Microsoft Research · Zhiyuan He, Yuqing Yang · 26 天前精选模型评分73
Microsoft Research Asia 开源 Agent Lightning v1.0，约 3500 行代码的智能体强化学习框架
Agent Lightning v1.0: A 3,500-Line Lightweight Agentic RL Framework for Training Agents with Real Harnesses
导读
Microsoft Research Asia 开源 Agent Lightning v1.0，这是一个约 3500 行代码的轻量智能体强化学习框架，让部署时使用的智能体框架直接参与训练，无需在训练框架里重写智能体。
推荐理由
框架把部署用的智能体直接接入强化学习训练，并给出约 6000 条样本提升 14.6 个百分点的完整流程，可据此判断改造成本。
分类置信度
高
- 官方或准官方来源，且无冲突类别。
- 正文、可信日期、主分类和证据摘录完整。
来源级别：官方一手
事件类型：产品更新
证据摘录
- “To address this, researchers at Microsoft Research Asia have introduced the Harnessed Agentic RL training paradigm and open-sourced a fully rebuilt Agent Lightning v1.0 (opens in new tab) .”
- “agents run as standard Kubernetes jobs on self-managed clusters, cloud Kubernetes, or local infrastructure, with no dependency on paid commercial sandbox services.”
- “using only about 6,000 training samples based on open sourced dataset.”
来源：Microsoft Research · microsoft.com