Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
RAGEN leverages reinforcement learning to train LLM reasoning agents in interactive, stochastic environments.
| Date | Stars |
|---|---|
| 2026-07-31 | 2760 |
| 2026-08-02 | 2761 |
| 2026-08-03 | 2761 |
| 2026-08-04 | 2765 |
| 2026-08-05 | 2766 |
| 2026-08-06 | 2766 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
<h1 align="center">RAGEN: Training Agents by Reinforcing Reasoning</h1> <h3 align="center"><em>Diagnose agent failure modes. Make your RL training better.</em></h3> <p align="center"><img src="public/ragen_logo.jpeg" width="300px" alt="RAGEN icon" /></p> <p align="center"> <strong>RAGEN</strong> (<b>R</b>easoning <b>AGEN</b>T) is a flexible RL framework for training reasoning agents. </p> <p align="center"> We develop <strong>diagnostics to understand <i>how</i> agent RL training works </strong>, and how to fix hidden issues. </p> <p align="center"> <a href="https://arxiv.org/abs/2604.06268"><img src="https://img.shields.io/badge/📄_V2_Paper-DC143C?style=for-the-badge&logoColor=white" alt="V2 Paper"></a> <a href="https://arxiv.org/abs/2504.20073"><img src="https://img.shields.io/badge/📄_v1_Paper-FF8C00?style=for-the-badge&logoColor=white" alt="v1 Paper"></a> <a href="https://ragen-ai.github.io/"><img src="https://img.shields.io/badge/📝_HomePage-FF5722?style=for-the-badge&logoColor=white" alt="Blog"></a> <!-- <a href="https://ragen-doc.readthedocs.io/"><img src="https://img.shields.io/badge/📚_Documentation-4285F4?style=for-the-badge&logoColor=white" alt="Documentation"></a> --> <a href="https://x.com/wzihanw/status/1915052871474712858"><img src="https://img.shields.io/badge/🔍_Post-34A853?style=for-the-badge&logoColor=white" alt="Post"></a> <a href="https://api.wandb.ai/links/zihanwang-ai-northwestern-university/a8er8l7b"><img src="https://img.shields.io/badge/🧪_Experiment_Log-AB47BC?style=for-the-badge&logoColor=white" alt="Experiment Log"></a> </p> > **Looking for the V1 README?** Please take a look [here](docs/readme_v1.md). ## News - **2026.3.12.** We are excited to release <font color="#DC143C">RAGEN-2</font>! We introduce a systematic study of reasoning collapse in agent RL and lightweight interventions for stable training. See the [<font color="#DC143C">v2 paper</font>](https://ragen-ai.github.io/v2). - **2025.4.20.** RAGEN V1 [paper](https://arxiv.org/abs/2504.20073) published on arXiv. - **2025.1.27.** Initial RAGEN release. [Post](https://x.com/wzihanw/status/1884092805598826609). ## About RAGEN is built around **StarPO** (**S**tate-**T**hinking-**A**ctions-**R**eward **P**olicy **O**ptimization), a unified RL framework for training multi-turn, trajectory-level agents with flexible control over reasoning processes, reward assignment mechanisms, and prompt-rollout structures. **RAGEN is flexible with:** - **StarPO framework.** Unified optimization for multi-turn agents, supporting both trajectory-level and turn-wise training. - **10 built-in environments.** Sokoban, FrozenLake, WebShop, DeepCoder, SearchQA, Lean, Bandit, Countdown, MetaMathQA, Sudoku. - **Gym-compatible interface.** Easy to add custom environments. **<font color="#DC143C">RAGEN-2</font> additionally introduces:** - **SNR-Adaptive Filtering (<font color="#DC143C">V2</font>).** Lightweight rollout filtering based on reward variance to mitigate noisy gradient updates. - **Reasoning collapse diagnostics (<font color="#DC143C">V2</font>).** Mutual information proxy metrics to detect and monitor template collapse during training. ## Algorithm ### StarPO: Reinforcing Reasoning via Trajectory-Level Optimization <p align="center"><img src="public/starpo_logo.png" width="800px" alt="StarPO Framework" /></p> <p align="center" style="font-size: 16px; max-width: 800px; margin: 0 auto;"> The StarPO (State-Thinking-Action-Reward Policy Optimization) framework with two interleaved stages: <b>rollout stage</b> and <b>update stage</b>. The LLM generates reasoning-guided actions to interact with the environment, collecting trajectory-level rewards to jointly optimize reasoning and action strategies. </p> **MDP Formulation.** Agent-environment interactions are formulated as Markov Decision Processes (MDPs) where states and actions are token sequences, allowing LLMs to reason over environment dynamics. The objective is to maximize expe
Excerpt of 14,736 characters
Read on GitHub
Zihan Wang
335
92
48
39
28
15
7
4
4
3
3
2
1
PanAndy · alibaba-inc · China
1
1
1
Harryis Wang · Tsinghua
1
1
1
1
Would you bet a product on this? Bounded 0–100 and slow moving.
matched fp:04ffede8afbc712f, desc:reinforcement learning