RLinf/RLinf

Health v2 · maintenance only

Adoption not part of the score

4655 stars · 685 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

RLinf is an open-source, flexible and scalable reinforcement learning training infrastructure for embodied AI (vision-language-action models) and agentic AI. It supports distributed RL training workflows like GRPO across multiple GPU accelerators including NVIDIA, AMD ROCm, Huawei Ascend, and Moore Threads.

Use cases

• train vision-language-action models with reinforcement learning
• run GRPO training for large language models
• RL training for robot manipulation policies on LIBERO
• reinforcement learning for video generation models
• train agentic AI models with RL post-training
• scalable distributed RL infrastructure for embodied AI research

When to choose

• you need a scalable RL training backbone for VLA or agentic models
• you want multi-vendor accelerator support (NVIDIA, AMD, Ascend, MUSA)
• you need GRPO-style post-training for LLMs or diffusion models
• you want an RL infrastructure officially integrated with Isaac Lab

When to avoid

• you only need simple single-GPU RL experiments with lightweight libraries like Gym or Stable-Baselines3
• you need supervised fine-tuning only without reinforcement learning
• you work outside Python-based ML training pipelines

Facets

framework · maturity active

reinforcement-learning llm-training machine-learning gpu-computing workflow-automation reinforcement-learning machine-learning robotics artificial-intelligence gpu-computing python rl-infrastructure embodied-ai agentic-ai vla-training grpo rlhf vision-language-action distributed-training robot-learning video-generation gpu linux docker

Member repositories

For agents

markdown · JSON · MCP: product_card(name="RLinf/RLinf")