ZEBRA: Leveraging Model-Behavioral Knowledge for Zero-Annotation Preference Dataset Construction

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Jung, Jeesu, Park, Chanjun, Jung, Sangkeun
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866916771296968704
author Jung, Jeesu
Park, Chanjun
Jung, Sangkeun
author_facet Jung, Jeesu
Park, Chanjun
Jung, Sangkeun
contents Recent efforts in LLM alignment have focused on constructing large-scale preference datasets via human or Artificial Intelligence (AI) annotators. However, such approaches rely on instance-wise supervision, incurring substantial annotation cost and limited interpretability. In this paper, we propose ZEBRA - a model behavior-wise zero-annotation framework that constructs preference data by leveraging model behavior knowledge derived from benchmark performances. ZEBRA binarizes response pairs by evaluating the quality and similarity of their origin models, entirely bypassing instance-level annotation. This allows scalable, controllable, and cost-effective alignment data generation. Empirical results show that ZEBRA achieves alignment performance comparable to instance-supervised methods, despite requiring no manual or model-based labeling.
format Preprint
id arxiv_https___arxiv_org_abs_2502_18744
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle ZEBRA: Leveraging Model-Behavioral Knowledge for Zero-Annotation Preference Dataset Construction
Jung, Jeesu
Park, Chanjun
Jung, Sangkeun
Artificial Intelligence
Computation and Language
Recent efforts in LLM alignment have focused on constructing large-scale preference datasets via human or Artificial Intelligence (AI) annotators. However, such approaches rely on instance-wise supervision, incurring substantial annotation cost and limited interpretability. In this paper, we propose ZEBRA - a model behavior-wise zero-annotation framework that constructs preference data by leveraging model behavior knowledge derived from benchmark performances. ZEBRA binarizes response pairs by evaluating the quality and similarity of their origin models, entirely bypassing instance-level annotation. This allows scalable, controllable, and cost-effective alignment data generation. Empirical results show that ZEBRA achieves alignment performance comparable to instance-supervised methods, despite requiring no manual or model-based labeling.
title ZEBRA: Leveraging Model-Behavioral Knowledge for Zero-Annotation Preference Dataset Construction
topic Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2502.18744