Metraipolis › archive › dc_8938ef2decf4

Beyond K/V Caches: Cross-Model State Handoff in Hybrid Linear-Attention LLMs — arXiv:2611.08133

Source: arXiv (en)
Original: https://arxiv.org/abs/2611.08133
License (as archived): CC0-1.0
Archived: 2026-11-06T14:43:14.429Z
SHA-256 of the archived text: 479ca250c22e4cba677d9df20b0b515acfb585ccf5850c5a741e61fe830919ef
Stable URL of this page: https://metraipolis.com/documento/dc_8938ef2decf4

How to cite

Beyond K/V Caches: Cross-Model State Handoff in Hybrid Linear-Attention LLMs — arXiv:2611.08133 — arXiv (en), https://arxiv.org/abs/2611.08133 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_8938ef2decf4 (Metraipolis archive dc_8938ef2decf4, sha256 479ca250c22e4cba677d9df20b0b515acfb585ccf5850c5a741e61fe830919ef)

BibTeX

@misc{dc8938ef2decf4,
  title = {Beyond K/V Caches: Cross-Model State Handoff in Hybrid Linear-Attention LLMs — arXiv:2611.08133},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2611.08133},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_8938ef2decf4; sha256 479ca250c22e4cba677d9df20b0b515acfb585ccf5850c5a741e61fe830919ef},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.08133. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

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 $21.91\times$.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.CL, nuovi lavori del giorno).
Categorie arXiv: Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2611.08133 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-11-06T14:18:57.592Z dalla pagina https://arxiv.org/list/cs.CL/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).