Adaptive Mass-Segmented KV Compression for Long-Context Reasoning

arXiv:2605.23200 · cs.LG, cs.AI · Submitted 2026-05-22 · Read on arXiv

cs.LG, cs.AI

Submitted: 2026-05-22

Updated: 2026-09-29

Comments: Accepted at NeurIPS 2026

Code: https://github.com/EIT-NLP/AdaptiveMassSegment

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Terminology

• Adaptive Mass-Segmented — 2× in this paper
• KV cache — 1× in this paper · explained in 3 episodes

Sources

• Efficient Memory Management for Large Language Model Serving with PagedAttention
• KVQuant: Towards 10 Million Context Length LLM Inference with KV Cache Quantization
• Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
• R-KV: Redundancy-aware KV Cache Compression for Reasoning Models
• TriAttention: Efficient Long Reasoning with Trigonometric KV Compression
• FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness
• Efficient Streaming Language Models with Attention Sinks
• PyramidKV: Dynamic KV Cache Compression based on Pyramidal Information Funneling
• Ada-KV: Optimizing KV Cache Eviction by Adaptive Budget Allocation for Efficient LLM Inference
• Not All Heads Matter: A Head-Level KV Cache Compression Method with Integrated Retrieval and Reasoning
• SnapKV: LLM Knows What You are Looking for Before Generation
• SABlock: Semantic-Aware KV Cache Eviction with Adaptive Compression Block Size
• ClusterKV: Manipulating LLM KV Cache in Semantic Space for Recallable Compression
• TreeKV: Smooth Key-Value Cache Compression with Tree Structures
• HeteroCache: A Dynamic Retrieval Approach to Heterogeneous KV Cache Compression for Long-Context LLM Inference
• Expected Attention: KV Cache Compression by Estimating Attention from Future Queries Distribution
• ChunkKV: Semantic-Preserving KV Cache Compression for Efficient Long-Context LLM Inference
• KeyDiff: Key Similarity-Based KV Cache Eviction for Long-Context LLM Inference in Resource-Constrained Environments
• SCOPE: Optimizing Key-Value Cache Compression in Long-context Generation
• G-KV: Decoding-Time KV Cache Eviction with Global Attention

Related papers

• Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
• AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
• Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
• Convergence issues in Relational Concept Analysis based on AOC-posets
• Beliefs Beyond Posteriors: The Predictive Cost of ELBO Binding
• Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks