Metraipolis › archive › dc_ed4efc2c79ce
Anchor: Augmented Cross-Layer Recurrent State for Long-Context Retrieval in Test-Time Training — arXiv:2611.04327
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2611.04327
- License (as archived)
- CC0-1.0
- Archived
- 2026-11-04T12:29:52.393Z
- SHA-256 of the archived text
119d3ea1634fe4ad9ba5ea90afd3b1224a166225a5c2c492572c3957b488cb2e- Stable URL of this page
https://metraipolis.com/documento/dc_ed4efc2c79ce
How to cite
Anchor: Augmented Cross-Layer Recurrent State for Long-Context Retrieval in Test-Time Training — arXiv:2611.04327 — arXiv (en), https://arxiv.org/abs/2611.04327 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_ed4efc2c79ce (Metraipolis archive dc_ed4efc2c79ce, sha256 119d3ea1634fe4ad9ba5ea90afd3b1224a166225a5c2c492572c3957b488cb2e)
BibTeX
@misc{dced4efc2c79ce,
title = {Anchor: Augmented Cross-Layer Recurrent State for Long-Context Retrieval in Test-Time Training — arXiv:2611.04327},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2611.04327},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_ed4efc2c79ce; sha256 119d3ea1634fe4ad9ba5ea90afd3b1224a166225a5c2c492572c3957b488cb2e},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.04327. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Parameter-efficient linearization turns a pretrained autoregressive Transformer into an efficient model by hybridizing sliding-window attention (SWA) with a constant-size recurrent memory. But does every layer's recurrent state contribute to long-context retrieval? Investigating an SWA and Test-Time Training (TTT) hybrid, we find that retrieval is sparse across depth, concentrated in a few anchor layers, and that the remaining states hold similar information expressed through layer-specific transformations. We exploit both observations in Anchor: TTT memory placed only at the anchor layers, and a cross-layer recurrent state written once and read by many layers, each through its own learned read map that realizes a layer-specific stretch and rotation, along with demonstrated generalization to other hybrid attentions. On Llama-3.1-8B and Mistral-7B, Anchor delivers 2.3-3.1x higher recall and 6.7-8.9 points higher MMLU accuracy than LoLCATs with a 10x smaller recurrent state, retains 91-95% of the teacher's MMLU accuracy, and cuts end-to-end latency by up to 4.3x relative to softmax attention. Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno). Categorie arXiv: Machine Learning (cs.LG) ; Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2611.04327 — 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 2026-11-04T12:26:47.270Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).