Metraipolis › archive › dc_2f7829c3ff06

TLRS-DPO: Token-Level Risk Scoring for Better Direct Preference Optimization — arXiv:2611.11232

Source
arXiv (en)
Original
https://arxiv.org/abs/2611.11232
License (as archived)
CC0-1.0
Archived
2026-11-10T18:43:13.316Z
SHA-256 of the archived text
fcb91a90ad814da25d16e694b3c95ce373ee9ada018f7fd531cb4ffa67e990a9
Stable URL of this page
https://metraipolis.com/documento/dc_2f7829c3ff06

How to cite

TLRS-DPO: Token-Level Risk Scoring for Better Direct Preference Optimization — arXiv:2611.11232 — arXiv (en), https://arxiv.org/abs/2611.11232 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_2f7829c3ff06 (Metraipolis archive dc_2f7829c3ff06, sha256 fcb91a90ad814da25d16e694b3c95ce373ee9ada018f7fd531cb4ffa67e990a9)
BibTeX
@misc{dc2f7829c3ff06,
  title = {TLRS-DPO: Token-Level Risk Scoring for Better Direct Preference Optimization — arXiv:2611.11232},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2611.11232},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_2f7829c3ff06; sha256 fcb91a90ad814da25d16e694b3c95ce373ee9ada018f7fd531cb4ffa67e990a9},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.11232. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Direct Preference Optimization (DPO) is a simple and widely used alternative to RLHF for aligning large language models (LLMs), but its sequence-level objective cannot precisely locate or suppress the local patterns that make an otherwise acceptable response unsafe or undesirable. We propose TLRS-DPO (Token-Level Risk Scoring for Better Direct Preference Optimization), which augments DPO with fine-grained token-level control while remaining fully offline and reference-based. TLRS-DPO introduces a token-level robust compensation term that converts the policy-reference log-ratio on rejected responses into localized risk scores, a curriculum-style schedule that gradually increases the strength of this compensation during training, and a risk-adaptive screening mechanism that selectively penalizes only high-risk tokens instead of entire sequences. This design bridges sequence-level preference fitting and token-level behavior control, and can be implemented as a drop-in modification to standard DPO trainers. Empirically, TLRS-DPO improves alignment quality and safety-relevant behavior with minimal impact on helpfulness compared to strong DPO baselines.

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); Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2611.11232 — 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 2026-11-10T18:37:37.631Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).