This work learns a selector that is able to produce, based on context, a per-layer cache configuration towards an overall compression budget, and achieves the Pareto frontier of accuracy against cache size, against learning-free and post-hoc baselines.
Abstract
As the context size of text processed with an LLM grows, the size of KV caches can outstrip the memory allocated for the original model weights. This impacts LLM throughput negatively, since decoding is memory-bound and decode cost grows with cache size. Recent work alleviates this bottleneck by discarding the least relevant tokens. Eviction introduces a tension, since a one-off decision to discard content may prove detrimental later. Instead, we focus on alternative choices that can lead to cache compression without evicting tokens. We achieve this by learning a selector that is able to produce, based on context, a per-layer cache configuration towards an overall compression budget. The selector operates along three axes: sharing one cache across layers (depth), caching at fewer bits (precision), or truncating the low-rank latent cache representations (rank). We call the resulting method KV-Kaizen, for the many small per-layer choices it compounds. We observe that these interventions taken independently and uniformly over all layers limit achievable compression because they degrade accuracy. Crucially, composing them locally and adaptively to the context can instead preserve accuracy while achieving large memory savings. At inference, the selector runs once, before pre-fill. In evaluations on instruction following and reasoning tasks, our selectors reach the Pareto frontier of accuracy against cache size, against learning-free and post-hoc baselines. On long-context tasks, KV-Kaizen improves on eviction and can be composed with it, reaching a 32x smaller decode-time cache on a 14B model while preserving accuracy. A 4x cache size reduction incurs no accuracy degradation from 7B parameters up, and a compressed model is more accurate than a smaller uncompressed one with the same cache size. Together, these findings support pre-training large models and compressing them only afterwards.
This work proposes Behavior-Preserving KV Cache Compression, a training-free framework that scores candidate evictions by estimating the compressed-cache logits induced by their removal and evaluating the resulting KL to the full-cache next-token distribution using pre-eviction forward statistics.
Doo Hwan Hwang, Junyoung Jang, Jun-Ho Na et al.· Proceedings of the 2026 Conf...· 1 citation
The key-value (KV) cache becomes a major memory bottleneck in long-context LLM inference, placing substantial pressure on memory capacity and bandwidth. To mitigate this bottleneck, vector quantization (VQ) has emerged as a promising approach for aggressive KV cache compression. However, existing VQ methods degrade sub...
Long-context inference with Large Language Models (LLMs) is bottlenecked by the linearly growing memory of the key-value (KV) cache. Existing compression methods reduce the cache through token eviction or approximation, but degrade sharply at aggressive compression budgets. We propose PatchKV, a training-free framework...
This study investigates the crucial deployment question of whether Cartridges and other CKVs can handle off-context queries, and proposes Cartridges++, simple modifications to cartridges that retain off-context abilities at small or negligible cost.
Sonia Laguna, João Monteiro, Marco Cuturi et al.· arXiv.org· 1 citation
TierKV is presented, a mobile LLM inference framework built on Predictive Multi-Tier Cache Optimization (PMCO), which improves prefill throughput by up to 17.6x over existing mobile LLM frameworks, reduces RAM-resident KV cache by 12.5-34%, thereby enabling substantially longer contexts under the same memory budget, wh...
Zhi-Hao Shu, Md Musfiqur Rahman Sanim, Jie Hu et al.· arXiv.org· 0 citations
P PuzzleKV is proposed, a training- and calibration-free method that treats each completed page as an independent compression unit that achieves more than 96% of Full KV performance across both evaluated models and all benchmark settings, with substantial gains over Global SVD on RULER and competitive performance on Lo...
Zi-Zhong J. Wang, Jie-Ying Wang, Zhao Zhang et al.· 0 citations