Compressing KV Cache for Long-Context LLM Inference with Inter-Layer Attention Similarity

• URL: http://arxiv.org/abs/2412.02252v2
• Date: Mon, 04 Aug 2025 02:17:56 GMT
• Title: Compressing KV Cache for Long-Context LLM Inference with Inter-Layer Attention Similarity
• Authors: Da Ma, Lu Chen, Situo Zhang, Yuxun Miao, Su Zhu, Zhi Chen, Hongshen Xu, Hanqi Li, Shuai Fan, Lei Pan, Kai Yu,
• Abstract summary: textscPoD allocates memory according to token importance.<n>textscPoD reduces KV cache memory usage by up to 35% without compromising performance.
• Score: 24.118503938098307
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: The rapid expansion of context window sizes in Large Language Models~(LLMs) has enabled them to tackle increasingly complex tasks involving lengthy documents. However, this progress comes at the cost of a substantial increase in memory usage during inference, primarily due to the linear growth of the key-value~(KV) cache. Existing KV cache compression methods often discard less relevant tokens, which can lead to significant performance degradation when critical information is lost. In this paper, we propose \textsc{PoD}~(Proximal tokens over Distant tokens), a novel KV cache compression framework that allocates memory according to token importance, retaining less important tokens in a more compact, shared form rather than discarding them entirely. Our approach is motivated by two key observations: (1) proximal tokens -- those at the beginning and end of the context -- are significantly more important for next-token prediction, and (2) attention scores for distant tokens are highly redundant across consecutive layers. Leveraging these insights, \textsc{PoD} preserves the full KV cache for proximal tokens, while for distant tokens, it shares key states across layers. Since attention scores are determined by both queries and keys, sharing key states enables multiple layers to reuse a single set of keys for distant tokens, substantially reducing KV cache memory without discarding essential context. We further introduce a lightweight post-training adaptation to enable the model to adjust to this new attention-sharing structure. Extensive experiments on both synthetic~(Needle in a Haystack) and real-world long-context benchmarks demonstrate that \textsc{PoD} reduces KV cache memory usage by up to 35\% without compromising performance. Our method is orthogonal to existing token-selection-based techniques and can be combined with them for further KV cache compression.

Related papers

• PACT: Pruning and Clustering-Based Token Reduction for Faster Visual Language Models [0.0]
  We introduce PACT, a method that reduces inference time and memory usage by pruning irrelevant tokens and merging visually redundant ones. Our approach uses a novel importance metric to identify unimportant tokens without relying on attention scores. We also propose a novel clustering algorithm, called Distance Bounded Density Peak Clustering, which efficiently clusters visual tokens.
  arXiv  Detail & Related papers  (2025-04-11T20:45:00Z)
• PromptDistill: Query-based Selective Token Retention in Intermediate Layers for Efficient Large Language Model Inference [23.828853446344663]
  We propose PromptDistill, a training-free method that improves inference efficiency while preserving generation quality. PromptDistill identifies and retains the most informative tokens by leveraging attention interactions in early layers, preserving their hidden states while reducing the computational burden in later layers.
  arXiv  Detail & Related papers  (2025-03-30T01:47:23Z)
• TopV: Compatible Token Pruning with Inference Time Optimization for Fast and Low-Memory Multimodal Vision Language Model [56.43860351559185]
  We introduce textbfTopV, a compatible textbfTOken textbfPruning with inference Time Optimization for fast and low-memory textbfVLM. Our framework incorporates a visual-aware cost function to measure the importance of each source visual token, enabling effective pruning of low-importance tokens.
  arXiv  Detail & Related papers  (2025-03-24T01:47:26Z)
• AttentionPredictor: Temporal Pattern Matters for Efficient LLM Inference [51.1972443343829]
  We propose AttentionPredictor, which is the first learning-based critical token identification approach.<n> AttentionPredictor accurately predicts the attention score while consuming negligible memory.<n>We also propose a cross-token critical cache prefetching framework that hides the token time overhead to accelerate the decoding stage.
  arXiv  Detail & Related papers  (2025-02-06T13:41:46Z)
• ChunkKV: Semantic-Preserving KV Cache Compression for Efficient Long-Context LLM Inference [28.96662510838151]
  We introduce ChunkKV, which reimagines KV cache compression by treating semantic chunks as basic compression units.<n>This approach preserves complete linguistic structures and contextual integrity, ensuring that essential meaning is retained even under aggressive compression.<n>ChunkKV outperforms state-of-the-art methods by up to 8.7% in precision while maintaining the same compression ratio.
  arXiv  Detail & Related papers  (2025-02-01T03:49:47Z)
• More Tokens, Lower Precision: Towards the Optimal Token-Precision Trade-off in KV Cache Compression [71.42818367729573]
  In large language models (LLMs), the memory usage of KV cache has become a critical bottleneck during inference.<n>The mainstream KV compression methods, including KV pruning and KV quantization, primarily focus on either token or precision dimension separately.<n>In this paper, we comprehensively investigate the token-precision trade-off in KV cache compression.
  arXiv  Detail & Related papers  (2024-12-17T09:20:31Z)
• SepLLM: Accelerate Large Language Models by Compressing One Segment into One Separator [65.62084602011596]
  Large Language Models (LLMs) have exhibited exceptional performance across a spectrum of natural language processing tasks. We have identified a key pattern: certain seemingly meaningless separator tokens (i.e., punctuations) contribute disproportionately to attention scores compared to semantically meaningful tokens. We introduce SepLLM, a plug-and-play framework that accelerates inference by compressing these segments and eliminating redundant tokens.
  arXiv  Detail & Related papers  (2024-12-16T18:58:57Z)
• [CLS] Token Tells Everything Needed for Training-free Efficient MLLMs [66.5266435598799]
  Multi-language Large Language Models (MLLMs) have recently demonstrated strong performance across a wide range of vision tasks. However, their efficient deployment remains a substantial challenge due to high computational costs and memory requirements. We introduce a simple yet effective method for train-free visual compression, called VTC- compression.
  arXiv  Detail & Related papers  (2024-12-08T05:29:39Z)
• ClusterKV: Manipulating LLM KV Cache in Semantic Space for Recallable Compression [10.003118268356017]
  Long context poses significant challenges for inference efficiency.<n>We introduce ClusterKV, which recalls tokens at the granularity of semantic clusters.<n>Experiment results show that ClusterKV attains negligible accuracy loss across various tasks with 32k context lengths.
  arXiv  Detail & Related papers  (2024-12-04T10:58:27Z)
• Recycled Attention: Efficient inference for long-context language models [54.00118604124301]
  We propose Recycled Attention, an inference-time method which alternates between full context attention and attention over a subset of input tokens. When performing partial attention, we recycle the attention pattern of a previous token that has performed full attention and attend only to the top K most attended tokens. Compared to previously proposed inference-time acceleration method which attends only to local context or tokens with high accumulative attention scores, our approach flexibly chooses tokens that are relevant to the current decoding step.
  arXiv  Detail & Related papers  (2024-11-08T18:57:07Z)
• Inference Optimal VLMs Need Fewer Visual Tokens and More Parameters [54.01228554126122]
  Vision Language Models (VLMs) have demonstrated strong capabilities across various visual understanding and reasoning tasks. To reduce inference costs, one can either downsize the Large Language Models (LLMs) or reduce the number of input tokens needed to represent the image. We take the first steps toward designing token compression algorithms tailored for high-compression settings.
  arXiv  Detail & Related papers  (2024-11-05T18:54:21Z)
• TidalDecode: Fast and Accurate LLM Decoding with Position Persistent Sparse Attention [7.4088392854630625]
  Large language models (LLMs) have driven significant advancements across diverse NLP tasks. This paper introduces TidalDecode, a system for fast and accurate LLM decoding through position persistent sparse attention.
  arXiv  Detail & Related papers  (2024-10-07T14:30:27Z)
• Efficient Inference of Vision Instruction-Following Models with Elastic Cache [76.44955111634545]
  We introduce Elastic Cache, a novel strategy for efficient deployment of instruction-following large vision-language models. We propose an importance-driven cache merging strategy to prune redundancy caches. For instruction encoding, we utilize the frequency to evaluate the importance of caches. Results on a range of LVLMs demonstrate that Elastic Cache not only boosts efficiency but also notably outperforms existing pruning methods in language generation.
  arXiv  Detail & Related papers  (2024-07-25T15:29:05Z)
• Training-Free Exponential Context Extension via Cascading KV Cache [49.608367376911694]
  We introduce a novel mechanism that leverages cascading sub-cache buffers to selectively retain the most relevant tokens. Our method reduces prefill stage latency by a factor of 6.8 when compared to flash attention on 1M tokens.
  arXiv  Detail & Related papers  (2024-06-24T03:59:17Z)
• A Simple and Effective $L_2$ Norm-Based Strategy for KV Cache Compression [13.981807478365452]
  Existing approaches to reduce the Key-Value cache size involve either fine-tuning the model to learn a compression strategy or leveraging attention scores to reduce the sequence length. We find a clear correlation between the $L$ and the attention scores over cached KV pairs, where a low $L$ of a key embedding leads to a high attention score during decoding. Our experimental results show that this simple strategy can reduce the KV cache size by 50% on language modelling and needle-in-a-haystack tasks and 90% on passkey retrieval tasks without losing accuracy.
  arXiv  Detail & Related papers  (2024-06-17T11:35:16Z)
• A Training-free Sub-quadratic Cost Transformer Model Serving Framework With Hierarchically Pruned Attention [43.211427581302715]
  We propose Hierarchically Pruned Attention (HiP) to increase context length in large language models. HiP reduces the time complexity of the attention mechanism to $O(T log T)$ and the space complexity to $O(T)$, where $T$ is the sequence length. We show that HiP significantly reduces both prefill and decoding latencies, as well as memory usage, while maintaining high-quality generation with minimal degradation.
  arXiv  Detail & Related papers  (2024-06-14T08:32:45Z)
• LeMeViT: Efficient Vision Transformer with Learnable Meta Tokens for Remote Sensing Image Interpretation [37.72775203647514]
  This paper proposes to use learnable meta tokens to formulate sparse tokens, which effectively learn key information and improve inference speed. By employing Dual Cross-Attention (DCA) in the early stages with dense visual tokens, we obtain the hierarchical architecture LeMeViT with various sizes. Experimental results in classification and dense prediction tasks show that LeMeViT has a significant $1.7 times$ speedup, fewer parameters, and competitive performance compared to the baseline models.
  arXiv  Detail & Related papers  (2024-05-16T03:26:06Z)
• Keyformer: KV Cache Reduction through Key Tokens Selection for Efficient Generative Inference [2.8241099113277666]
  "Keyformer" is an innovative inference-time approach to mitigate the challenges associated with KV cache size and memory bandwidth utilization. We evaluate Keyformer's performance across three foundational models: GPT-J, Cerebras-GPT, and MPT.
  arXiv  Detail & Related papers  (2024-03-14T02:42:42Z)
• Get More with LESS: Synthesizing Recurrence with KV Cache Compression for Efficient LLM Inference [78.65321721142624]
  We focus on a memory bottleneck imposed by the key-value ( KV) cache. Existing KV cache methods approach this problem by pruning or evicting large swaths of relatively less important KV pairs. We propose LESS, a simple integration of a constant sized cache with eviction-based cache methods.
  arXiv  Detail & Related papers  (2024-02-14T18:54:56Z)
• 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)
• Efficient Streaming Language Models with Attention Sinks [72.20260088848987]
  StreamingLLM is an efficient framework that enables Large Language Models to generalize to infinite sequence lengths without any fine-tuning. We show that StreamingLLM can enable Llama-2, MPT, Falcon, and Pythia to perform stable and efficient language modeling with up to 4 million tokens and more.
  arXiv  Detail & Related papers  (2023-09-29T17:59:56Z)

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.