Metraipolis › archive › dc_09a4d4e20ea1
Output-aware Residual Stream Pruning for Large Language Models — arXiv:2609.35579
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.35579
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-07T22:39:15.033Z
- SHA-256 of the archived text
a229198971f62cf024ef30c9dc5e4414724e592669a6518ae62c8ea9913f41f1- Stable URL of this page
https://metraipolis.com/documento/dc_09a4d4e20ea1
How to cite
Output-aware Residual Stream Pruning for Large Language Models — arXiv:2609.35579 — arXiv (en), https://arxiv.org/abs/2609.35579 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_09a4d4e20ea1 (Metraipolis archive dc_09a4d4e20ea1, sha256 a229198971f62cf024ef30c9dc5e4414724e592669a6518ae62c8ea9913f41f1)
BibTeX
@misc{dc09a4d4e20ea1,
title = {Output-aware Residual Stream Pruning for Large Language Models — arXiv:2609.35579},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.35579},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_09a4d4e20ea1; sha256 a229198971f62cf024ef30c9dc5e4414724e592669a6518ae62c8ea9913f41f1},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.35579. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Residual stream pruning methods reduce inference cost by shrinking the model's hidden dimension, but existing approaches typically choose these dimensions by minimizing activation reconstruction error. This criterion implicitly treats all perturbation directions as equally important, ignoring the sensitivity of downstream layers. We introduce a sensitivity-aware approach to residual-stream pruning that directly accounts for this direction-dependent sensitivity. Using a second-order approximation to the output KL divergence, we characterize the effect of a residual-stream perturbation through both its activation covariance and the local sensitivity of the model output. The resulting subspace selection objective couples these two quantities, but is difficult to optimize directly. We derive a tractable spectral upper bound that reduces subspace selection to an eigendecomposition of a sensitivity-weighted covariance matrix, retaining the efficiency and structural simplicity of rotation-based pruning methods. Across several instruction-tuned language model families, our method consistently reduces calibration KL divergence relative to activation-only pruning and improves perplexity and downstream task performance over a range of compression levels. Our results show that preserving activation energy alone is insufficient for residual-stream pruning, and that explicitly accounting for how perturbations propagate to the model output provides a more effective criterion for selecting dimensions to remove. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-29 in cs.LG). Categorie arXiv: Machine Learning (cs.LG). Pagina del lavoro: https://arxiv.org/abs/2609.35579 — 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 2026-10-07T22:14:28.479Z dalla pagina https://arxiv.org/catchup/cs.LG/2026-09-29?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).