Cover Image

Two Experts Are All You Need for Steering Thinking: Reinforcing Cognitive Effort in MoE Reasoning Models Without Additional Training

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Wang, Mengru, Chen, Xingyu, Wang, Yue, He, Zhiwei, Xu, Jiahao, Liang, Tian, Liu, Qiuzhi, Yao, Yunzhi, Wang, Wenxuan, Ma, Ruotian, Mi, Haitao, Zhang, Ningyu, Tu, Zhaopeng, Li, Xiaolong, Yu, Dong
Format:	Preprint
Published:	2025
Subjects:	Artificial Intelligence Computation and Language Computer Vision and Pattern Recognition Information Retrieval Machine Learning
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

The First Few Tokens Are All You Need: An Efficient and Effective Unsupervised Prefix Fine-Tuning Method for Reasoning Models
by: Ji, Ke, et al.
Published: (2025)

Do NOT Think That Much for 2+3=? On the Overthinking of o1-Like LLMs
by: Chen, Xingyu, et al.
Published: (2024)

Beyond Prompt Engineering: Robust Behavior Control in LLMs via Steering Target Atoms
by: Wang, Mengru, et al.
Published: (2025)

Dancing with Critiques: Enhancing LLM Reasoning with Stepwise Natural Language Self-Critique
by: Li, Yansi, et al.
Published: (2025)

MoE Lens -- An Expert Is All You Need
by: Chaudhari, Marmik, et al.
Published: (2026)

DeepMath-103K: A Large-Scale, Challenging, Decontaminated, and Verifiable Mathematical Dataset for Advancing Reasoning
by: He, Zhiwei, et al.
Published: (2025)

Trust, But Verify: A Self-Verification Approach to Reinforcement Learning with Verifiable Rewards
by: Liu, Xiaoyuan, et al.
Published: (2025)

Too Good to be Bad: On the Failure of LLMs to Role-Play Villains
by: Yi, Zihao, et al.
Published: (2025)

Thoughts Are All Over the Place: On the Underthinking of o1-Like LLMs
by: Wang, Yue, et al.
Published: (2025)

Steering MoE LLMs via Expert (De)Activation
by: Fayyaz, Mohsen, et al.
Published: (2025)

Insight Over Sight: Exploring the Vision-Knowledge Conflicts in Multimodal LLMs
by: Liu, Xiaoyuan, et al.
Published: (2024)

DeepTheorem: Advancing LLM Reasoning for Theorem Proving Through Natural Language and Reinforcement Learning
by: Zhang, Ziyin, et al.
Published: (2025)

EasyEdit2: An Easy-to-use Steering Framework for Editing Large Language Models
by: Xu, Ziwen, et al.
Published: (2025)

Chain-of-Jailbreak Attack for Image Generation Models via Editing Step by Step
by: Wang, Wenxuan, et al.
Published: (2024)

Social Welfare Function Leaderboard: When LLM Agents Allocate Social Welfare
by: Shi, Zhengliang, et al.
Published: (2025)

Unveiling the Pitfalls of Knowledge Editing for Large Language Models
by: Li, Zhoubo, et al.
Published: (2023)

Steer-MoE: Efficient Audio-Language Alignment with a Mixture-of-Experts Steering Module
by: Feng, Ruitao, et al.
Published: (2025)

Unveiling and Consulting Core Experts in Retrieval-Augmented MoE-based LLMs
by: Zhou, Xin, et al.
Published: (2024)

The Hunger Game Debate: On the Emergence of Over-Competition in Multi-Agent Systems
by: Ma, Xinbei, et al.
Published: (2025)

Draft Model Knows When to Stop: Self-Verification Speculative Decoding for Long-Form Generation
by: Zhang, Ziyin, et al.
Published: (2024)

OD-MoE: On-Demand Expert Loading for Cacheless Edge-Distributed MoE Inference
by: Wang, Liujianfu, et al.
Published: (2025)

Refuse Whenever You Feel Unsafe: Improving Safety in LLMs via Decoupled Refusal Training
by: Yuan, Youliang, et al.
Published: (2024)

Think Fast and Slow: Step-Level Cognitive Depth Adaptation for LLM Agents
by: Yang, Ruihan, et al.
Published: (2026)

Knowledge Circuits in Pretrained Transformers
by: Yao, Yunzhi, et al.
Published: (2024)

From Data to Behavior: Predicting Unintended Model Behaviors Before Training
by: Wang, Mengru, et al.
Published: (2026)

MoE-Compression: How the Compression Error of Experts Affects the Inference Accuracy of MoE Model?
by: Ma, Songkai, et al.
Published: (2025)

BLR-MoE: Boosted Language-Routing Mixture of Experts for Domain-Robust Multilingual E2E ASR
by: Ma, Guodong, et al.
Published: (2025)

Why Steering Works: Toward a Unified View of Language Model Parameter Dynamics
by: Xu, Ziwen, et al.
Published: (2026)

Harder Tasks Need More Experts: Dynamic Routing in MoE Models
by: Huang, Quzhe, et al.
Published: (2024)

Effective MoE-based LLM Compression by Exploiting Heterogeneous Inter-Group Experts Routing Frequency and Information Density
by: Mi, Zhendong, et al.
Published: (2026)

Finding Fantastic Experts in MoEs: A Unified Study for Expert Dropping Strategies and Observations
by: Jaiswal, Ajay, et al.
Published: (2025)

Elastic MoE: Unlocking the Inference-Time Scalability of Mixture-of-Experts
by: Gu, Naibin, et al.
Published: (2025)

Guiding the Experts: Semantic Priors for Efficient and Focused MoE Routing
by: Min, Chengxi, et al.
Published: (2025)

Pangu Pro MoE: Mixture of Grouped Experts for Efficient Sparsity
by: Tang, Yehui, et al.
Published: (2025)

Unchosen Experts Can Contribute Too: Unleashing MoE Models' Power by Self-Contrast
by: Shi, Chufan, et al.
Published: (2024)

MoE-Prism: Disentangling Monolithic Experts for Elastic MoE Services via Model-System Co-Designs
by: Xia, Xinfeng, et al.
Published: (2025)

SPC: Evolving Self-Play Critic via Adversarial Games for LLM Reasoning
by: Chen, Jiaqi, et al.
Published: (2025)

ECG-MoE: Mixture-of-Expert Electrocardiogram Foundation Model
by: Xu, Yuhao, et al.
Published: (2026)

Janus: Disaggregating Attention and Experts for Scalable MoE Inference
by: Zhang, Zhexiang, et al.
Published: (2025)

Cross-Platform Fused MoE Dispatch in Triton: Portable Expert Routing Without CUDA
by: Mitra, Subhadip
Published: (2026)