BigHugger
GH Repository · OpenPipe

ART

Agent Reinforcement Trainer: train multi-step agents for real-world tasks using GRPO. Give your agents on-the-job training. Reinforcement learning for Qwen3.6, GPT-OSS, Llama, and more!

stars
10,721
30-day movement
+93/day
Related entries
60
Connections
3
Pythonqwen3agent-appqwenrlgrpoagentic-aiagentreinforcement-learningpython/uvpythonlorallms

ART (Agent Reinforcement Trainer) is an open-source Python toolchain from OpenPipe for training multi-step agents on real-world tasks using GRPO reinforcement learning. It targets LLMs such as Qwen3, GPT-OSS, and Llama, using LoRA-based training, and ships Claude directory and claude.md config files.

You want to fine-tune agents with reinforcement learning rather than prompt engineering alone.

Use it to

  • Train multi-step agents with GRPO
  • Fine-tune Qwen3 or Llama via LoRA
  • Apply on-the-job RL to real-world tasks
  • Set up training runs in a Python/uv toolchain

For Developers training LLM-powered agents with reinforcement learning

Role
agent-app
Language
Python
Licence
Apache-2.0
Forks
983
Open issues
67
Last push
2026-09-15
Latest release
v0.1.9 · 2025-07-11
topicsreinforcement-learninggrpoagentslorallmsqwen