Rethinking the Trust Region in LLM Reinforcement Learning
Paper • 2602.04879 • Published • 37
Ruobing Xie
Ruobing-Xie
AI & ML interests
Recommender System; Large Language Model; Natural Language Processing; Information Retrieval
Organizations
None yet
upvoted 3
papers 3 months ago
Hybrid Linear Attention Done Right: Efficient Distillation and Effective Architectures for Extremely Long Contexts
Paper • 2601.22156 • Published • 14
Locate, Steer, and Improve: A Practical Survey of Actionable Mechanistic Interpretability in Large Language Models
Paper • 2601.14004 • Published • 47
upvoted an article 6 months ago
Article
Why Did MiniMax M2 End Up as a Full Attention Model?
•
80
upvoted a paper 7 months ago
upvoted a paper 9 months ago
upvoted 3
papers 11 months ago
Qwen3 Technical Report
Paper • 2505.09388 • Published • 339
The Climb Carves Wisdom Deeper Than the Summit: On the Noisy Rewards in Learning to Reason
Paper • 2505.22653 • Published • 43
The Entropy Mechanism of Reinforcement Learning for Reasoning Language Models
Paper • 2505.22617 • Published • 132
upvoted 8
papers about 1 year ago
Transformers without Normalization
Paper • 2503.10622 • Published • 172
MM-Eureka: Exploring Visual Aha Moment with Rule-based Large-scale Reinforcement Learning
Paper • 2503.07365 • Published • 61
HMoE: Heterogeneous Mixture of Experts for Language Modeling
Paper • 2408.10681 • Published • 10
Kimi k1.5: Scaling Reinforcement Learning with LLMs
Paper • 2501.12599 • Published • 128
DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning
Paper • 2501.12948 • Published • 447
Autonomy-of-Experts Models
Paper • 2501.13074 • Published • 44
Hunyuan3D 2.0: Scaling Diffusion Models for High Resolution Textured 3D Assets Generation
Paper • 2501.12202 • Published • 50
Towards Best Practices for Open Datasets for LLM Training
Paper • 2501.08365 • Published • 62
upvoted 3
papers over 1 year ago
Scaling Laws for Floating Point Quantization Training
Paper • 2501.02423 • Published • 26
PhD: A Prompted Visual Hallucination Evaluation Dataset
Paper • 2403.11116 • Published • 3
Hunyuan-Large: An Open-Source MoE Model with 52 Billion Activated Parameters by Tencent
Paper • 2411.02265 • Published • 25