---
title: "Beyond K/V Caches: Cross-Model State Handoff in Hybrid Linear-Attention LLMs"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2611.08133v1"
permalink: "https://feed.astralor.com/items/itm_238b9b07556bcf972a08dddb"
published: "2026-11-05T15:00:50.000Z"
score: 59
channel: "tech"
tags: ["LLM","安全","Beyond","Caches","Cross-Model State Handoff","Hybrid Linear-Attention LLMs","Production","Context"]
---

# Beyond K/V Caches: Cross-Model State Handoff in Hybrid Linear-Attention LLMs

## 速览

Production deployments often use different models to balance cost and quality through cascading, routing, and switching models during long conversations. Context accumulates across turns. Each switch requires the target…

## 关键点

- Production deployments often use different models to balance cost and quality through cascading, routing, and switching models during long conversations.
- Context accumulates across turns.
- Each switch requires the target to repeat prefill over this context.

## 全文

Production deployments often use different models to balance cost and quality through cascading, routing, and switching models during long conversations. Context accumulates across turns. Each switch requires the target to repeat prefill over this context. Existing methods transfer full-attention (FA) K/V caches across models. Hybrid linear-attention models also retain recurrent states (REC) and convolution histories (Conv). The interaction between Conv and REC complicates cross-model state transfer. Conv perturbations can change keys and values in recurrent layers and alter REC updates. Conv histories are overwritten, but changes in REC can still affect later outputs. More accurate REC reconstruction need not improve continuation. Building on this, we propose Hybrid State Transfer (HST) for all three state types. It uses structured REC maps and preserves history order during Conv alignment. We calibrate the adapters using the frozen target's outputs during continuation. FA cache map updates are merged offline. Across five benchmarks, Qwen3.5 4B$\rightarrow$27B and 9B$\rightarrow$27B achieve mean task retentions of 89.88% and 94.11%, respectively. With source caches already available, separate timing on synthetic 16K inputs yields handoff body speedups of $21.66\times$ and $21.91\times$.

---

引用前请核对原文;本文由星际流动整理:https://feed.astralor.com/items/itm_238b9b07556bcf972a08dddb