Metraipolis › archive › dc_affb60ed08fb

Separating Stream Stability from Long-Term Recall in Language Models — arXiv:2609.07282

Source: arXiv (en)
Original: https://arxiv.org/abs/2609.07282
License (as archived): CC0-1.0
Archived: 2026-10-16T12:45:30.831Z
SHA-256 of the archived text: 9bd564264b80c87ecda4de16e60342d535a768265d60961b77ef2193c9ac1d3d
Stable URL of this page: https://metraipolis.com/documento/dc_affb60ed08fb

How to cite

Separating Stream Stability from Long-Term Recall in Language Models — arXiv:2609.07282 — arXiv (en), https://arxiv.org/abs/2609.07282 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_affb60ed08fb (Metraipolis archive dc_affb60ed08fb, sha256 9bd564264b80c87ecda4de16e60342d535a768265d60961b77ef2193c9ac1d3d)

BibTeX

@misc{dcaffb60ed08fb,
  title = {Separating Stream Stability from Long-Term Recall in Language Models — arXiv:2609.07282},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.07282},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_affb60ed08fb; sha256 9bd564264b80c87ecda4de16e60342d535a768265d60961b77ef2193c9ac1d3d},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.07282. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Methods for streaming language models are often discussed alongside long-context and memory systems, although they solve different problems. An attention sink can stabilize autoregressive generation over an indefinitely long stream while the model remains unable to use content that has left its recent-token cache. We argue that this distinction should be explicit in system claims and evaluation. We introduce three horizons: the stability horizon, over which predictive behavior remains well behaved; the access horizon, over which past content can still causally affect the output; and the utility horizon, over which a task retains acceptable performance. We show constructively that the stability horizon can be infinite while the access and utility horizons are finite. We then propose ThreeH, an evaluation contract that measures all three horizons under a common state and compute budget. Applying the framework to attention-sink streaming clarifies its strength, constant-memory, stable generation, without treating anchor tokens as semantic memory. The framework exposes roles for cache policies, recurrent state, retrieval, and external memory. Experiments on 128K-token streams, delayed binding recall, and delayed decisions show that attention sinks preserve local modeling but not content beyond the active cache; recurrent and retrieval state extend the semantic horizon.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-09 in cs.CL).
Categorie arXiv: Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2609.07282 — 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 2026-10-16T12:30:25.105Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-09-09?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).