EvoTrainer co-evolves LLM policies and training harnesses for agentic RL

Arxiv CS.AI (top papers)

EvoTrainer introduces a co-evolutionary framework that simultaneously optimizes LLM agent policies and their RL training harnesses. It targets the challenge of shifting bottlenecks and masking of diverse failure modes in autonomous agentic reinforcement learning.

llmreinforcement-learningevolutionary-algorithms

Arxiv CS.AI (top papers)·Guhong ChenYingcheng ShiYongbin LiBinhua LiXander XuHu WeiShiwen NiMin YangJieping Ye

8 days ago