verl-project/verl

Verl’s **HybridFlow** is an open‑source reinforcement‑learning post‑training framework that lets developers fine‑tune, adapt, and evaluate pretrained RL agents