// ⚖ Alignment & Preference Optimization
DAPO
A set of GRPO fixes and an open system for long-CoT RL.
2025 Research frontier core
IN PLAIN WORDS
Introduces Clip-Higher, dynamic sampling, token-level loss and overlong reward shaping to fix entropy collapse, reaching 50 on AIME 2024 with Qwen2.5-32B.
Used in: Open reasoning RL (DAPO recipe ships in verl); Clip-Higher and dynamic sampling are widely reused by later GRPO variants
01Original paper
DAPO: An Open-Source LLM Reinforcement Learning System at Scale
Qiying Yu, Zheng Zhang, Ruofei Zhu, et al. · 2025
arXiv ↗ ✓ Verified · arXiv API
02Where it sits
Learn the left column first; the right column builds on this one.
Learn first
Then you can learn
- Nothing in the atlas builds on it yet
Related:RL with Verifiable Rewards (RLVR)
Left: prerequisites · right: builds on this · dashed: related. Full map →
03Learning resources
Official docs, explainers, courses and reference code. All links checked.
04Where the course teaches it
On learning paths:Post-training & Reasoning Models
For AI / data
- Verification: verified, checked 2026-10-21.
- JSON: /api/learn/algorithms.json (id:
dapo)