Proximal Supervised Fine-Tuning

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Zhu, Wenhong, Xie, Ruobing, Wang, Rui, Sun, Xingwu, Wang, Di, Liu, Pengfei
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911585708015616
author Zhu, Wenhong
Xie, Ruobing
Wang, Rui
Sun, Xingwu
Wang, Di
Liu, Pengfei
author_facet Zhu, Wenhong
Xie, Ruobing
Wang, Rui
Sun, Xingwu
Wang, Di
Liu, Pengfei
contents Supervised fine-tuning (SFT) of foundation models often leads to poor generalization, where prior capabilities deteriorate after tuning on new tasks or domains. Inspired by trust-region policy optimization (TRPO) and proximal policy optimization (PPO) in reinforcement learning (RL), we propose Proximal SFT (PSFT). This fine-tuning objective incorporates the benefits of trust-region, effectively constraining policy drift during SFT while maintaining competitive tuning. By viewing SFT as a special case of policy gradient methods with constant positive advantages, we derive PSFT that stabilizes optimization and leads to generalization, while leaving room for further optimization in subsequent post-training stages. Experiments across mathematical and human-value domains show that PSFT matches SFT in-domain, outperforms it in out-of-domain generalization, remains stable under prolonged training without causing entropy collapse, and provides a stronger foundation for the subsequent optimization.
format Preprint
id arxiv_https___arxiv_org_abs_2508_17784
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Proximal Supervised Fine-Tuning
Zhu, Wenhong
Xie, Ruobing
Wang, Rui
Sun, Xingwu
Wang, Di
Liu, Pengfei
Machine Learning
Artificial Intelligence
Computation and Language
Supervised fine-tuning (SFT) of foundation models often leads to poor generalization, where prior capabilities deteriorate after tuning on new tasks or domains. Inspired by trust-region policy optimization (TRPO) and proximal policy optimization (PPO) in reinforcement learning (RL), we propose Proximal SFT (PSFT). This fine-tuning objective incorporates the benefits of trust-region, effectively constraining policy drift during SFT while maintaining competitive tuning. By viewing SFT as a special case of policy gradient methods with constant positive advantages, we derive PSFT that stabilizes optimization and leads to generalization, while leaving room for further optimization in subsequent post-training stages. Experiments across mathematical and human-value domains show that PSFT matches SFT in-domain, outperforms it in out-of-domain generalization, remains stable under prolonged training without causing entropy collapse, and provides a stronger foundation for the subsequent optimization.
title Proximal Supervised Fine-Tuning
topic Machine Learning
Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2508.17784