R1 Zero GRPO Resources

A curated collection of resources related to R1 Zero and GRPO (Generative Reward-Penalty Optimization) implementations and research.

Official Implementations

• Open R1 - Official implementation by Hugging Face
• X-R1 - C++ implementation
• R1-Onevision - Vision-language model implementation
• Open R1 Multimodal - Multimodal implementation

Training Tools & Frameworks

• LLaMA Factory - Training framework with quickstart guide
• EasyR1 - Simplified R1 implementation
• VERL - Volcengine's implementation
• VLM-R1 - Vision-Language Model implementation

T2I-R1: Reinforcing Image Generation with Collaborative Semantic-level and Token-level CoT https://github.com/CaraJ7/T2I-R1

Thinking with Images

DeepEyes: Incentivizing "Thinking with Images" via Reinforcement Learning
paperprojectcodewechat

Pixel Reasoner: Incentivizing Pixel-Space Reasoning with Curiosity-Driven Reinforcement Learning projectpapercode

VisualToolAgent (VisTA): A Reinforcement Learning Framework for Visual‘Tool Selection

Thinking with Generated Images

•SAM-R1: Leveraging SAM for Reward Feedback in Multimodal Segmentation via Reinforcement Learning

•Omni-R1: Reinforcement Learning for Omnimodal Reasoning via Two-System Collaboration

•GRIT: Teaching MLLMs to Think with Images

•Chain-of-Focus: Adaptive Visual Search and Zooming for Multimodal Reasoning via RL

•Pixel Reasoner: Incentivizing Pixel-Space Reasoning with Curiosity-Driven Reinforcement Learning

•UniVG-R1: Reasoning Guided Universal Visual Grounding with Reinforcement Learning

•OpenThinkIMG: Learning to Think with Images via Visual Tool Reinforcement Learning

•Perception-R1: Pioneering Perception Policy with Reinforcement Learning

Documentation

• Swift GRPO Documentation - Official GRPO documentation

Additional Resources

• Awesome LLM Resources - Comprehensive collection of LLM resources

https://github.com/qiwang067/awesome-visual-rl

https://github.com/datawhalechina/easy-rl?tab=readme-ov-file 强化学习教程

训练框架

https://github.com/Simple-Efficient/RL-Factory

物理规则推理模型 https://github.com/nvidia-cosmos/cosmos-reason1