RAGEN (LLM Agent Training System)

What is the LLM agent training system RAGEN?

RAGEN is a modular system designed for training and evaluating large language model (LLM) agents using multi-turn reinforcement learning. It introduces a reinforcement learning algorithm called StarPO (state-thinking-action-reward policy-optimization) that incorporates the agent's internal reasoning processes into the learning framework. RAGEN aims to enhance the stability and broad applicability of LLM agents, enabling them to perform complex tasks through iterative learning and adaptation (paper).