Enregistré dans:
Détails bibliographiques
Auteurs principaux: Zheng, Tong, Huang, Chengsong, Dai, Runpeng, He, Yun, Liu, Rui, Ni, Xin, Bao, Huiwen, Wang, Kaishen, Zhu, Hongtu, Huang, Jiaxin, Huang, Furong, Huang, Heng
Format: Preprint
Publié: 2026
Sujets:
Accès en ligne:https://arxiv.org/abs/2602.03845
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866912895760072704
author Zheng, Tong
Huang, Chengsong
Dai, Runpeng
He, Yun
Liu, Rui
Ni, Xin
Bao, Huiwen
Wang, Kaishen
Zhu, Hongtu
Huang, Jiaxin
Huang, Furong
Huang, Heng
author_facet Zheng, Tong
Huang, Chengsong
Dai, Runpeng
He, Yun
Liu, Rui
Ni, Xin
Bao, Huiwen
Wang, Kaishen
Zhu, Hongtu
Huang, Jiaxin
Huang, Furong
Huang, Heng
contents Parallel thinking has emerged as a promising paradigm for reasoning, yet it imposes significant computational burdens. Existing efficiency methods primarily rely on local, per-trajectory signals and lack principled mechanisms to exploit global dynamics across parallel branches. We introduce 2D probing, an interface that exposes the width-depth dynamics of parallel thinking by periodically eliciting intermediate answers from all branches. Our analysis reveals three key insights: non-monotonic scaling across width-depth allocations, heterogeneous reasoning branch lengths, and early stabilization of global consensus. Guided by these insights, we introduce $\textbf{Parallel-Probe}$, a training-free controller designed to optimize online parallel thinking. Parallel-Probe employs consensus-based early stopping to regulate reasoning depth and deviation-based branch pruning to dynamically adjust width. Extensive experiments across three benchmarks and multiple models demonstrate that Parallel-Probe establishes a superior Pareto frontier for test-time scaling. Compared to standard majority voting, it reduces sequential tokens by up to $\textbf{35.8}$% and total token cost by over $\textbf{25.8}$% while maintaining competitive accuracy.
format Preprint
id arxiv_https___arxiv_org_abs_2602_03845
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Parallel-Probe: Towards Efficient Parallel Thinking via 2D Probing
Zheng, Tong
Huang, Chengsong
Dai, Runpeng
He, Yun
Liu, Rui
Ni, Xin
Bao, Huiwen
Wang, Kaishen
Zhu, Hongtu
Huang, Jiaxin
Huang, Furong
Huang, Heng
Computation and Language
Parallel thinking has emerged as a promising paradigm for reasoning, yet it imposes significant computational burdens. Existing efficiency methods primarily rely on local, per-trajectory signals and lack principled mechanisms to exploit global dynamics across parallel branches. We introduce 2D probing, an interface that exposes the width-depth dynamics of parallel thinking by periodically eliciting intermediate answers from all branches. Our analysis reveals three key insights: non-monotonic scaling across width-depth allocations, heterogeneous reasoning branch lengths, and early stabilization of global consensus. Guided by these insights, we introduce $\textbf{Parallel-Probe}$, a training-free controller designed to optimize online parallel thinking. Parallel-Probe employs consensus-based early stopping to regulate reasoning depth and deviation-based branch pruning to dynamically adjust width. Extensive experiments across three benchmarks and multiple models demonstrate that Parallel-Probe establishes a superior Pareto frontier for test-time scaling. Compared to standard majority voting, it reduces sequential tokens by up to $\textbf{35.8}$% and total token cost by over $\textbf{25.8}$% while maintaining competitive accuracy.
title Parallel-Probe: Towards Efficient Parallel Thinking via 2D Probing
topic Computation and Language
url https://arxiv.org/abs/2602.03845