long Context Modeling

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: You, Zeng, Chen, Yaofo, Chen, Qiuwu, Sun, Ying, Zhang, Shuhai, Li, Yingjian, Wang, Yaowei, Tan, Mingkui
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908969460563968
author You, Zeng
Chen, Yaofo
Chen, Qiuwu
Sun, Ying
Zhang, Shuhai
Li, Yingjian
Wang, Yaowei
Tan, Mingkui
author_facet You, Zeng
Chen, Yaofo
Chen, Qiuwu
Sun, Ying
Zhang, Shuhai
Li, Yingjian
Wang, Yaowei
Tan, Mingkui
contents Large language models (LLMs) face significant challenges in processing long contexts due to the linear growth of the key-value (KV) cache and quadratic complexity of self-attention. Existing approaches address these bottlenecks separately: Multi-head Latent Attention (MLA) reduces the KV cache by projecting tokens into a low-dimensional latent space, while sparse attention reduces computation. However, sparse methods cannot operate natively on MLA's compressed latent structure, missing opportunities for joint optimization. In this paper, we propose Latent-Condensed Attention (LCA), which directly condenses context within MLA's latent space, where the representation is disentangled into semantic latent vectors and positional keys. LCA separately aggregates semantic vectors via query-aware pooling and preserves positional keys via anchor selection. This approach jointly reduces both computational cost and KV cache without adding parameters. Beyond MLA, LCA's design is architecture-agnostic and readily extends to other attention mechanisms such as GQA. Theoretically, we prove a length-independent error bound. Experiments show LCA achieves up to 2.5$\times$ prefilling speedup and 90% KV cache reduction at 128K context while maintaining competitive performance.
format Preprint
id arxiv_https___arxiv_org_abs_2604_12452
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Latent-Condensed Transformer for Efficient Long Context Modeling
You, Zeng
Chen, Yaofo
Chen, Qiuwu
Sun, Ying
Zhang, Shuhai
Li, Yingjian
Wang, Yaowei
Tan, Mingkui
Computation and Language
Large language models (LLMs) face significant challenges in processing long contexts due to the linear growth of the key-value (KV) cache and quadratic complexity of self-attention. Existing approaches address these bottlenecks separately: Multi-head Latent Attention (MLA) reduces the KV cache by projecting tokens into a low-dimensional latent space, while sparse attention reduces computation. However, sparse methods cannot operate natively on MLA's compressed latent structure, missing opportunities for joint optimization. In this paper, we propose Latent-Condensed Attention (LCA), which directly condenses context within MLA's latent space, where the representation is disentangled into semantic latent vectors and positional keys. LCA separately aggregates semantic vectors via query-aware pooling and preserves positional keys via anchor selection. This approach jointly reduces both computational cost and KV cache without adding parameters. Beyond MLA, LCA's design is architecture-agnostic and readily extends to other attention mechanisms such as GQA. Theoretically, we prove a length-independent error bound. Experiments show LCA achieves up to 2.5$\times$ prefilling speedup and 90% KV cache reduction at 128K context while maintaining competitive performance.
title Latent-Condensed Transformer for Efficient Long Context Modeling
topic Computation and Language
url https://arxiv.org/abs/2604.12452