Metraipolis › archive › dc_701fcf66b99b

Fewer Heads, Fewer Memories: Pruning Redundant Attention Heads Suppresses Verbatim Memorisation — arXiv:2610.40974

Source: arXiv (en)
Original: https://arxiv.org/abs/2610.40974
License (as archived): CC0-1.0
Archived: 2026-11-02T07:33:20.701Z
SHA-256 of the archived text: bd8763590e51e3a63725b7c39960f31bb12e74eca4cab3d565b446cb01d636f8
Stable URL of this page: https://metraipolis.com/documento/dc_701fcf66b99b

How to cite

Fewer Heads, Fewer Memories: Pruning Redundant Attention Heads Suppresses Verbatim Memorisation — arXiv:2610.40974 — arXiv (en), https://arxiv.org/abs/2610.40974 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_701fcf66b99b (Metraipolis archive dc_701fcf66b99b, sha256 bd8763590e51e3a63725b7c39960f31bb12e74eca4cab3d565b446cb01d636f8)

BibTeX

@misc{dc701fcf66b99b,
  title = {Fewer Heads, Fewer Memories: Pruning Redundant Attention Heads Suppresses Verbatim Memorisation — arXiv:2610.40974},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.40974},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_701fcf66b99b; sha256 bd8763590e51e3a63725b7c39960f31bb12e74eca4cab3d565b446cb01d636f8},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.40974. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Large language models have been shown to memorise and reproduce verbatim sequences from their training data, raising concerns about privacy leakage, intellectual property exposure, and the limits of generalisation. Recent theoretical work suggests that attention heads may contribute to this behaviour by functioning as associative memory mechanisms, with memorisation capacity scaling with the number of heads. At the same time, empirical work has shown that many attention heads are redundant and can be pruned with limited impact on downstream performance. Despite these parallel findings, the effect of pruning redundant heads on memorised-sequence reproduction remains unexplored. In this work, we demonstrate that zeroing out a substantial fraction of attention heads prior to inference can significantly suppress verbatim memorisation while preserving, or even improving, language modelling performance on the benchmark used to identify redundant heads. Evaluating GPT-2-XL(1.5B) on known memorised sequences, we find that pruning 729 task-redundant heads reduces the memorisation count by 91.5%, while improving WikiText-103 perplexity by 0.92. These findings suggest that attention-head redundancy can be exploited to mitigate memorisation at inference time, without retraining or modifying the training data.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.CL, nuovi lavori del giorno).
Categorie arXiv: Computation and Language (cs.CL) ; Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2610.40974 — 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 2026-11-02T07:27:12.620Z dalla pagina https://arxiv.org/list/cs.CL/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).