Metraipolis › archive › dc_9fe01b655ca1
Self-Indexing Latent Memories for Access-Aware Continual Instruction Tuning of LLMs — arXiv:2611.11684
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2611.11684
- License (as archived)
- CC0-1.0
- Archived
- 2026-11-10T18:41:29.393Z
- SHA-256 of the archived text
6acb124a0a83cca9a1be629bbc1c4381568ace606f58c1f0c551d67da2cb3b29- Stable URL of this page
https://metraipolis.com/documento/dc_9fe01b655ca1
How to cite
Self-Indexing Latent Memories for Access-Aware Continual Instruction Tuning of LLMs — arXiv:2611.11684 — arXiv (en), https://arxiv.org/abs/2611.11684 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_9fe01b655ca1 (Metraipolis archive dc_9fe01b655ca1, sha256 6acb124a0a83cca9a1be629bbc1c4381568ace606f58c1f0c551d67da2cb3b29)
BibTeX
@misc{dc9fe01b655ca1,
title = {Self-Indexing Latent Memories for Access-Aware Continual Instruction Tuning of LLMs — arXiv:2611.11684},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2611.11684},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_9fe01b655ca1; sha256 6acb124a0a83cca9a1be629bbc1c4381568ace606f58c1f0c551d67da2cb3b29},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.11684. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Continual instruction tuning often preserves old skills with task-private modules. Inference still requires selecting the right module from the prompt alone. We introduce Task-Isolated Latent Adapter Continual Training (TILA-CT), a replay-free method that makes each task module self-indexing. Each task receives a latent memory and adapter bank in a frozen LLM. During adaptation, the memory gathers causal-prefix information and injects task-conditioned context. For access, its persistent state provides the routing prototype used by a shared prompt-only router. On the 8-task CT-Pipeline stream, TILA-CT achieves 65.19 acquisition and 63.98 final macro score, with 1.38 points of average forgetting. A middle-reversed order gives 65.38 acquisition and 64.17 final macro score. Across 6 backbones, final scores remain within 1.5 points of acquisition. Ablations support memory-derived routing keys and causal gathering. Final macro routing accuracy is 93.9\%, and oracle routing recovers 1.21 final macro points. These results separate preserved task computation from learned access to it. Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno). Categorie arXiv: Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2611.11684 — 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 2026-11-10T18:37:37.518Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).