FROST: Filtering Reasoning Outliers with Attention for Efficient Reasoning

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Luo, Haozheng, Jiang, Zhuolin, Hasan, Md Zahid, Chen, Yan, Sarkar, Soumalya
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866912922616201216
author Luo, Haozheng
Jiang, Zhuolin
Hasan, Md Zahid
Chen, Yan
Sarkar, Soumalya
author_facet Luo, Haozheng
Jiang, Zhuolin
Hasan, Md Zahid
Chen, Yan
Sarkar, Soumalya
contents We propose FROST, an attention-aware method for efficient reasoning. Unlike traditional approaches, FROST leverages attention weights to prune uncritical reasoning paths, yielding shorter and more reliable reasoning trajectories. Methodologically, we introduce the concept of reasoning outliers and design an attention-based mechanism to remove them. Theoretically, FROST preserves and enhances the model's reasoning capacity while eliminating outliers at the sentence level. Empirically, we validate FROST on four benchmarks using two strong reasoning models (Phi-4-Reasoning and GPT-OSS-20B), outperforming state-of-the-art methods such as TALE and ThinkLess. Notably, FROST achieves an average 69.68% reduction in token usage and a 26.70% improvement in accuracy over the base model. Furthermore, in evaluations of attention outlier metrics, FROST reduces the maximum infinity norm by 15.97% and the average kurtosis by 91.09% compared to the base model. Code is available at https://github.com/robinzixuan/FROST
format Preprint
id arxiv_https___arxiv_org_abs_2601_19001
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle FROST: Filtering Reasoning Outliers with Attention for Efficient Reasoning
Luo, Haozheng
Jiang, Zhuolin
Hasan, Md Zahid
Chen, Yan
Sarkar, Soumalya
Computation and Language
Artificial Intelligence
Machine Learning
We propose FROST, an attention-aware method for efficient reasoning. Unlike traditional approaches, FROST leverages attention weights to prune uncritical reasoning paths, yielding shorter and more reliable reasoning trajectories. Methodologically, we introduce the concept of reasoning outliers and design an attention-based mechanism to remove them. Theoretically, FROST preserves and enhances the model's reasoning capacity while eliminating outliers at the sentence level. Empirically, we validate FROST on four benchmarks using two strong reasoning models (Phi-4-Reasoning and GPT-OSS-20B), outperforming state-of-the-art methods such as TALE and ThinkLess. Notably, FROST achieves an average 69.68% reduction in token usage and a 26.70% improvement in accuracy over the base model. Furthermore, in evaluations of attention outlier metrics, FROST reduces the maximum infinity norm by 15.97% and the average kurtosis by 91.09% compared to the base model. Code is available at https://github.com/robinzixuan/FROST
title FROST: Filtering Reasoning Outliers with Attention for Efficient Reasoning
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2601.19001