微软Agent Lightning v1.0：将任意harness接入RL训练
Very interesting new work from Microsoft.
Very interesting new work from Microsoft.
微软的新工作非常有趣。
(bookmark it)
（收藏它）
This work is related to this emerging theme of leveraging harnesses for model post-training.
这项工作与利用框架进行模型后训练这一新兴主题相关。
Modern agents run inside a harness that owns tools, context, and control flow. When you train them, the harness owns the environment loop and the trainer only sees LLM request and response pairs.
现代代理在拥有工具、上下文和控制流的框架内运行。当你训练它们时，框架拥有环境循环，训练器只看到LLM请求和响应对。
How it works.
工作原理。
Agent Lightning v1.0 connects any harness to RL through an endpoint proxy in about 3,500 lines, then works through what breaks in that setup, retokenization, sample merging, advantage calculation, loss normalization, and backend scheduling.
Agent Lightning v1.0通过端点代理将任何框架连接到强化学习，大约3500行代码，然后解决该设置中出现的问题，包括重新分词、样本合并、优势计算、损失归一化和后端调度。
Using 6K training examples and modest compute, it moves Qwen3.5-9B on SWE-bench Verified from 41.8% to 56.4%.
使用6K训练示例和适度的计算资源，它在SWE-bench Verified上将Qwen3.5-9B从41.8%提升到56.4%。
Paper: https://arxiv.org/abs/2608.17528
论文：https://arxiv.org/abs/2608.17528
Track more trending AI papers in our academy: https://academy.dair.ai/
在我们的学院追踪更多热门AI论文：https://academy.dair.ai/
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