Products

1000 个 DeepSeek 智能体一周写出 60 万行代码,PyBot 无训练对齐 Smollm 360M

Why we picked it

Hugging Face 团队拿 1000 个 DeepSeek Flash 智能体花一周写了 60 万行 Python,造出一个零权重、零训练的 PyBot,成绩对标 Smollm 360M。做工程和微调的朋友可以看看这个思路,它跟跑小模型的路线完全不同,仓库已开源,适合动手验证哪些任务用规则就够了。

Hugging Face 的 Leandro von Werra 发布了 PyBot,一个不含任何权重、不做训练的纯规则聊天机器人,全部代码达 60 万行 Python,却能对标上 360M 参数的 Smollm。项目的开发方式是亮点:由 1000 个 DeepSeek Flash 智能体组成的集群,只用一周时间就写完了整个代码库,仓库已开源在 huggingface/py。它的思路是把模型能力"编译"进硬编码规则里,用代码量换取推理时的零参数开销。

Body

Julien Chaumonde 借此重申他十年前的判断:软件工程是 AI 的子集——既然智能体集群能在一周产出人类团队数年规模的代码,传统工程活动的边界正在被重新定义。

这个实验的具体价值在于对比口径:PyBot 没有训练成本,却在能力上追平 Smollm 360M,说明在特定场景下"规则+海量生成代码"与小型模型各有取舍,开发者可以据此评估自己的任务到底需要模型还是需要工程化方案。

Original · Julien Chaumond

People used to call me crazy when I said this 10 years ago, but: Software engineering is a subset of AI Exhibit 42,729: Leandro von Werra @lvwerra PyBot: No weights. No training. A chatbot built with only simple rules and no AI, entirely written in 600k lines of Python code. PyBot matches Smollm with 360M parameters! How? A swarm of 1,000 DeepSeek Flash agents wrote the whole code in just a week. github.com/huggingface/py…

Industry Coordinates & Source Verification

This report is categorized under AITOP's "Products" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include DeepSeek, Smollm, Hugging Face, agents.

Source attribution: curated from Julien Chaumond (@julien_c).