Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Fodeh, Samah, Ma, Linhai, Puthiaraju, Ganesh, Talakokkul, Srivani, Khan, Afshan, Hagaman, Ashley, Lowe, Sarah R., Roundtree, Aimee Kendall
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2603.00025
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866918360810258432
author Fodeh, Samah
Ma, Linhai
Puthiaraju, Ganesh
Talakokkul, Srivani
Khan, Afshan
Hagaman, Ashley
Lowe, Sarah R.
Roundtree, Aimee Kendall
author_facet Fodeh, Samah
Ma, Linhai
Puthiaraju, Ganesh
Talakokkul, Srivani
Khan, Afshan
Hagaman, Ashley
Lowe, Sarah R.
Roundtree, Aimee Kendall
contents Direct Preference Optimization is an offline post-SFT method for aligning language models from preference pairs, with strong results in instruction following and summarization. However, DPO's sequence-level implicit reward can be brittle for token-critical structured prediction settings such as medical annotation, which often exhibit (i) low-separation preference pairs, where chosen and rejected completions differ by minimal edit distance (often 1-3 tokens), and (ii) token-importance skew, where sparse semantic tokens (hierarchical labels and evidence Spans) carry disproportionate task importance relative to high-frequency structural tokens (JSON scaffolding). In this regime, standard DPO suffers from margin collapse (insufficient log-probability separation between near-identical preferences), likelihood squeezing (the margin objective shifts the absolute likelihoods of both completions together), and gradient dilution, where uniform sequence-level weighting diffuses learning signal across shared scaffolding while rare, confusable label tokens receive weak, noisy updates. We introduce Token-Adaptive Barrier Preference Optimization (TAB-PO), which augments DPO with token-weighted, reference-adjusted advantages that prioritize high-value semantic tokens, and a conditional token-level barrier that regularizes under-confident tokens balancing SFT-anchored likelihood and preference-driven separation in low-separation, importance-skewed regimes. We evaluate TAB-PO on medical communication annotation, a task requiring joint prediction of hierarchical labels and evidence Spans from patient-provider messages. TAB-PO achieves a ~ 4% relative improvement in micro-F1 over SFT and consistently outperforms recent preference-optimization baselines.
format Preprint
id arxiv_https___arxiv_org_abs_2603_00025
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle TAB-PO: Preference Optimization with a Token-Level Adaptive Barrier for Token-Critical Structured Generation
Fodeh, Samah
Ma, Linhai
Puthiaraju, Ganesh
Talakokkul, Srivani
Khan, Afshan
Hagaman, Ashley
Lowe, Sarah R.
Roundtree, Aimee Kendall
Computation and Language
Direct Preference Optimization is an offline post-SFT method for aligning language models from preference pairs, with strong results in instruction following and summarization. However, DPO's sequence-level implicit reward can be brittle for token-critical structured prediction settings such as medical annotation, which often exhibit (i) low-separation preference pairs, where chosen and rejected completions differ by minimal edit distance (often 1-3 tokens), and (ii) token-importance skew, where sparse semantic tokens (hierarchical labels and evidence Spans) carry disproportionate task importance relative to high-frequency structural tokens (JSON scaffolding). In this regime, standard DPO suffers from margin collapse (insufficient log-probability separation between near-identical preferences), likelihood squeezing (the margin objective shifts the absolute likelihoods of both completions together), and gradient dilution, where uniform sequence-level weighting diffuses learning signal across shared scaffolding while rare, confusable label tokens receive weak, noisy updates. We introduce Token-Adaptive Barrier Preference Optimization (TAB-PO), which augments DPO with token-weighted, reference-adjusted advantages that prioritize high-value semantic tokens, and a conditional token-level barrier that regularizes under-confident tokens balancing SFT-anchored likelihood and preference-driven separation in low-separation, importance-skewed regimes. We evaluate TAB-PO on medical communication annotation, a task requiring joint prediction of hierarchical labels and evidence Spans from patient-provider messages. TAB-PO achieves a ~ 4% relative improvement in micro-F1 over SFT and consistently outperforms recent preference-optimization baselines.
title TAB-PO: Preference Optimization with a Token-Level Adaptive Barrier for Token-Critical Structured Generation
topic Computation and Language
url https://arxiv.org/abs/2603.00025