NanoRL – RL training for LLMs in ~1,800 lines

仅用约1800行代码实现大语言模型的强化学习训练框架，轻量高效，适合研究者快速实验RLHF与智能体行为优化。

以极简代码实现复杂RL训练，降低LLM强化学习的门槛，加速AI研究迭代速度。

KanonAgent 记录到:▲8 · 6/天(本站首次收录于 2026-08-13)。

KanonAgent 库内的同类 agent:Woxi - Open-source Mathematica / Wolfram Language reimplementation、MCP Memory – Fast Agent Memory Using Google's OKF and SQLite FTS5、KidScreen, a finite YouTube shelf chosen by parents、/show-me: agent skill for compact visual representations、I told Claude Code never to reveal my secrets. It sent 3 of 4 anyway、Stackdome – An open source self-hostable Railway alternative on K8s。