Rehearse Everything, Remember Nothing: Attic-KV Rehearses What Will Be Read

arXiv:2610.12133 · cs.CL, cs.LG · Submitted 2026-10-08 · Read on arXiv

cs.CL, cs.LG

Submitted: 2026-10-08

Updated: 2026-10-09

Comments: 15 pages, 6 figures

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

Terminology

• KV cache — 33× in this paper · explained in 3 episodes
• Benchmark — 13× in this paper · explained in 2 episodes
• Anchor Tokens — 4× in this paper
• Ablation — 3× in this paper · explained in 1 episode
• Empirical Methods in Natural Language Processing — 3× in this paper
• Pruning — 3× in this paper · explained in 4 episodes
• Adam — 1× in this paper · explained in 2 episodes
• Aggregation — 1× in this paper
• Distillation — 1× in this paper · explained in 2 episodes
• International Conference on Learning Representations — 1× in this paper
• KV cache reuse — 1× in this paper · explained in 2 episodes
• Meeting of the Association for Computational Linguistics — 1× in this paper
• RAG — 1× in this paper · explained in 3 episodes
• Transformer — 1× in this paper · explained in 2 episodes

Sources

• RestoreKV: Recovering Full-Cache Behavior Under Aggressive Query-Agnostic KV Cache Eviction
• LongBench: A Bilingual, Multitask Benchmark for Long Context Understanding
• Monte Carlo Estimation for KV Cache Eviction
• PyramidKV: Dynamic KV Cache Compression based on Pyramidal Information Funneling
• Compactor: Calibrated Query-Agnostic KV Cache Compression with Approximate Leverage Scores
• Expected Attention: KV Cache Compression by Estimating Attention from Future Queries Distribution
• Cartridges: Lightweight and general-purpose long context representations via self-study
• Draft-based Approximate Inference for LLMs
• Prompt Cache: Modular Attention Reuse for Low-Latency Inference
• Back from the Future: Key-Value Cache Management by Counter-Causal Surprise
• The Llama 3 Herd of Models
• RULER: What's the Real Context Size of Your Long-Context Language Models?
• Behavior-Preserving KV Cache Compression
• KVzap: Fast, Adaptive, and Faithful KV Cache Pruning
• RAGCache: Efficient Knowledge Caching for Retrieval-Augmented Generation
• Fast KVzip: Efficient and Accurate LLM Inference with Gated KV Eviction
• KVzip: Query-Agnostic KV Cache Compression with Context Reconstruction
• Cartridges++: KV Cache Compression without Off-Context Derailment
• PatchKV: Weight-Space Compensation of KV Cache
• LooGLE: Can Long-Context Language Models Understand Long Contexts?

Related papers

• Exploring Solution Divergence and Its Effect on Large Language Model Problem Solving
• Ishigaki-IDS-Bench: A Benchmark for Generating Information Delivery Specification from BIM Information Requirements
• Subliminal Steering: Stronger Encoding of Hidden Signals
• MedStruct-S: A Benchmark for Key Discovery, Key-Conditioned QA and Semi-Structured Extraction from OCR Clinical Reports
• The End of Transformers? On Challenging Attention and the Rise of Sub-Quadratic Architectures
• Untangling the Mechanisms of Misleading Context in Medical Question Answering