Cover Image

Semantic Integrity Matters: Benchmarking and Preserving High-Density Reasoning in KV Cache Compression

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Liu, Xiang, Tang, Zhenheng, Chen, Hong, Dong, Peijie, Li, Zeyu, Zhou, Xiuze, Li, Bo, Hu, Xuming, Chu, Xiaowen
Format:	Preprint
Published:	2025
Subjects:	Computation and Language Artificial Intelligence
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

ChunkKV: Semantic-Preserving KV Cache Compression for Efficient Long-Context LLM Inference
by: Liu, Xiang, et al.
Published: (2025)

Can Compressed LLMs Truly Act? An Empirical Evaluation of Agentic Capabilities in LLM Compression
by: Dong, Peijie, et al.
Published: (2025)

The Lottery LLM Hypothesis, Rethinking What Abilities Should LLM Compression Preserve?
by: Tang, Zhenheng, et al.
Published: (2025)

LongGenBench: Long-context Generation Benchmark
by: Liu, Xiang, et al.
Published: (2024)

FlowKV: Enhancing Multi-Turn Conversational Coherence in LLMs via Isolated Key-Value Cache Management
by: Liu, Xiang, et al.
Published: (2025)

NestedKV: Nested Memory Routing for Long-Context KV Cache Compression
by: Chen, Hong, et al.
Published: (2026)

SONIC: Segmented Optimized Nexus for Information Compression in Key-Value Caching
by: Chen, Hong, et al.
Published: (2026)

Should We Really Edit Language Models? On the Evaluation of Edited Language Models
by: Li, Qi, et al.
Published: (2024)

AnTKV: Anchor Token-Aware Sub-Bit Vector Quantization for KV Cache in Large Language Models
by: Li, Zeyu, et al.
Published: (2025)

VMRNN: Integrating Vision Mamba and LSTM for Efficient and Accurate Spatiotemporal Forecasting
by: Tang, Yujin, et al.
Published: (2024)

Decoding Knowledge Attribution in Mixture-of-Experts: A Framework of Basic-Refinement Collaboration and Efficiency Analysis
by: Li, Junzhuo, et al.
Published: (2025)

Pruner-Zero: Evolving Symbolic Pruning Metric from scratch for Large Language Models
by: Dong, Peijie, et al.
Published: (2024)

LPZero: Language Model Zero-cost Proxy Search from Zero
by: Dong, Peijie, et al.
Published: (2024)

Position: LLM Inference Should Be Evaluated as Energy-to-Token Production
by: Liu, Xiang, et al.
Published: (2026)

R-KV: Redundancy-aware KV Cache Compression for Reasoning Models
by: Cai, Zefan, et al.
Published: (2025)

Which Heads Matter for Reasoning? RL-Guided KV Cache Compression
by: Du, Wenjie, et al.
Published: (2025)

Dynamic Expert Specialization: Towards Catastrophic Forgetting-Free Multi-Domain MoE Adaptation
by: Li, Junzhuo, et al.
Published: (2025)

Mediator: Memory-efficient LLM Merging with Less Parameter Conflicts and Uncertainty Based Routing
by: Lai, Kunfeng, et al.
Published: (2025)

DiffAdapt: Difficulty-Adaptive Reasoning for Token-Efficient LLM Inference
by: Liu, Xiang, et al.
Published: (2025)

ClusterKV: Manipulating LLM KV Cache in Semantic Space for Recallable Compression
by: Liu, Guangda, et al.
Published: (2024)

AttentionPredictor: Temporal Patterns Matter for KV Cache Compression
by: Yang, Qingyue, et al.
Published: (2025)

KVSculpt: KV Cache Compression as Distillation
by: Jiang, Bo, et al.
Published: (2026)

Reasoning Language Model Inference Serving Unveiled: An Empirical Study
by: Li, Qi, et al.
Published: (2025)

Multi-view Hypergraph-based Contrastive Learning Model for Cold-Start Micro-video Recommendation
by: Lyu, Sisuo, et al.
Published: (2024)

FuseFL: One-Shot Federated Learning through the Lens of Causality with Progressive Model Fusion
by: Tang, Zhenheng, et al.
Published: (2024)

Are Dilemmas and Conflicts in LLM Alignment Solvable? A View from Priority Graph
by: Tang, Zhenheng, et al.
Published: (2026)

Beyond Speedup -- Utilizing KV Cache for Sampling and Reasoning
by: Xing, Zeyu, et al.
Published: (2026)

Not All Heads Matter: A Head-Level KV Cache Compression Method with Integrated Retrieval and Reasoning
by: Fu, Yu, et al.
Published: (2024)

LongFlow: Efficient KV Cache Compression for Reasoning Models
by: Su, Yi, et al.
Published: (2026)

SemantiCache: Efficient KV Cache Compression via Semantic Chunking and Clustered Merging
by: Wu, Shunlong, et al.
Published: (2026)

PyramidKV: Dynamic KV Cache Compression based on Pyramidal Information Funneling
by: Cai, Zefan, et al.
Published: (2024)

Hold Onto That Thought: Assessing KV Cache Compression On Reasoning
by: Liu, Minghui, et al.
Published: (2025)

ReasonCache: Accelerating Large Reasoning Model Serving through KV Cache Sharing
by: Chen, Kaiwen, et al.
Published: (2025)

KV-CoRE: Benchmarking Data-Dependent Low-Rank Compressibility of KV-Caches in LLMs
by: Chen, Jian, et al.
Published: (2026)

PyramidInfer: Pyramid KV Cache Compression for High-throughput LLM Inference
by: Yang, Dongjie, et al.
Published: (2024)

ThinKV: Thought-Adaptive KV Cache Compression for Efficient Reasoning Models
by: Ramachandran, Akshat, et al.
Published: (2025)

Is Your LLM-as-a-Recommender Agent Trustable? LLMs' Recommendation is Easily Hacked by Biases (Preferences)
by: Tang, Zichen, et al.
Published: (2026)

OmniReview: A Large-scale Benchmark and LLM-enhanced Framework for Realistic Reviewer Recommendation
by: Huang, Yehua, et al.
Published: (2026)

CaliDrop: KV Cache Compression with Calibration
by: Su, Yi, et al.
Published: (2025)

The Pitfalls of KV Cache Compression
by: Chen, Alex, et al.
Published: (2025)