让大模型智能体与数据动态互演，自动发现并解决复杂任务中的失败模式。
CoEvolve: Training LLM Agents via Agent-Data Mutual Evolution
- 通过回放轨迹提取遗忘与不确定信号，驱动任务自动生成。
- 在多个环境上实现15.58%~19.43%的性能提升，效果稳定显著。
- 适合需要持续进化能力的智能体训练场景，如复杂交互系统。
大型语言模型智能体的强化学习通常依赖静态数据分布，无法随智能体行为演变而调整，导致复杂环境交互覆盖不足。为此，我们提出CoEvolve框架，实现智能体与数据的协同演化。该框架从回放轨迹中提取遗忘和不确定性等反馈信号，识别易出错的交互模式，并据此生成新任务。这些任务经环境验证后用于更新数据分布，实现智能体与数据的联合适应。在AppWorld和BFCL基准上，基于Qwen2.5-7B、Qwen3-4B和Qwen3-30B-A3B的实验表明，相比强基线模型，性能分别提升19.43%、15.58%和18.14%，效果显著且一致。
原文摘要 · Abstract (English)
Reinforcement learning for LLM agents is typically conducted on a static data distribution, which fails to adapt to the agent's evolving behavior and leads to poor coverage of complex environment interactions. To address these challenges, we propose CoEvolve, an agent-data mutual evolution framework that enables LLM agents to improve through closed-loop, interaction-driven training. Specifically, CoEvolve extracts feedback signals such as forgetting and uncertainty from rollout trajectories to identify failure-prone interaction patterns, and utilizes them to guide LLM-based task synthesis. The synthesized tasks are validated through environment interaction and utilized to update the data distribution, enabling joint adaptation of the agent and its data. Extensive experiments on AppWorld and BFCL across Qwen2.5-7B, Qwen3-4B, and Qwen3-30B-A3B demonstrate consistent and significant improvements over strong base models, yielding absolute gains of 19.43%, 15.58%, and 18.14%, respectively.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。