Metraipolis › archive › dc_3876387ef314
Training-Free Task Vectors for LLM Behavioral Control — arXiv:2609.09054
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.09054
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-16T13:58:28.231Z
- SHA-256 of the archived text
dc4fe48d20d5b9c75dbe4492c059a538a5c85380f62d0af3a4f5e868a349bc92- Stable URL of this page
https://metraipolis.com/documento/dc_3876387ef314
How to cite
Training-Free Task Vectors for LLM Behavioral Control — arXiv:2609.09054 — arXiv (en), https://arxiv.org/abs/2609.09054 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_3876387ef314 (Metraipolis archive dc_3876387ef314, sha256 dc4fe48d20d5b9c75dbe4492c059a538a5c85380f62d0af3a4f5e868a349bc92)
BibTeX
@misc{dc3876387ef314,
title = {Training-Free Task Vectors for LLM Behavioral Control — arXiv:2609.09054},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.09054},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_3876387ef314; sha256 dc4fe48d20d5b9c75dbe4492c059a538a5c85380f62d0af3a4f5e868a349bc92},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.09054. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Task vectors enable post-training model editing by identifying semantically meaningful directions in weight space, typically computed as the difference between a fine-tuned model and its pretrained initialization. However, this reliance on fine-tuning makes discovering such directions costly and limits the practicality of post-training model editing. To address this limitation, we introduce Training-Free Task Vectors (TFTVs), a novel method to compute task-vector-like directions without requiring fine-tuning. Our method maps activation steering vectors to rank-one weight-space edits using only forward-pass statistics, while satisfying arithmetic properties that directly support learning via addition, forgetting via subtraction, and the composition of multiple edits. Empirically, we evaluate TFTVs on large language model behavioral control tasks and show that they consistently amplify, suppress, and compose target behaviors while preserving general knowledge and problem-solving skills. We also validate our method against other editing and steering baselines, experimentally demonstrating that TFTVs achieve stronger trait control with better or competitive utility preservation. We hope our work opens new directions for the community in post-training model editing and broader training-free model control. Code is available on the project website: this http URL . Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-09 in cs.LG). Categorie arXiv: Machine Learning (cs.LG) ; Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2609.09054 — 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 2026-10-16T13:39:43.083Z dalla pagina https://arxiv.org/catchup/cs.LG/2026-09-09?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).