Cover Image

Beyond the 80/20 Rule: High-Entropy Minority Tokens Drive Effective Reinforcement Learning for LLM Reasoning

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Wang, Shenzhi, Yu, Le, Gao, Chang, Zheng, Chujie, Liu, Shixuan, Lu, Rui, Dang, Kai, Chen, Xionghui, Yang, Jianxin, Zhang, Zhenru, Liu, Yuqiong, Yang, An, Zhao, Andrew, Yue, Yang, Song, Shiji, Yu, Bowen, Huang, Gao, Lin, Junyang
Format:	Preprint
Published:	2025
Subjects:	Computation and Language Artificial Intelligence Machine Learning
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

HopChain: Multi-Hop Data Synthesis for Generalizable Vision-Language Reasoning
by: Wang, Shenzhi, et al.
Published: (2026)

Group Sequence Policy Optimization
by: Zheng, Chujie, et al.
Published: (2025)

Disentangling Reasoning Tokens and Boilerplate Tokens For Language Model Fine-tuning
by: Ye, Ziang, et al.
Published: (2024)

Revealing Behavioral Plasticity in Large Language Models: A Token-Conditional Perspective
by: Mao, Liyuan, et al.
Published: (2026)

Soft Adaptive Policy Optimization
by: Gao, Chang, et al.
Published: (2025)

Stabilizing Reinforcement Learning with LLMs: Formulation and Practices
by: Zheng, Chujie, et al.
Published: (2025)

ProcessBench: Identifying Process Errors in Mathematical Reasoning
by: Zheng, Chujie, et al.
Published: (2024)

The Lessons of Developing Process Reward Models in Mathematical Reasoning
by: Zhang, Zhenru, et al.
Published: (2025)

Outcome Accuracy is Not Enough: Aligning the Reasoning Process of Reward Models
by: Wang, Binghai, et al.
Published: (2026)

Beyond Turn Limits: Training Deep Search Agents with Dynamic Context Window
by: Tang, Qiaoyu, et al.
Published: (2025)

Efficient Reinforcement Learning with Semantic and Token Entropy for LLM Reasoning
by: Cao, Hongye, et al.
Published: (2025)

Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model?
by: Yue, Yang, et al.
Published: (2025)

PsychoGAT: A Novel Psychological Measurement Paradigm through Interactive Fiction Games with LLM Agents
by: Yang, Qisen, et al.
Published: (2024)

DeeR-VLA: Dynamic Inference of Multimodal Large Language Models for Efficient Robot Execution
by: Yue, Yang, et al.
Published: (2024)

Model Surgery: Modulating LLM's Behavior Via Simple Parameter Editing
by: Wang, Huanqian, et al.
Published: (2024)

WorldPM: Scaling Human Preference Modeling
by: Wang, Binghai, et al.
Published: (2025)

START: Self-taught Reasoner with Tools
by: Li, Chengpeng, et al.
Published: (2025)

From Reasoning Chains to Verifiable Subproblems: Curriculum Reinforcement Learning Enables Credit Assignment for LLM Reasoning
by: Jiang, Xitai, et al.
Published: (2026)

Leveraging Verifier-Based Reinforcement Learning in Image Editing
by: Guo, Hanzhong, et al.
Published: (2026)

Enhancing In-Context Learning Performance with just SVD-Based Weight Pruning: A Theoretical Perspective
by: Yao, Xinhao, et al.
Published: (2024)

CIDR interpolation: an enhanced SSA-based temporal filling framework for restoring continuity in downscaled GRACE(-FO) TWSA products
by: Gao, Yu, et al.
Published: (2025)

RuleReasoner: Reinforced Rule-based Reasoning via Domain-aware Dynamic Sampling
by: Liu, Yang, et al.
Published: (2025)

CatCode: A Comprehensive Evaluation Framework for LLMs On the Mixture of Code and Text
by: Lin, Zhenru, et al.
Published: (2024)

Qwen2.5-Math Technical Report: Toward Mathematical Expert Model via Self-Improvement
by: Yang, An, et al.
Published: (2024)

DocThinker: Explainable Multimodal Large Language Models with Rule-based Reinforcement Learning for Document Understanding
by: Yu, Wenwen, et al.
Published: (2025)

Rethinking Data Selection at Scale: Random Selection is Almost All You Need
by: Xia, Tingyu, et al.
Published: (2024)

DeepStack: Deeply Stacking Visual Tokens is Surprisingly Simple and Effective for LMMs
by: Meng, Lingchen, et al.
Published: (2024)

Towards Token-Level Text Anomaly Detection
by: Cao, Yang, et al.
Published: (2026)

DISTINGUIENDO SIMILITUDES Y DIFERENCIAS; COMBINANDO ORIENTE Y OCCIDENTE. PERSPECTIVAS PRESENTES Y FUTURAS DE LA CULTURA CHINA
by: Li Shenzhi
Published: (2012)

Effectively Compress KV Heads for LLM
by: Yu, Hao, et al.
Published: (2024)

Rule Learning for Knowledge Graph Reasoning under Agnostic Distribution Shift
by: Liu, Shixuan, et al.
Published: (2025)

ERPO: Token-Level Entropy-Regulated Policy Optimization for Large Reasoning Models
by: Yu, Song, et al.
Published: (2026)

TokenPure: Watermark Removal through Tokenized Appearance and Structural Guidance
by: Yang, Pei, et al.
Published: (2025)

Absolute Zero: Reinforced Self-play Reasoning with Zero Data
by: Zhao, Andrew, et al.
Published: (2025)

Beyond Human Preferences: Exploring Reinforcement Learning Trajectory Evaluation and Improvement through LLMs
by: Shen, Zichao, et al.
Published: (2024)

Qwen3 Technical Report
by: Yang, An, et al.
Published: (2025)

Emphasizing Crucial Features for Efficient Image Restoration
by: Gao, Hu, et al.
Published: (2024)

Language Models can Self-Lengthen to Generate Long Texts
by: Quan, Shanghaoran, et al.
Published: (2024)

EchoWorld: Learning Motion-Aware World Models for Echocardiography Probe Guidance
by: Yue, Yang, et al.
Published: (2025)

CheXWorld: Exploring Image World Modeling for Radiograph Representation Learning
by: Yue, Yang, et al.
Published: (2025)