Metraipolis › archive › dc_0b511237ed7b

Choosing the Number of Attention Heads from Intrinsic Dimension: A Small-Scale Study — arXiv:2610.19966

Source
arXiv (en)
Original
https://arxiv.org/abs/2610.19966
License (as archived)
CC0-1.0
Archived
2026-10-15T14:23:42.491Z
SHA-256 of the archived text
7b5634c994b92ebe4e60ca0c5c93e01dd6daf1641b0c76d5d39c35dd2a41171a
Stable URL of this page
https://metraipolis.com/documento/dc_0b511237ed7b

How to cite

Choosing the Number of Attention Heads from Intrinsic Dimension: A Small-Scale Study — arXiv:2610.19966 — arXiv (en), https://arxiv.org/abs/2610.19966 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_0b511237ed7b (Metraipolis archive dc_0b511237ed7b, sha256 7b5634c994b92ebe4e60ca0c5c93e01dd6daf1641b0c76d5d39c35dd2a41171a)
BibTeX
@misc{dc0b511237ed7b,
  title = {Choosing the Number of Attention Heads from Intrinsic Dimension: A Small-Scale Study — arXiv:2610.19966},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.19966},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_0b511237ed7b; sha256 7b5634c994b92ebe4e60ca0c5c93e01dd6daf1641b0c76d5d39c35dd2a41171a},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.19966. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

The number of attention heads in a transformer is usually set by convention or grid search. We test a simple heuristic for setting it in advance: estimate the intrinsic dimension of token representations without labels, and use roughly that many heads. We estimate intrinsic dimension with a nested-dropout VAE trained on DistilBERT embeddings of WikiText-2, choosing the KL weight at the knee of the VAE's rate-distortion curve rather than by hand. This gives $k^*=9.2$ (95% CI [8.6, 9.8]). In a small 4-layer transformer trained from scratch, 10 heads and 32 heads are statistically equivalent on SST-2 and on MRPC (paired TOST over 10 seeds, $\pm1$ point margin), with 69% fewer attention parameters for the 10-head model. Training 10 heads from the start also slightly outperforms training 32 heads and pruning to 10 (0.6 points, $p=0.008$). The evidence is narrow: two classification tasks, one small architecture, no pretrained models, and no formal link between intrinsic dimension and head count. We present the result as a practical heuristic worth testing at larger scale, not as a general principle.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.LG, nuovi lavori del giorno).
Categorie arXiv: Computation and Language (cs.CL) ; Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2610.19966 — 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 2026-10-15T13:53:35.929Z dalla pagina https://arxiv.org/list/cs.LG/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).