Metraipolis › archive › dc_ff3ed3cf3bd0
Why Constrained Updates Still Forget: A Functional-Path Analysis of Continual SFT — arXiv:2610.30191
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2610.30191
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-25T20:55:41.331Z
- SHA-256 of the archived text
224088039e5db726ed90cd207b48525996cde4589016753e69d3a6f6afb39bff- Stable URL of this page
https://metraipolis.com/documento/dc_ff3ed3cf3bd0
How to cite
Why Constrained Updates Still Forget: A Functional-Path Analysis of Continual SFT — arXiv:2610.30191 — arXiv (en), https://arxiv.org/abs/2610.30191 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_ff3ed3cf3bd0 (Metraipolis archive dc_ff3ed3cf3bd0, sha256 224088039e5db726ed90cd207b48525996cde4589016753e69d3a6f6afb39bff)
BibTeX
@misc{dcff3ed3cf3bd0,
title = {Why Constrained Updates Still Forget: A Functional-Path Analysis of Continual SFT — arXiv:2610.30191},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2610.30191},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_ff3ed3cf3bd0; sha256 224088039e5db726ed90cd207b48525996cde4589016753e69d3a6f6afb39bff},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.30191. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Supervised fine-tuning adapts language models to specialized tasks but can disrupt previously acquired behavior. Existing approaches constrain parameters, gradients, or replay objectives, yet the connection between the protected object, the intervening trajectory, and retained outputs remains unclear. We develop a functional-path account that decomposes old-loss changes along realized updates into the controlled direction, gradient staleness, probe mismatch, and finite-step effects. The account yields a conditional refresh budget and identifies the restoring direction supplied by informative replay. Matched QLoRA interventions separate constraint granularity from placement before or after Adam: placement strongly affects retention across the studied code-label audits, whereas the granularity interaction depends on the endpoint. Checkpointed generation separates endpoint recovery from higher sampled-path accuracy, and matched replay schedules reveal loss exposure concealed by final scores. A sequential-task comparison extends this distinction to backward transfer: subspace regularization and projection improve average retention while trading off different old tasks; supervised replay protects both more consistently in this setting. Additional response interfaces distinguish content gains from exact-response retention. These findings connect update-level protection to task-wise behavior and motivate mitigation through explicit functional anchors, trajectory evaluation, and joint old/new auditing. Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno). Categorie arXiv: Machine Learning (stat.ML) ; Artificial Intelligence (cs.AI); Machine Learning (cs.LG). Pagina del lavoro: https://arxiv.org/abs/2610.30191 — 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 2026-10-25T20:47:56.964Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).