Guardado en:
Detalles Bibliográficos
Autores principales: Wu, Wei, Chen, Liyi, Xiao, Congxi, Wang, Tianfu, Wang, Qimeng, Lu, Chengqiang, Gao, Yan, Wu, Yi, Hu, Yao, Xiong, Hui
Formato: Preprint
Publicado: 2026
Materias:
Acceso en línea:https://arxiv.org/abs/2601.03969
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866910217873129472
author Wu, Wei
Chen, Liyi
Xiao, Congxi
Wang, Tianfu
Wang, Qimeng
Lu, Chengqiang
Gao, Yan
Wu, Yi
Hu, Yao
Xiong, Hui
author_facet Wu, Wei
Chen, Liyi
Xiao, Congxi
Wang, Tianfu
Wang, Qimeng
Lu, Chengqiang
Gao, Yan
Wu, Yi
Hu, Yao
Xiong, Hui
contents Large reasoning models enhanced by reinforcement learning with verifiable rewards have achieved significant performance gains by extending their chain-of-thought. However, this paradigm incurs substantial deployment costs as models often exhibit excessive verbosity on simple queries. Existing efficient reasoning methods relying on explicit length penalties often introduce optimization conflicts and leave the generative mechanisms driving overthinking largely unexamined. In this paper, we identify a phenomenon termed length shift where models increasingly generate unnecessary reasoning on trivial inputs during training. To address this, we introduce Dynamic Outlier Truncation (DOT), a training-time intervention that selectively suppresses redundant tokens. This method targets only the extreme tail of response lengths within fully correct rollout groups while preserving long-horizon reasoning capabilities for complex problems. To complement this intervention and ensure stable convergence, we further incorporate auxiliary KL regularization and predictive dynamic sampling. Experimental results across multiple model scales demonstrate that our approach significantly pushes the efficiency-performance Pareto frontier outward. Notably, on the AIME-24, our method reduces inference token usage by 78% while simultaneously increasing accuracy compared to the initial policy and surpassing state-of-the-art efficient reasoning methods.
format Preprint
id arxiv_https___arxiv_org_abs_2601_03969
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Anti-Length Shift: Dynamic Outlier Truncation for Training Efficient Reasoning Models
Wu, Wei
Chen, Liyi
Xiao, Congxi
Wang, Tianfu
Wang, Qimeng
Lu, Chengqiang
Gao, Yan
Wu, Yi
Hu, Yao
Xiong, Hui
Artificial Intelligence
Computation and Language
Large reasoning models enhanced by reinforcement learning with verifiable rewards have achieved significant performance gains by extending their chain-of-thought. However, this paradigm incurs substantial deployment costs as models often exhibit excessive verbosity on simple queries. Existing efficient reasoning methods relying on explicit length penalties often introduce optimization conflicts and leave the generative mechanisms driving overthinking largely unexamined. In this paper, we identify a phenomenon termed length shift where models increasingly generate unnecessary reasoning on trivial inputs during training. To address this, we introduce Dynamic Outlier Truncation (DOT), a training-time intervention that selectively suppresses redundant tokens. This method targets only the extreme tail of response lengths within fully correct rollout groups while preserving long-horizon reasoning capabilities for complex problems. To complement this intervention and ensure stable convergence, we further incorporate auxiliary KL regularization and predictive dynamic sampling. Experimental results across multiple model scales demonstrate that our approach significantly pushes the efficiency-performance Pareto frontier outward. Notably, on the AIME-24, our method reduces inference token usage by 78% while simultaneously increasing accuracy compared to the initial policy and surpassing state-of-the-art efficient reasoning methods.
title Anti-Length Shift: Dynamic Outlier Truncation for Training Efficient Reasoning Models
topic Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2601.03969