Metraipolis › archive › dc_f92549e2fae3

Output-Aware Block Influence: Jacobian-Lens Weighting for Depth Pruning — arXiv:2611.00015

Source
arXiv (en)
Original
https://arxiv.org/abs/2611.00015
License (as archived)
CC0-1.0
Archived
2026-11-03T11:04:27.497Z
SHA-256 of the archived text
ac92b1fe86797eea5558840335a41b8928e0926f63473f41c3a28edf3c7c4679
Stable URL of this page
https://metraipolis.com/documento/dc_f92549e2fae3

How to cite

Output-Aware Block Influence: Jacobian-Lens Weighting for Depth Pruning — arXiv:2611.00015 — arXiv (en), https://arxiv.org/abs/2611.00015 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_f92549e2fae3 (Metraipolis archive dc_f92549e2fae3, sha256 ac92b1fe86797eea5558840335a41b8928e0926f63473f41c3a28edf3c7c4679)
BibTeX
@misc{dcf92549e2fae3,
  title = {Output-Aware Block Influence: Jacobian-Lens Weighting for Depth Pruning — arXiv:2611.00015},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2611.00015},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_f92549e2fae3; sha256 ac92b1fe86797eea5558840335a41b8928e0926f63473f41c3a28edf3c7c4679},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.00015. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Depth pruning removes entire transformer blocks and therefore depends on a reliable criterion for identifying which blocks can be removed. Block Influence (BI) ranks blocks by how minimally they rotate the hidden state. However, local redundancy does not necessarily imply downstream irrelevance. We address this limitation using the Jacobian Lens (J-Lens) neural artifact, a map transporting perturbations at a layer to the final residual stream. At a common sparsity level, pruning scores applying J-Lens transportation to $BI$, what we call $J-BI$, were found to improve post-pruning accuracy by up to 19 points over $BI$ while reducing damage to perplexity for Qwen3-8B, Llama-3.1-8B-IT, and Gemma-3-12B-IT. We also perform a single-block ablation study and use J-Lens probing to explain why the proposed scores improved performance.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno).
Categorie arXiv: Machine Learning (cs.LG) ; Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2611.00015 — 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 2026-11-03T10:59:45.901Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).