Lethe: Layer- and Time-Adaptive KV Cache Pruning for Reasoning-Intensive LLM Serving

• URL: http://arxiv.org/abs/2511.06029v2
• Date: Thu, 13 Nov 2025 01:20:52 GMT
• Title: Lethe: Layer- and Time-Adaptive KV Cache Pruning for Reasoning-Intensive LLM Serving
• Authors: Hui Zeng, Daming Zhao, Pengfei Yang, WenXuan Hou, Tianyang Zheng, Hui Li, Weiye Ji, Jidong Zhai,
• Abstract summary: Generative reasoning with large language models (LLMs) often involves long decoding sequences.<n>We propose Lethe, a dynamic KV cache management framework that introduces adaptivity along both the spatial and temporal dimensions of decoding.<n>Lethe achieves a favorable balance between efficiency and generation quality across diverse models and tasks, increases throughput by up to 2.56x.
• Score: 11.750209684686707
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Generative reasoning with large language models (LLMs) often involves long decoding sequences, leading to substantial memory and latency overheads from accumulating key-value (KV) caches. While existing KV compression methods primarily focus on reducing prefill memory from long input sequences, they fall short in addressing the dynamic and layer-sensitive nature of long-form generation, which is central to reasoning tasks. We propose Lethe, a dynamic KV cache management framework that introduces adaptivity along both the spatial and temporal dimensions of decoding. Along the spatial dimension, Lethe performs layerwise sparsity-aware allocation, assigning token pruning budgets to each transformer layer based on estimated attention redundancy. Along the temporal dimension, Lethe conducts multi-round token pruning during generation, driven by a Recency-Aware Selective Retention} (RASR) mechanism. RASR extends traditional recency-based heuristics by also considering token relevance derived from evolving attention patterns, enabling informed decisions about which tokens to retain or evict. Empirical results demonstrate that Lethe achieves a favorable balance between efficiency and generation quality across diverse models and tasks, increases throughput by up to 2.56x.

Related papers

• Training-free Context-adaptive Attention for Efficient Long Context Modeling [57.703159205740185]
  Training-free Context-adaptive Attention (TCA-Attention) is a training-free sparse attention mechanism that selectively attends to only the informative tokens for efficient long-context inference.<n>TCA-Attention achieves a 2.8$times$ speedup and reduces KV cache by 61% at 128K context length while maintaining performance comparable to full attention.
  arXiv  Detail & Related papers  (2025-12-10T01:54:57Z)
• Cache What Lasts: Token Retention for Memory-Bounded KV Cache in LLMs [26.951325519894525]
  We propose a novel approach that learns each token's intrinsic importance at creation time via a lightweight retention gate.<n>We show that it consistently outperforms strong eviction and learnable retrieval baselines, especially in low-memory regimes.<n>It even surpasses full-cache models in some settings, showing that selective retention can serve as a form of regularization.
  arXiv  Detail & Related papers  (2025-12-03T00:20:35Z)
• KV-Latent: Dimensional-level KV Cache Reduction with Frequency-aware Rotary Positional Embedding [72.12756830560217]
  Large language models (LLMs) based on Transformer Decoders have become the preferred choice for conversational generative AI.<n>Despite the overall superiority of the Decoder architecture, the gradually increasing Key-Value cache during inference has emerged as a primary efficiency bottleneck.<n>By down-sampling the Key-Value vector dimensions into a latent space, we can significantly reduce the KV Cache footprint and improve inference speed.
  arXiv  Detail & Related papers  (2025-07-15T12:52:12Z)
• Beyond Homogeneous Attention: Memory-Efficient LLMs via Fourier-Approximated KV Cache [67.47789629197857]
  We propose a training-free framework that exploits the heterogeneous roles of transformer head dimensions.<n>By projecting the long-context-insensitive dimensions onto Fourier bases, FourierAttention approximates their temporal evolution with fixed-length spectral coefficients.<n>We show that FourierAttention achieves the best long-context accuracy on LongBench and Needle-In-A-Haystack.
  arXiv  Detail & Related papers  (2025-06-13T15:35:54Z)
• Efficient Pretraining Length Scaling [21.4715211093876]
  We present the Parallel Hidden Decoding Transformer (textitPHD-Transformer), a novel framework that enables efficient length scaling during pre-training.<n>textitPHD-Transformer achieves this through an innovative KV cache management strategy that distinguishes between original tokens and hidden decoding tokens.
  arXiv  Detail & Related papers  (2025-04-21T09:41:26Z)
• Dialogue Without Limits: Constant-Sized KV Caches for Extended Responses in LLMs [6.222287867011644]
  We propose MorphKV, an inference-time technique that maintains a constant-sized KV cache while preserving accuracy.<n>Unlike retention or lossy compression, MorphKV iteratively refines the KV cache via lightweight updates guided by attention patterns of recent tokens.<n>Our studies show 52.9$%$ memory savings and 18.2$%$ higher accuracy on average compared to state-of-the-art prior works.
  arXiv  Detail & Related papers  (2025-03-02T18:12:50Z)
• ZipVL: Efficient Large Vision-Language Models with Dynamic Token Sparsification [29.163757099307553]
  The efficiency of large vision-language models (LVLMs) is constrained by the computational bottleneck of the attention mechanism during the prefill phase.<n>We present ZipVL, an efficient inference framework designed for LVLMs through a dynamic ratio allocation strategy of important tokens.
  arXiv  Detail & Related papers  (2024-10-11T07:24:21Z)
• ThinK: Thinner Key Cache by Query-Driven Pruning [63.13363917871414]
  Large Language Models (LLMs) have revolutionized the field of natural language processing, achieving unprecedented performance across a variety of applications.<n>This paper focuses on the long-context scenario, addressing the inefficiencies in KV cache memory consumption during inference.<n>We propose ThinK, a novel query-dependent KV cache pruning method designed to minimize attention weight loss while selectively pruning the least significant channels.
  arXiv  Detail & Related papers  (2024-07-30T17:59:08Z)
• D2O: Dynamic Discriminative Operations for Efficient Long-Context Inference of Large Language Models [28.244034916473804]
  Generative inference in Large Language Models (LLMs) is impeded by the growing memory demands of Key-Value (KV) cache.<n>Traditional KV cache eviction strategies discard less critical KV pairs based on attention scores, leading to issues such as context loss or hallucinations.<n>We introduce Dynamic Discriminative Operations (D2O), a KV cache compression method that optimize KV cache size dynamically and discriminatively at two levels without fine-tuning.
  arXiv  Detail & Related papers  (2024-06-18T20:01:51Z)
• SubGen: Token Generation in Sublinear Time and Memory [48.35076900702408]
  Large language models (LLMs) have extensive memory requirements for token generation. In this work, we focus on developing an efficient compression technique for the KV cache. We have devised a novel caching method with sublinear complexity, employing online clustering on key tokens and online $ell$ sampling on values. Not only does this algorithm ensure a sublinear memory footprint and sublinear time complexity, but we also establish a tight error bound for our approach.
  arXiv  Detail & Related papers  (2024-02-08T22:17:40Z)
• Model Tells You What to Discard: Adaptive KV Cache Compression for LLMs [82.08922896531618]
  We introduce adaptive KV cache compression, a plug-and-play method that reduces the memory footprint of generative inference for Large Language Models (LLMs) We conduct targeted profiling to discern the intrinsic structure of attention modules. Based on the recognized structure, we then construct the KV cache in an adaptive manner: evicting long-range contexts on attention heads emphasizing local contexts, discarding non-special tokens on attention heads centered on special tokens, and only employing the standard KV cache for attention heads that broadly attend to all tokens.
  arXiv  Detail & Related papers  (2023-10-03T05:17:08Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.