Saved in:
Bibliographic Details
Main Authors: Liu, Tao, Wu, Taiqiang, Yang, Runming, Sun, Shaoning, Wang, Junjie, Yang, Yujiu
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2601.09195
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915982752088064
author Liu, Tao
Wu, Taiqiang
Yang, Runming
Sun, Shaoning
Wang, Junjie
Yang, Yujiu
author_facet Liu, Tao
Wu, Taiqiang
Yang, Runming
Sun, Shaoning
Wang, Junjie
Yang, Yujiu
contents Supervised fine-tuning (SFT) is a fundamental post-training strategy to align Large Language Models (LLMs) with human intent. However, traditional SFT often ignores the one-to-many nature of language by forcing alignment with a single reference answer, leading to the model overfitting to non-core expressions. Although our empirical analysis suggests that introducing multiple reference answers can mitigate this issue, the prohibitive data and computational costs necessitate a strategic shift: prioritizing the mitigation of single-reference overfitting over the costly pursuit of answer diversity. To achieve this, we reveal the intrinsic connection between token probability and semantic importance: high-probability tokens carry the core logical framework, while low-probability tokens are mostly replaceable expressions. Based on this insight, we propose ProFit, which selectively masks low-probability tokens to prevent surface-level overfitting. Extensive experiments confirm that ProFit consistently outperforms traditional SFT baselines on general reasoning and mathematical benchmarks.
format Preprint
id arxiv_https___arxiv_org_abs_2601_09195
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle ProFit: Leveraging High-Value Signals in SFT via Probability-Guided Token Selection
Liu, Tao
Wu, Taiqiang
Yang, Runming
Sun, Shaoning
Wang, Junjie
Yang, Yujiu
Computation and Language
Artificial Intelligence
Supervised fine-tuning (SFT) is a fundamental post-training strategy to align Large Language Models (LLMs) with human intent. However, traditional SFT often ignores the one-to-many nature of language by forcing alignment with a single reference answer, leading to the model overfitting to non-core expressions. Although our empirical analysis suggests that introducing multiple reference answers can mitigate this issue, the prohibitive data and computational costs necessitate a strategic shift: prioritizing the mitigation of single-reference overfitting over the costly pursuit of answer diversity. To achieve this, we reveal the intrinsic connection between token probability and semantic importance: high-probability tokens carry the core logical framework, while low-probability tokens are mostly replaceable expressions. Based on this insight, we propose ProFit, which selectively masks low-probability tokens to prevent surface-level overfitting. Extensive experiments confirm that ProFit consistently outperforms traditional SFT baselines on general reasoning and mathematical benchmarks.
title ProFit: Leveraging High-Value Signals in SFT via Probability-Guided Token Selection
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2601.09195