Metraipolis › archive › dc_8d623e925212

Post-Training, Finetuning, and Steering:A Field Guide for Curating Training Data — arXiv:2610.19143

Source
arXiv (en)
Original
https://arxiv.org/abs/2610.19143
License (as archived)
CC0-1.0
Archived
2026-10-15T21:57:43.597Z
SHA-256 of the archived text
e2abaf2f4fd07d488846b055fef9f33ff394de227d2f35afe8757d739fad374a
Stable URL of this page
https://metraipolis.com/documento/dc_8d623e925212

How to cite

Post-Training, Finetuning, and Steering:A Field Guide for Curating Training Data — arXiv:2610.19143 — arXiv (en), https://arxiv.org/abs/2610.19143 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_8d623e925212 (Metraipolis archive dc_8d623e925212, sha256 e2abaf2f4fd07d488846b055fef9f33ff394de227d2f35afe8757d739fad374a)
BibTeX
@misc{dc8d623e925212,
  title = {Post-Training, Finetuning, and Steering:A Field Guide for Curating Training Data — arXiv:2610.19143},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.19143},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_8d623e925212; sha256 e2abaf2f4fd07d488846b055fef9f33ff394de227d2f35afe8757d739fad374a},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.19143. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

While Vision-Language-Action models have enabled impressive cross-task transfer in zero-shot settings, \emph{adaptation} methods are often necessary to achieve consistent performance on novel tasks. However, even though a wealth of methods for adaptation have been proposed, it is unclear how to choose the best method based on available trajectories. We close this gap by studying the impact of the adaptation method under a variety of common data scenarios: on- and off-policy data, offline vs online adaptation, and whether the data contains optimal or suboptimal trajectories. Our evaluation spans simulated tasks in the LIBERO suite and real tasks on a Franka platform. In the offline data regime, critic-based steering approaches outperform supervised finetuning when failure data is available. However, offline RL techniques for training critic functions fail to outperform simpler Monte-Carlo-based approaches. Finally, in the online setting, critic-based steering is significantly more sample efficient compared to policy-gradient-based finetuning methods, but has a lower performance ceiling than full finetuning.

Tema degli esploratori: statistica e scienza dei dati (elenco arXiv cs.RO, nuovi lavori del giorno).
Categorie arXiv: Robotics (cs.RO).
Pagina del lavoro: https://arxiv.org/abs/2610.19143 — 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 2026-10-15T21:38:44.428Z dalla pagina https://arxiv.org/list/cs.RO/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).