Just Let Linear States Forget the Distant Past: Prefix Caching via Suffix Replay for Hybrid LLMs

arXiv:2609.33477 · cs.AI, cs.PF · Submitted 2026-09-27 · Read on arXiv

cs.AI, cs.PF

Submitted: 2026-09-27

Updated: 2026-09-27

Code: https://github.com/vllm-project/vllm

Terminology

• Hybrid Models — 11× in this paper · explained in 2 episodes
• KV cache — 11× in this paper · explained in 3 episodes
• Transformer — 8× in this paper · explained in 2 episodes
• Decode — 7× in this paper · explained in 2 episodes
• Sparsity — 5× in this paper · explained in 4 episodes
• Trace — 5× in this paper · explained in 2 episodes
• Alignment — 3× in this paper · explained in 1 episode
• In International Conference on Learning Representations — 3× in this paper
• Attention Mechanism — 2× in this paper · explained in 3 episodes
• Benchmark — 2× in this paper · explained in 2 episodes
• Hybrid Architectures — 2× in this paper · explained in 3 episodes
• In International Conference on Machine Learning — 2× in this paper
• Abstraction — 1× in this paper · explained in 1 episode
• Calibration — 1× in this paper · explained in 7 episodes
• Communication Overhead — 1× in this paper · explained in 2 episodes
• Consistency — 1× in this paper · explained in 3 episodes
• Core — 1× in this paper · explained in 2 episodes
• Decoupling — 1× in this paper · explained in 2 episodes
• Fidelity — 1× in this paper · explained in 2 episodes
• Foundation Model — 1× in this paper · explained in 3 episodes
• Foundation Models — 1× in this paper · explained in 7 episodes
• KV cache reuse — 1× in this paper · explained in 2 episodes
• Multi-agent Collaborative Framework — 1× in this paper · explained in 2 episodes
• RAG — 1× in this paper · explained in 3 episodes
• Reconstruction Error — 1× in this paper · explained in 2 episodes
• Transformer Models — 1× in this paper · explained in 2 episodes
• USENIX Conference on File and Storage Technologies — 1× in this paper

Sources

• Transformers are SSMs: Generalized Models and Efficient Algorithms Through Structured State Space Duality
• Cost-Efficient Large Language Model Serving for Multi-turn Conversations with CachedAttention
• Prompt Cache: Modular Attention Reuse for Low-Latency Inference
• MetaGPT: Meta Programming for A Multi-Agent Collaborative Framework
• EPIC: Efficient Position-Independent Caching for Serving Large Language Models
• Kimi Linear: An Expressive, Efficient Attention Architecture
• Efficient Memory Management for Large Language Model Serving with PagedAttention
• Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks
• Jamba: A Hybrid Transformer-Mamba Language Model
• LinearKV: One Cached State Suffices for Position-Independent Caching in Hybrid LLMs
• HYPIC: Accelerating Hybrid-Attention LLM Serving with Position-Independent Caching
• Olmo Hybrid: From Theory to Practice and Back
• MiniMax-01: Scaling Foundation Models with Lightning Attention
• Skeleton-of-Thought: Prompting LLMs for Efficient Parallel Generation
• Nemotron-H: A Family of Accurate and Efficient Hybrid Mamba-Transformer Models
• Training language models to follow instructions with human feedback
• Marconi: Prefix Caching for the Era of Hybrid LLMs
• Mooncake: A KVCache-centric Disaggregated Architecture for LLM Serving
• Sparse Prefix Caching for Hybrid and Recurrent LLM Serving
• Preble: Efficient Distributed Prompt Scheduling for LLM Serving

Related papers

• MAVEN-T: Reinforced Heterogeneous Distillation for Real-Time Multi-Agent Trajectory Prediction
• Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models
• The Clinician's Veto: Navigating Trust, Liability, and Uncertainty in Autonomous AI Prescribing
• MindHelper: Closed-Loop Embodied Mental-State Reasoning for Precision Intervention
• Incumbent Advantage: Brand Bias and Cognitive Manipulation Dynamics in LLM Recommendation Systems
• VSAL: A Vision Solver with Adaptive Layouts for Graph Property Detection