微软发布 Agent Lightning v1.0：在真实部署环境中训练智能体
New Microsoft paper says agent training should happen inside the agent's normal…
New Microsoft paper says agent training should happen inside the agent's normal operating setup. Otherwise, you may be training something different from what you actually deploy.
微软最新论文指出，智能体训练应在智能体的正常运行环境中进行。否则，你训练的可能与实际部署的并不一致。
"In traditional agentic RL, the training engine owns the environment interaction loop. In harnessed agentic RL, the harness owns this loop, while the training engine observes only a sequence of LLM request-response pairs."
"在传统智能体强化学习中，训练引擎掌控环境交互循环。在受控智能体强化学习（harnessed agentic RL）中，Harness 掌控该循环，而训练引擎仅观察一系列 LLM 请求-响应对。"
Microsoft’s solution is Agent Lightning v1.0, a lightweight framework for harnessed agentic RL that trains an existing agent through its real deployment harness, instead of rebuilding the agent loop inside the RL trainer.
微软的解决方案是 Agent Lightning v1.0，这是一个用于受控智能体强化学习的轻量级框架，它通过智能体的真实部署 Harness 来训练现有智能体，而不是在 RL 训练器内部重建智能体循环。
Agent Lightning v1.0 sits between the agent harness and the model, recording the harness’s LLM calls and feeding them into the RL trainer without taking over the harness itself.
Agent Lightning v1.0 位于智能体 Harness 和模型之间，记录 Harness 的 LLM 调用并将其输入到 RL 训练器中，而不接管 Harness 本身。
So it makes RL work correctly with arbitrary deployment-time harnesses, including the messy cases where 1 rollout becomes multiple training samples.
因此，它使得 RL 能够与任意部署时的 Harness 正确协作，包括那些混乱的情况，例如一次 rollout 产生多个训练样本。
– arxiv. org/abs/2608.17528
– arxiv.org/abs/2608.17528
Title: "Agent Lightning v1.0: Towards Harnessed Agentic RL"
标题："Agent Lightning v1.0: Towards Harnessed Agentic RL"
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力