Metraipolis › archive › dc_021c13809c36

Ranking and Partitioning of Attention Heads for Sparse and Continual Reinforcement Learning Fine-tuning — arXiv:2610.21336

Source
arXiv (en)
Original
https://arxiv.org/abs/2610.21336
License (as archived)
CC0-1.0
Archived
2026-10-16T10:02:16.801Z
SHA-256 of the archived text
2210f45a00c1bfd94b54252b69ed1a1248129eefe18726f90c49e24079a987e1
Stable URL of this page
https://metraipolis.com/documento/dc_021c13809c36

How to cite

Ranking and Partitioning of Attention Heads for Sparse and Continual Reinforcement Learning Fine-tuning — arXiv:2610.21336 — arXiv (en), https://arxiv.org/abs/2610.21336 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_021c13809c36 (Metraipolis archive dc_021c13809c36, sha256 2210f45a00c1bfd94b54252b69ed1a1248129eefe18726f90c49e24079a987e1)
BibTeX
@misc{dc021c13809c36,
  title = {Ranking and Partitioning of Attention Heads for Sparse and Continual Reinforcement Learning Fine-tuning — arXiv:2610.21336},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.21336},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_021c13809c36; sha256 2210f45a00c1bfd94b54252b69ed1a1248129eefe18726f90c49e24079a987e1},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.21336. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

We study sparse and continual reinforcement learning fine-tuning of language models through attention-head ranking and partitioning. For sparse learning, we rank attention heads by gradient norm to identify a small subset of heads that can be trained effectively. For continual learning, we partition heads across tasks to reduce cross-task interference and preserve previously learned capabilities. Theoretically, we relate gradient norm to training effectiveness, and further characterize why sparse, separated head updates can remain stable. Empirically, across math, instruction-following, and coding tasks, tuning higher-ranked heads consistently outperforms tuning lower-ranked ones while often matching denser baselines, and head partitioning improves retention while preserving adaptation to new tasks.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.CL, nuovi lavori del giorno).
Categorie arXiv: Machine Learning (cs.LG) ; Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2610.21336 — 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 2026-10-16T08:44:47.753Z dalla pagina https://arxiv.org/list/cs.CL/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).