Metraipolis › archive › dc_f54f8bf34342

Modern Transformers Are Implicit Hybrids: From Functional Differentiation to Principled Hybrid Architecture Design — arXiv:2609.02986

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.02986
License (as archived)
CC0-1.0
Archived
2026-10-17T02:07:42.393Z
SHA-256 of the archived text
094e0ff6b0a5003d92ff96589e446a7c25085b7024119e75aa3b3019d45ca3b6
Stable URL of this page
https://metraipolis.com/documento/dc_f54f8bf34342

How to cite

Modern Transformers Are Implicit Hybrids: From Functional Differentiation to Principled Hybrid Architecture Design — arXiv:2609.02986 — arXiv (en), https://arxiv.org/abs/2609.02986 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_f54f8bf34342 (Metraipolis archive dc_f54f8bf34342, sha256 094e0ff6b0a5003d92ff96589e446a7c25085b7024119e75aa3b3019d45ca3b6)
BibTeX
@misc{dcf54f8bf34342,
  title = {Modern Transformers Are Implicit Hybrids: From Functional Differentiation to Principled Hybrid Architecture Design — arXiv:2609.02986},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.02986},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_f54f8bf34342; sha256 094e0ff6b0a5003d92ff96589e446a7c25085b7024119e75aa3b3019d45ca3b6},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.02986. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Hybrid architectures combining Full Attention (FA) and Linear Attention (LA) are increasingly prominent, yet their allocation remains heuristic. We seek an evidence-grounded basis in head-level functional organization learned by RoPE-based Transformers. Behavioral probes do not yield a complete taxonomy, so we propose two intervention metrics: RoPE Frequency Importance Score (RFIS), measuring how each frequency affects a head's attention distribution, and RoPE Positional Dependence (RPD), isolating dependence on rotary positional modulation. On Qwen3-series models and Llama3.1, RFIS suggests and RPD verifies a complete taxonomy of retrieval and positional heads separated by a salient mid-low-frequency band. Controlled Transformers show that this boundary follows the training-length positional scale; we term it the Global Positional Band (GPBand). The analysis suggests a potential cause of zero-shot length-extrapolation failure and yields two principles: positional modeling should operate only locally, with global access through position-independent retrieval; and both functions should be assigned at head granularity with layer-specific allocation. We instantiate them in Head-wise Hybrid Architecture (HwH), using NoPE FA for global retrieval and LA for local positional modeling. With an FA-to-LA ratio below 1:3, HwH retains strong language modeling and commonsense reasoning while improving retrieval and substantially strengthening zero-shot long-context extrapolation over Transformer, LA, and a layer-wise hybrid baseline. Ablations validate both principles and component roles, highlighting principled hybrid architecture design as a promising route toward future foundation models.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-04 in cs.LG).
Categorie arXiv: Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2609.02986 — 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 2026-10-17T01:47:25.512Z dalla pagina https://arxiv.org/catchup/cs.LG/2026-09-04?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).