KARA: Efficient Reasoning LLM Serving via Sliding-Window KV Cache Compression

• URL: http://arxiv.org/abs/2607.01237v2
• Date: Fri, 03 Jul 2026 02:55:22 GMT
• Title: KARA: Efficient Reasoning LLM Serving via Sliding-Window KV Cache Compression
• Abstract summary: Reasoning language models generate long chain-of-thought (CoT)<n>CoT accumulates a massive KV cache during the decoding phase.<n> KV cache compression has emerged as a promising technique for reducing memory overhead.
• Score: 3.9838043625801967
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Reasoning language models often generate long chain-of-thought (CoT), which accumulates a massive KV cache during the decoding phase and incurs high decoding latency and limited throughput. To address these issues, KV cache compression has emerged as a promising technique for reducing memory overhead by selectively removing unimportant KV pairs while preserving useful ones for subsequent decoding. Nevertheless, we identify two key limitations in existing KV cache compression methods: 1) their threshold-triggered compression policy may provide limited throughput improvement or even reduce throughput, and may fully eliminate KV pairs from certain blocks of the sequence, potentially worsening information loss. 2) they typically retain either isolated KV pairs or fixed-size chunks with rigid boundaries, failing to preserve important flexible-sized chunks at arbitrary token positions. To overcome these limitations, we propose Kara, a sliding-window KV cache compression method that performs decoding-time compression by operating only on the recently generated context. Kara leverages bidirectional attention to score and select informative KV pairs in the window. To enable flexible preservation of important semantic information, we design a Token2Chunk module to expand a subset of selected KV pairs into chunks. Furthermore, we adapt Kara to PagedAttention and develop KvLLM, an inference framework built upon vLLM, which reduces KV cache memory usage and effectively improves output throughput. Extensive experiments demonstrate consistent performance improvements of proposed Kara and KvLLM.

Related papers

• DeltaKV: Residual-Based KV Cache Compression via Long-Range Similarity [50.52392445266824]
  We propose a residual-based KV cache compression framework motivated by long-range inter-token similarity and highly shared latent components in KV representations.<n>Instead of discarding tokens, DeltaKV encodes semantic residuals relative to retrieved historical references, preserving fidelity while substantially reducing storage.<n>Experiments show that DeltaKV reduces KV cache memory to 29% of the original while maintaining near-lossless accuracy on LongBench, SCBench, and AIME.
  arXiv  Detail & Related papers  (2026-02-08T15:14:36Z)
• KV-CAR: KV Cache Compression using Autoencoders and KV Reuse in Large Language Models [3.5171501100868876]
  The KV cache grows with sequence length and embedding dimension, often exceeding the memory footprint of the model itself.<n>We present KV CAR, a unified and agnostic architecture framework that significantly reduces KV cache storage while maintaining model fidelity.<n> Evaluations on GPT 2 and TinyLLaMA models across Wikitext, C4, PIQA, and Winogrande datasets demonstrate that KV CAR achieves up to 47.85 percent KV cache memory reduction.
  arXiv  Detail & Related papers  (2025-12-07T08:40:52Z)
• KVReviver: Reversible KV Cache Compression with Sketch-Based Token Reconstruction [20.53279247581787]
  We propose KVReviver, a reversible KV cache compression method based on the sketch algorithm.<n>In 2k-length contexts, it requires only 10% of KV Cache budget while maintaining identical end-to-end inference accuracy.<n>For 32k-length contexts, it achieves equivalent or comparable accuracy 2% accuracy loss.
  arXiv  Detail & Related papers  (2025-12-01T03:59:20Z)
• SWAN: Sparse Winnowed Attention for Reduced Inference Memory via Decompression-Free KV-Cache Compression [7.603859408568262]
  Large Language Models (LLMs) face a significant bottleneck during autoregressive inference due to the massive memory footprint of the Key-Value (KV) cache.<n>We introduce SWAN, a novel, fine-tuning-free framework that eliminates this overhead.<n>Our method uses an offline matrix to rotate and prune the KV-cache, which is then used directly in the attention computation without any reconstruction.
  arXiv  Detail & Related papers  (2025-11-24T09:41:24Z)
• 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)
• ReCalKV: Low-Rank KV Cache Compression via Head Reordering and Offline Calibration [69.57122277845293]
  We propose ReCalKV, a post-training low-rank KV cache compression approach with tailored strategies for Keys and Values.<n>For Keys, we propose Similarity aware Recontext (HSR), which clusters structurally similar heads into groups, enabling more accurate low-rank approximation.<n>For Values, we propose Offline Head-wise Value (OVC), which efficiently calibrates the value projection matrix using calibration data without training.
  arXiv  Detail & Related papers  (2025-05-30T08:49:27Z)
• KeepKV: Eliminating Output Perturbation in KV Cache Compression for Efficient LLMs Inference [16.53643930310808]
  KeepKV is a novel adaptive KV cache merging method designed to eliminate output perturbation while preserving performance under strict memory constraints.<n>We show that KeepKV substantially reduces memory usage, enhances inference throughput by more than 2x and keeps superior generation quality even with 10% KV cache budgets.
  arXiv  Detail & Related papers  (2025-04-14T06:58:00Z)
• DBudgetKV: Dynamic Budget in KV Cache Compression for Ensuring Optimal Performance [125.81664663201282]
  We introduce a new KV cache compression method dubbed DBudgetKV.<n>It features an attention-based metric to signal when the remaining KV cache is unlikely to match the full-cache performance.<n>Our method achieves lossless KV pruning effectively and robustly, exceeding 25% compression ratio on average.
  arXiv  Detail & Related papers  (2025-02-24T06:33:39Z)
• ChunkKV: Semantic-Preserving KV Cache Compression for Efficient Long-Context LLM Inference [61.412894960600205]
  Large Language Models (LLMs) require significant GPU memory when processing long texts.<n>ChunkKV reimagines KV cache compression by treating semantic chunks as basic compression units.<n>Result: ChunkKV outperforms state-of-the-art methods by up to 8.7% in precision.
  arXiv  Detail & Related papers  (2025-02-01T03:49:47Z)
• KVSharer: Efficient Inference via Layer-Wise Dissimilar KV Cache Sharing [58.29726147780976]
  We propose a plug-and-play method called textit KVSharer, which shares the KV cache between layers to achieve layer-wise compression. Experiments show that textit KVSharer can reduce KV cache computation by 30%, thereby lowering memory consumption. We verify that textit KVSharer is compatible with existing intra-layer KV cache compression methods, and combining both can further save memory.
  arXiv  Detail & Related papers  (2024-10-24T08:06:41Z)
• LoRC: Low-Rank Compression for LLMs KV Cache with a Progressive Compression Strategy [59.1298692559785]
  Key-Value ( KV) cache is crucial component in serving transformer-based autoregressive large language models (LLMs) Existing approaches to mitigate this issue include: (1) efficient attention variants integrated in upcycling stages; (2) KV cache compression at test time; and (3) KV cache compression at test time. We propose a low-rank approximation of KV weight matrices, allowing plug-in integration with existing transformer-based LLMs without model retraining. Our method is designed to function without model tuning in upcycling stages or task-specific profiling in test stages.
  arXiv  Detail & Related papers  (2024-10-04T03:10:53Z)
• CSKV: Training-Efficient Channel Shrinking for KV Cache in Long-Context Scenarios [13.144156413032896]
  We introduce CSKV, a training-efficient Channel Shrinking technique for KV cache compression. We show that CSKV can reduce the memory overhead of the KV cache by 80% while maintaining the model's long-context capability. Our method can be seamlessly combined with quantization to further reduce the memory overhead, achieving a compression ratio of up to 95%.
  arXiv  Detail & Related papers  (2024-09-16T17:36:50Z)

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.