paper-with-me

홈 › Papers

CASK: Core-Aware Selective KV Compression for Reasoning Traces

2026-04-13 · Buseong Kim, Heejun Gwon arxiv

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.

📄 PDF Abstract BibTeX arXiv:2604.10900

Code (0)

등록된 구현이 없습니다.

Similar Papers 제목 키워드 기반

LogicAsker: Evaluating and Improving the Logical Reasoning Ability of Large Language Models

2024-01-01 · Yuxuan Wan, Wenxuan Wang, Yiliu Yang, Youliang Yuan 외

We introduce LogicAsker, a novel approach for evaluating and enhancing the logical reasoning capabilities of large language models (LLMs) such as ChatGPT and GPT-4. Despite LLMs' prowess in tasks like writing assistance,…

Code GenerationIn-Context LearningLogical ReasoningMachine Translation

STAC: Selective Spatiotemporal Aggregation and Compression for Video Reasoning Segmentation

2026-07-03 · Syed Ariff Syed Hesham, Yun Liu, Guolei Sun, Jing Yang 외 arxiv

Video reasoning segmentation demands pixel-accurate object tracking across hundreds of frames under complex natural language queries, producing dense spatiotemporal tokens whose quadratic self-attention cost makes long-v…

Natural Language QueriesObject Tracking

When Compression Helps and When It Hurts: Condition-Aware Analysis of Chain-of-Thought Distillation

2026-06-19 · Siyang Lyu, Zhijing Sun, Xinghao Chen, Tong Liu 외 arxiv

Chain-of-Thought (CoT) distillation transfers multi-step reasoning from large reasoning models to smaller students, but verbose teacher traces inflate both training and inference cost. Existing CoT compression methods fa…

Selective Latent Thinking: Adaptive Compression of LLM Reasoning Chains

2026-05-25 · Hui Xie, Jie Liu, Ziyue Qiao, Joaquin Vanschore arxiv

Explicit chain-of-thought (CoT) reasoning substantially improves the reasoning ability of large language models (LLMs), but incurs high inference cost due to lengthy autoregressive traces. Existing latent reasoning metho…

Reinforcement LearningMathematical Reasoning

Can Pruning Improve Reasoning? Revisiting Long-CoT Compression with Capability in Mind for Better Reasoning

2025-05-20 · Shangziqi Zhao, Jiahao Yuan, Guisong Yang, Usman Naseem

Long chain-of-thought (Long-CoT) reasoning improves accuracy in LLMs, yet its verbose, self-reflective style often hinders effective distillation into small language models (SLMs). We revisit Long-CoT compression through…

Large Language ModelMathematical Reasoning