Searcharxiv⌕ Search
Home/Archive/arXiv

arXiv · 2608.25230

Trust the Mass: Forced Weights in KV-Cache Eviction

Jack Shi·Jerry Gu
View original source ↗Read source document ↗Export metadata ↓

Abstract

Every deployed sparse-attention or KV-cache-eviction rule keeps a subset of the keys, discards the rest, and renormalizes the attention weights over the kept set. Enumerating the exact best subset under that constraint on $168{,}192$ attention rows from five models shows that keeping the largest weights is already near-optimal, since the best subset closes only a median $2$ to $5\%$ of the remaining gap to full attention. If selection closes this little, published margins between eviction methods must come from elsewhere, so we measure the bytes each method holds. In the shared evaluation pipeline, the strongest query-agnostic methods hold the full cache because their per-head selections are stored as masks, and only ragged per-head storage frees that memory. Enforcing a nominal budget on one fixed selection costs $14$ to $62$ benchmark points. We trace an $87.6$-point retrieval margin to rankings computed while the question is visible. ContourKV, a training-free allocator built from the dropped-mass statistic, wins $93$ of $160$ paired comparisons against that state of the art and loses $22$ at the byte count of the budget-enforcing baselines, and it ties the strongest of them.

Explore related subjects

cs.LGcs.CL

Keep this discovery

BibTeXRIS

Jack Shi, Jerry Gu. 2026-08-28. Trust the Mass: Forced Weights in KV-Cache Eviction. https://arxiv.org/abs/2608.25230

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

Discover connections

cs.CL subjectcs.LG subject

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

KEEP EXPLORING

Related discoveries

arXiv2026

Leveraging Turn-taking Dynamics for Intent Recognition in Multi-party Conversations

We propose a multi-task learning approach for multi-party dialogue intent recognition that leverages an auxiliary task that models turn-taking dynamics. Specifically, we introduce turn-transition entropy, a self-supervised target computed from the sequence of speaker transitions, which quantifies the predictability of interaction patterns. Experiments on multiple pre-trained models demonstrate that incorporating this auxiliary task improves intent recognition performance, outperforming existing approaches which ignore multi-party interaction dynamics. We find that our proposed continuous target can be learned as a single-task objective, suggesting that it is an actual signal carrying useful information.

cs.CL↗
arXiv2024

ContrastWSD: Enhancing Metaphor Detection with Word Sense Disambiguation Following the Metaphor Identification Procedure

This paper presents ContrastWSD, a RoBERTa-based metaphor detection model that integrates the Metaphor Identification Procedure (MIP) and Word Sense Disambiguation (WSD) to extract and contrast the contextual meaning with the basic meaning of a word to determine whether it is used metaphorically in a sentence. By utilizing the word senses derived from a WSD model, our model enhances the metaphor detection process and outperforms other methods that rely solely on contextual embeddings or integrate only the basic definitions and other external knowledge. We evaluate our approach on various benchmark datasets and compare it with strong baselines, indicating the effectiveness in advancing metaphor detection.

cs.CL↗
arXiv2026

Bayesian Sparse Low-Rank Adaptation for Large Language Model Uncertainty Estimation

Large language models (LLMs) exhibit remarkable reasoning capabilities, but their task-specific fine-tuning is notoriously plagued by overconfidence, severely hindering trustworthy deployment. We propose Data-Adaptive Lower-Rank Adaptation (DALorRA), a simple and effective variational Bayesian sparse framework that shifts the paradigm of uncertainty quantification from the dense parameter space to the lightweight rank level of low-rank adaptation (LoRA). With the insight that LoRA essentially aggregates multiple rank-one components that may provide superfluous model capacity, DALorRA imposes stochastic masking on rank dimensions, enabling Bayesian regularization of model capacity during training and ensemble-like calibration during inference. Extensive experiments demonstrate DALorRA's excellent calibration of LLMs without compromising reasoning accuracy.

cs.LG↗