Enregistré dans:
Détails bibliographiques
Auteurs principaux: Vieira, Inacio, Castaldo, Antonio, O'Doherty, James, Castilho, Sheila
Format: Preprint
Publié: 2025
Sujets:
Accès en ligne:https://arxiv.org/abs/2510.27556
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866911243574444032
author Vieira, Inacio
Castaldo, Antonio
O'Doherty, James
Castilho, Sheila
author_facet Vieira, Inacio
Castaldo, Antonio
O'Doherty, James
Castilho, Sheila
contents LLMs often require adaptation to domain-specific requirements, a process that can be expensive when relying solely on SFT. We present an empirical study on applying CPO to simulate a post-editing workflow for data-efficient domain adaptation. Our approach synthesizes preference pairs by treating the base model's own raw output as the 'rejected' translation and the human-approved TM entry as the 'chosen' one. This method provides direct feedback on the model's current knowledge, guiding it to align with domain-specific standards. Experiments in English-Brazilian Portuguese and English-Korean show that, by using just 14.7k preference pairs, the model achieves performance close to that of a model trained on 160k+ samples with SFT, demonstrating significant data efficiency. Although we showcase its effectiveness in MT, this application of CPO naturally generalizes to other generative tasks where a model's initial drafts can serve as a contrastive signal against a golden reference.
format Preprint
id arxiv_https___arxiv_org_abs_2510_27556
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Data-Efficient Domain Adaptation for LLM-based MT using Contrastive Preference Optimization
Vieira, Inacio
Castaldo, Antonio
O'Doherty, James
Castilho, Sheila
Computation and Language
LLMs often require adaptation to domain-specific requirements, a process that can be expensive when relying solely on SFT. We present an empirical study on applying CPO to simulate a post-editing workflow for data-efficient domain adaptation. Our approach synthesizes preference pairs by treating the base model's own raw output as the 'rejected' translation and the human-approved TM entry as the 'chosen' one. This method provides direct feedback on the model's current knowledge, guiding it to align with domain-specific standards. Experiments in English-Brazilian Portuguese and English-Korean show that, by using just 14.7k preference pairs, the model achieves performance close to that of a model trained on 160k+ samples with SFT, demonstrating significant data efficiency. Although we showcase its effectiveness in MT, this application of CPO naturally generalizes to other generative tasks where a model's initial drafts can serve as a contrastive signal against a golden reference.
title Data-Efficient Domain Adaptation for LLM-based MT using Contrastive Preference Optimization
topic Computation and Language
url https://arxiv.org/abs/2510.27556