CASK: Core-Aware Selective KV Compression for Reasoning Traces

• URL: http://arxiv.org/abs/2604.10900v1
• Date: Mon, 13 Apr 2026 02:03:16 GMT
• Title: CASK: Core-Aware Selective KV Compression for Reasoning Traces
• Abstract summary: CASK partitions the decode-time reasoning trace into a protected core that anchors answer formation and intermediate state, and mergeable scratch with high redundancy.<n>On the H100 reasoning gate, CASK shows higher full- KV continuation fidelity than TriAttention at matched budgets on both AIME24 and AIME25, with recurring cask@384 > triattention@512 crossings.
• Score: 0.0
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: In large language models performing long-form reasoning, the KV cache grows rapidly with decode length, creating bottlenecks in memory and inference stability. Existing reasoning-oriented KV compression has mostly followed an eviction-centered view: estimate token importance more accurately, then discard lower-ranked entries. Our analysis suggests that scorer refinement alone often fails to substantially reorganize the actual keep-set and may therefore not be the main lever for preserving reasoning behavior. We instead frame reasoning KV compression as a behavior-preserving structured consolidation problem. CASK partitions the decode-time reasoning trace into a protected core that anchors answer formation and intermediate state, and mergeable scratch with high redundancy. The core is preserved, while selective consolidation is applied only to the scratch. To address prompt-heavy regimes where the prefix can exhaust the budget before decode-stage compression becomes active, CASK further uses a two-stage design: prefix eviction followed by decode-stage consolidation. On the H100 reasoning gate, CASK shows higher full-KV continuation fidelity than TriAttention at matched budgets on both AIME24 and AIME25, with recurring cask@384 > triattention@512 crossings. In prompt-heavy replay, multi_news and vcsum act as decode-active witnesses, while qmsum and gov_report expose the prefix_budget_exhausted boundary. The overall evidence supports a simple conclusion: effective reasoning KV compression depends less on more elaborate scorer engineering than on combining core preservation with selective scratch consolidation to lower the usable budget frontier.

Related papers

• BeaconKV: Key-Value Cache Compression Guided by Beacon Queries for Efficient Large Reasoning Model Inference [20.58039191111583]
  Large Reasoning Models (LRMs) achieve superior problem-solving through extended Chain-of-Thought (CoT) generation.<n>The resulting key-value ( KV) cache grows linearly with sequence length and creates severe memory bottlenecks.<n>We propose BeaconKV, a training-free KV cache compression method that maintains beacon queries, compact representatives for each global query cluster.
  arXiv  Detail & Related papers  (2026-09-04T10:23:16Z)
• AnchorKV: Safety-Aware KV Cache Compression via Soft Penalty with a Refusal Anchor [20.7510970936151]
  Large language models (LLMs) outperform earlier architectures on generative inference and long-context tasks.<n>Their large size introduces significant challenges in memory usage, energy cost, and on-device deployment.<n>We propose AnchorKV, a drop-in modification to KV cache compression that biases token retention scores away from directions in key space associated with harmful prompts.
  arXiv  Detail & Related papers  (2026-06-16T12:43:18Z)
• ReasonAlloc: Hierarchical Decoding-Time KV Cache Budget Allocation for Reasoning Models [51.947287846779545]
  ReasonAlloc recasts decoding-time KV compression as a hierarchical budget allocation problem.<n>An offline layer-wise preallocation strategy captures an architecture-driven demand pattern.<n>An online head-wise strategy reallocates resources during decoding to information-rich heads based on real-time utility.
  arXiv  Detail & Related papers  (2026-06-09T17:44:23Z)
• GRKV: Global Regression for Training-Free KV Cache Compression in Long-Context LLMs [97.36238579001544]
  Large language models (LLMs) with extended context lengths rely on the key-value ( KV) cache to support attention over prior tokens.<n>Maintaining the KV cache incurs substantial memory overhead, motivating KV-cache compression methods.<n>We propose GRKV, a training-free KV-cache merging method that directly minimizes the discrepancy between compressed-cache and full-cache attention outputs.
  arXiv  Detail & Related papers  (2026-05-29T10:16:30Z)
• Make Each Token Count: Towards Improving Long-Context Performance with KV Cache Eviction [65.710271475739]
  We introduce a global retention-based KV eviction method that learns each token's future utility under a unified memory budget.<n>We show that our method substantially reduces KV memory while matching or surpassing full-cache inference.<n>These results suggest that learned, globally calibrated KV eviction is not only a compression technique, but also a mechanism for improving long-context reasoning.
  arXiv  Detail & Related papers  (2026-05-10T16:47:50Z)
• Kwai Summary Attention Technical Report [69.40814939510126]
  Long-context ability, has become one of the most important direction of next-generation Large Language Models.<n>Standard softmax attention exhibits quadratic time complexity with respect to sequence length.<n>We propose Kwai Summary Attention (KSA), a novel attention mechanism that reduces sequence modeling cost by compressing historical contexts.
  arXiv  Detail & Related papers  (2026-04-27T12:59:53Z)
• 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)
• ForesightKV: Optimizing KV Cache Eviction for Reasoning Models by Learning Long-Term Contribution [84.41751286055909]
  We develop a training-based KV cache eviction framework that learns to predict which KV pairs to evict during longtext generations.<n>We formulate cache eviction as a Markov Decision Process and apply the GRPO algorithm to mitigate the significant language modeling loss increase on low-entropy tokens.
  arXiv  Detail & Related papers  (2026-02-03T07:16:51Z)
• Crystal-KV: Efficient KV Cache Management for Chain-of-Thought LLMs via Answer-First Principle [22.39292106468653]
  Chain-of-Thought (CoT) reasoning significantly improves accuracy on complex tasks.<n>CoT emphasizes the final answer, rendering conventional KV compression strategies ineffective.<n>We present Crystal-KV, an efficient KV cache management framework tailored for CoT reasoning.
  arXiv  Detail & Related papers  (2026-01-05T07:42:58Z)
• SkipKV: Selective Skipping of KV Generation and Storage for Efficient Inference with Large Reasoning Models [25.509962883211]
  Large reasoning models (LRMs) often cost significant key-value (KV) cache overhead, due to their linear growth with the chain-of-thought (CoT) reasoning process.<n>We present textbfSkipKV, a KV compression method for selective textiteviction and textitgeneration operating at a coarse-grained sentence-level sequence removal.
  arXiv  Detail & Related papers  (2025-12-08T19:32:06Z)
• G-KV: Decoding-Time KV Cache Eviction with Global Attention [57.47409249054187]
  Large language models (LLMs) excel in complex tasks but encounter significant computational and memory challenges due to long sequence lengths.<n> KV cache compression has emerged as an effective approach to greatly enhance the efficiency of reasoning.<n>We propose G-KV, a KV cache eviction method that employs a global scoring mechanism, combining local and historical attention scores to more accurately assess token importance.
  arXiv  Detail & Related papers  (2025-11-29T14:21:33Z)
• Which Heads Matter for Reasoning? RL-Guided KV Cache Compression [15.865990296257413]
  Reasoning large language models exhibit complex reasoning behaviors through the extended chain-of-thought generation.<n>Existing KV cache compression methods underperform on reasoning models.<n>We propose RLKV, a novel reasoning-critical head identification framework.
  arXiv  Detail & Related papers  (2025-10-09T17:50:00Z)
• OjaKV: Context-Aware Online Low-Rank KV Cache Compression with Oja's Rule [54.37983890753086]
  We introduce OjaKV, a framework that integrates a strategic hybrid storage policy with online subspace adaptation.<n>OjaKV preserves crucial first and most recent tokens in full-rank, maintaining high-fidelity anchors for attention.<n>It applies low-rank compression by incrementally adapting the projection basis using Oja's algorithm for online principal component analysis.
  arXiv  Detail & Related papers  (2025-09-25T21:42:27Z)
• LazyEviction: Lagged KV Eviction with Attention Pattern Observation for Efficient Long Reasoning [21.761205124793175]
  extended reasoning sequences introduce significant GPU memory overhead due to increased key-value (KV) cache.<n>Existing KV cache compression methods mitigate memory bottlenecks but struggle in long reasoning tasks.<n>We propose LazyEviction, an observation window-based lagged eviction framework retaining latent recurring tokens by prioritized eviction based on tokens' recurrence patterns.
  arXiv  Detail & Related papers  (2025-06-19T02:25:04Z)
• 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)
• FlowKV: Enhancing Multi-Turn Conversational Coherence in LLMs via Isolated Key-Value Cache Management [48.904743679691414]
  FlowKV is a novel multi-turn isolation mechanism for KV Cache management.<n>It preserves the accumulated compressed KV cache from past turns.<n>It prevents the re-compression of older context and thereby mitigating catastrophic forgetting.
  arXiv  Detail & Related papers  (2025-05-21T10:20:46Z)

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.