STROT 将 LLM 指令编译为本地运行的决策工作流

Amit Rath 的开源项目在设置步骤中使用大型模型，而在执行任务时使用小型本地模型，而其 Pac-Man 演示展示了这种划分的局限性。

By · Published · Updated

Primary source: GitHub

为何重要

STROT 将大型模型置于工作流设置中，并将重复的决策保留在本地，这可能会降低有界任务的成本和延迟。它自己的演示还表明了为什么编译和金标准测试不能替代在真实、不断变化的数据上进行的评估。

STROT compiles LLM instructions into locally run decision workflows — Amit Rath's open-source project puts a large model in the setup step and a small local model on the job, while its Pac-Man demo shows the limits of that split.

Amit Rath's STROT uses a large language model to turn a developer's instructions into a Python artifact, then runs routine classification and routing decisions with a smaller model on local hardware. Rath, listed as the project's author and maintainer on PyPI, has now carried the idea from a 2025 research framework into an open-source tool developers can install and test.…

When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.

Reader comments

Conversation for this story loads after sign-in.