verl-agent
verl-agent is an extension of veRL, designed for training LLM/VLM agents via RL. verl-agent is also the official code for paper "Group-in-Group Policy Optimization for LLM Agent Training"
Reinforcement-learning training for agents over long, multi-turn episodes.
You want a standalone framework rather than an extension of veRL.
About verl-agent
verl-agent is an extension of veRL, specifically designed for training large language model (LLM) agents via reinforcement learning (RL).
Unlike prior approaches that simply concatenate full interaction histories, verl-agent proposes step-independent multi-turn rollout mechanism, which allows for fully customizable per-step input structures, history management, and memory modules. This design makes verl-agent highly scalable for very long-horizon, multi-turn RL training (e.g., tasks in ALFWorld can require up to 50 steps to complete).
verl-agent provides a diverse set of RL algorithms (including our new algorithm GiGPO) and a rich suite of agent environments, enabling the development of reasoning agents in both visual and text-based tasks.
verl-agent is an open-source project written primarily in Python, with 2.3k stars on GitHub. It was last updated in June 2026.
pip3 install vllm==0.11.0
