Advanced Research

• Beyond the 80/20 Rule: High-Entropy Minority Tokens Drive Effective Reinforcement Learning for LLM Reasoning
• Clipped Importance Sampling Policy Optimization (CISPO)
• DAPO: An Open-Source LLM Reinforcement Learning System at Scale
• DeepEyes: Incentivizing “Thinking with Images” via Reinforcement Learning
• FIPO: Future-KL Influenced Policy Optimization
• Group Sequence Policy Optimization (GSPO)
• On-Policy RL Meets Off-Policy Experts: Harmonizing SFT and RL via Dynamic Weighting (CHORD)
• REINFORCE Leave-One-Out (RLOO)
• REINFORCE++: An Efficient RLHF Algorithm with Robustness to Both Prompt and Reward Models
• Rewards as Labels: Revisiting RLVR from a Classification Perspective
• Router Replay (R2/R3)
• Soft Adaptive Policy Optimization (SAPO)
• Training-Inference-Mismatch
• TreePO: Bridging the Gap of Policy Optimization and Efficacy and Inference Efficiency with Heuristic Tree-based Modeling