Beyond the 80/20 Rule: High-Entropy Minority Tokens Drive Effective Reinforcement Learning for LLM Reasoning
Shenzhi Wang, Le Yu, Chang Gao, Chujie Zheng, Shixuan Liu, Rui Lu, Kai Dang, Xionghui Chen and 10 more
Who it is from
Alibaba Groupaffiliation declared by the authors, according to OpenAlex- Nanjing Universityaffiliation declared by the authors, according to OpenAlex
- National University of Defense Technologyaffiliation declared by the authors, according to OpenAlex
- Sichuan Universityaffiliation declared by the authors, according to OpenAlex
Other papers from the same organizations
- Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any ResolutionAlibaba Group · 6,193 citations
- Intern VL: Scaling up Vision Foundation Models and Aligning for Generic Visual-Linguistic TasksChinese Academy of Sciences · 3,956 citations
- VBench: Comprehensive Benchmark Suite for Video Generative ModelsNanjing University · 2,431 citations
- Qwen3 Embedding: Advancing Text Embedding and Reranking Through Foundation ModelsAlibaba Group · 2,422 citations
- MME: A Comprehensive Evaluation Benchmark for Multimodal Large Language ModelsBeijing Institute of Technology · 1,963 citations
- DeepSeekMoE: Towards Ultimate Expert Specialization in Mixture-of-Experts Language ModelsNanjing University · 1,603 citations