
RLinf
RLinf is a flexible and scalable open-source infrastructure designed for post-training foundation models (LLMs, VLMs, VLAs) via reinforcement learning.
Stars: 423

RLinf is a flexible and scalable open-source infrastructure designed for post-training foundation models via reinforcement learning. It provides a robust backbone for next-generation training, supporting open-ended learning, continuous generalization, and limitless possibilities in intelligence development. The tool offers unique features like Macro-to-Micro Flow, flexible execution modes, auto-scheduling strategy, embodied agent support, and fast adaptation for mainstream VLA models. RLinf is fast with hybrid mode and automatic online scaling strategy, achieving significant throughput improvement and efficiency. It is also flexible and easy to use with multiple backend integrations, adaptive communication, and built-in support for popular RL methods. The roadmap includes system-level enhancements and application-level extensions to support various training scenarios and models. Users can get started with complete documentation, quickstart guides, key design principles, example gallery, advanced features, and guidelines for extending the framework. Contributions are welcome, and users are encouraged to cite the GitHub repository and acknowledge the broader open-source community.
README:
RLinf is a flexible and scalable open-source infrastructure designed for post-training foundation models via reinforcement learning. The 'inf' in RLinf stands for Infrastructure
, highlighting its role as a robust backbone for next-generation training. It also stands for Infinite
, symbolizing the system’s support for open-ended learning, continuous generalization, and limitless possibilities in intelligence development.
- [2025/08] RLinf is open-sourced. The formal v0.1 will be released soon. The paper RLinf: Flexible and Efficient Large-scale Reinforcement Learning via Macro-to-Micro Flow Transformation will also be released accordingly.
RLinf is unique with:
-
Macro-to-Micro Flow: a new paradigm M2Flow, which executes macro-level logical flows through micro-level execution flows, decoupling logical workflow construction (programmable) from physical communication and scheduling (efficiency).
-
Flexible Execution Modes
- Collocated mode: shares all GPUs across all workers.
- Disaggregated mode: enables fine-grained pipelining.
- Hybrid mode: a customizable combination of different placement modes, integrating both collocated and disaggregated modes.
-
Auto-scheduling Strategy: automatically selects the most suitable execution mode based on the training workload, without the need for manual resource allocation.
-
Embodied Agent Support
- Fast adaptation support for mainstream VLA models: OpenVLA, OpenVLA-OFT, and π₀.
- Support for mainstream CPU & GPU-based simulators via standardized RL interfaces: ManiSkill3, LIBERO.
- Enabling the first RL fine-tuning of the $\pi_0$ model family with a flow-matching action expert.
RLinf is fast with:
- Hybrid mode with fine-grained pipelining: achieves a 120%+ throughput improvement compared to other frameworks.
- Automatic Online Scaling Strategy: dynamically scales training resources, with GPU switching completed within seconds, further improving efficiency by 20–40% while preserving the on-policy nature of RL algorithms.
RLinf is flexible and easy to use with:
-
Multiple Backend Integrations
- FSDP + Hugging Face: rapid adaptation to new models and algorithms, ideal for beginners and fast prototyping.
- Megatron + SGLang: optimized for large-scale training, delivering maximum efficiency for expert users with demanding workloads.
-
Adaptive communication via the asynchronous communication channel
-
Built-in support for popular RL methods, including PPO, GRPO, DAPO, Reinforce++, and more.
- [ ] Support for heterogeneous GPUs
- [ ] Support for asynchronous pipeline execution
- [ ] Support for Mixture of Experts (MoE)
- [ ] Support for vLLM inference backend
- [ ] Support for Vision-Language Models (VLMs) training
- [ ] Support for deep searcher agent training
- [ ] Support for multi-agent training
- [ ] Support for integration with more embodied simulators (e.g., Meta-World, GENESIS)
- [ ] Support for more Vision Language Action models (VLAs), such as GR00T
- [ ] Support for world model
- [ ] Support for real-world RL embodied intelligence
Complete documentation for RLinf can be found Here.
Quickstart
- Installation
- Quickstart 1: PPO Training of VLAs on Maniskill3
- Quickstart 2: GRPO Training of LLMs on MATH
- Multi-node Training
- Model Evaluation
Key Design
- Unified User Interface Usage
- Flexible Execution Modes
- Enable Automatic Scheduling
- Elastic Communication
Example Gallery
Advanced Features
- 5D Parallelism Configuration for Megatron-LM
- LoRA Integration for efficient fine-tuning
- Switch between different versions of SGLang
- Checkpoint Resume and Recovery Support
Extending The Framework:
- Adding new Environments
- Adding new Models with FSDP+Huggingface backend
- Adding new Models with Megatron+SGLang backend
Blogs
Type | Status |
---|---|
Reasoning RL-MATH | |
Embodied RL-VLA |
We welcome contributions to RLinf. Please read contribution guide before taking action.
If you find RLinf helpful, please cite the GitHub repository:
@misc{RLinf_repo,
title = {RLinf: Reinforcement Learning Infrastructure for Agentic AI},
howpublished = {\url{https://github.com/RLinf/RLinf}},
note = {GitHub repository},
year = {2025}
}
Paper: A full paper describing RLinf will be released by September 20, 2025. We will update this section with the official citation and BibTeX when they become available.
Acknowledgements RLinf has been inspired by, and benefits from, the ideas and tooling of the broader open-source community. In particular, we would like to thank the teams and contributors behind VeRL, AReaL, Megatron-LM, SGLang, and PyTorch Fully Sharded Data Parallel (FSDP), and if we have inadvertently missed your project or contribution, please open an issue or a pull request so we can properly credit you.
Contact: We welcome applications from Postdocs, PhD/Master's students, and interns. Join us in shaping the future of RL infrastructure and embodied AI!
- Chao Yu: [email protected]
- Yu Wang: [email protected]
For Tasks:
Click tags to check more tools for each tasksFor Jobs:
Alternative AI tools for RLinf
Similar Open Source Tools

RLinf
RLinf is a flexible and scalable open-source infrastructure designed for post-training foundation models via reinforcement learning. It provides a robust backbone for next-generation training, supporting open-ended learning, continuous generalization, and limitless possibilities in intelligence development. The tool offers unique features like Macro-to-Micro Flow, flexible execution modes, auto-scheduling strategy, embodied agent support, and fast adaptation for mainstream VLA models. RLinf is fast with hybrid mode and automatic online scaling strategy, achieving significant throughput improvement and efficiency. It is also flexible and easy to use with multiple backend integrations, adaptive communication, and built-in support for popular RL methods. The roadmap includes system-level enhancements and application-level extensions to support various training scenarios and models. Users can get started with complete documentation, quickstart guides, key design principles, example gallery, advanced features, and guidelines for extending the framework. Contributions are welcome, and users are encouraged to cite the GitHub repository and acknowledge the broader open-source community.

AReaL
AReaL (Ant Reasoning RL) is an open-source reinforcement learning system developed at the RL Lab, Ant Research. It is designed for training Large Reasoning Models (LRMs) in a fully open and inclusive manner. AReaL provides reproducible experiments for 1.5B and 7B LRMs, showcasing its scalability and performance across diverse computational budgets. The system follows an iterative training process to enhance model performance, with a focus on mathematical reasoning tasks. AReaL is equipped to adapt to different computational resource settings, enabling users to easily configure and launch training trials. Future plans include support for advanced models, optimizations for distributed training, and exploring research topics to enhance LRMs' reasoning capabilities.

dash-infer
DashInfer is a C++ runtime tool designed to deliver production-level implementations highly optimized for various hardware architectures, including x86 and ARMv9. It supports Continuous Batching and NUMA-Aware capabilities for CPU, and can fully utilize modern server-grade CPUs to host large language models (LLMs) up to 14B in size. With lightweight architecture, high precision, support for mainstream open-source LLMs, post-training quantization, optimized computation kernels, NUMA-aware design, and multi-language API interfaces, DashInfer provides a versatile solution for efficient inference tasks. It supports x86 CPUs with AVX2 instruction set and ARMv9 CPUs with SVE instruction set, along with various data types like FP32, BF16, and InstantQuant. DashInfer also offers single-NUMA and multi-NUMA architectures for model inference, with detailed performance tests and inference accuracy evaluations available. The tool is supported on mainstream Linux server operating systems and provides documentation and examples for easy integration and usage.

joliGEN
JoliGEN is an integrated framework for training custom generative AI image-to-image models. It implements GAN, Diffusion, and Consistency models for various image translation tasks, including domain and style adaptation with conservation of semantics. The tool is designed for real-world applications such as Controlled Image Generation, Augmented Reality, Dataset Smart Augmentation, and Synthetic to Real transforms. JoliGEN allows for fast and stable training with a REST API server for simplified deployment. It offers a wide range of options and parameters with detailed documentation available for models, dataset formats, and data augmentation.

DataDreamer
DataDreamer is a powerful open-source Python library designed for prompting, synthetic data generation, and training workflows. It is simple, efficient, and research-grade, allowing users to create prompting workflows, generate synthetic datasets, and train models with ease. The library is built for researchers, by researchers, focusing on correctness, best practices, and reproducibility. It offers features like aggressive caching, resumability, support for bleeding-edge techniques, and easy sharing of datasets and models. DataDreamer enables users to run multi-step prompting workflows, generate synthetic datasets for various tasks, and train models by aligning, fine-tuning, instruction-tuning, and distilling them using existing or synthetic data.

Mooncake
Mooncake is a serving platform for Kimi, a leading LLM service provided by Moonshot AI. It features a KVCache-centric disaggregated architecture that separates prefill and decoding clusters, leveraging underutilized CPU, DRAM, and SSD resources of the GPU cluster. Mooncake's scheduler balances throughput and latency-related SLOs, with a prediction-based early rejection policy for highly overloaded scenarios. It excels in long-context scenarios, achieving up to a 525% increase in throughput while handling 75% more requests under real workloads.

fluid
Fluid is an open source Kubernetes-native Distributed Dataset Orchestrator and Accelerator for data-intensive applications, such as big data and AI applications. It implements dataset abstraction, scalable cache runtime, automated data operations, elasticity and scheduling, and is runtime platform agnostic. Key concepts include Dataset and Runtime. Prerequisites include Kubernetes version > 1.16, Golang 1.18+, and Helm 3. The tool offers features like accelerating remote file accessing, machine learning, accelerating PVC, preloading dataset, and on-the-fly dataset cache scaling. Contributions are welcomed, and the project is under the Apache 2.0 license with a vendor-neutral approach.

repromodel
ReproModel is an open-source toolbox designed to boost AI research efficiency by enabling researchers to reproduce, compare, train, and test AI models faster. It provides standardized models, dataloaders, and processing procedures, allowing researchers to focus on new datasets and model development. With a no-code solution, users can access benchmark and SOTA models and datasets, utilize training visualizations, extract code for publication, and leverage an LLM-powered automated methodology description writer. The toolbox helps researchers modularize development, compare pipeline performance reproducibly, and reduce time for model development, computation, and writing. Future versions aim to facilitate building upon state-of-the-art research by loading previously published study IDs with verified code, experiments, and results stored in the system.

oat
Oat is a simple and efficient framework for running online LLM alignment algorithms. It implements a distributed Actor-Learner-Oracle architecture, with components optimized using state-of-the-art tools. Oat simplifies the experimental pipeline of LLM alignment by serving an Oracle online for preference data labeling and model evaluation. It provides a variety of oracles for simulating feedback and supports verifiable rewards. Oat's modular structure allows for easy inheritance and modification of classes, enabling rapid prototyping and experimentation with new algorithms. The framework implements cutting-edge online algorithms like PPO for math reasoning and various online exploration algorithms.

clearml
ClearML is a suite of tools designed to streamline the machine learning workflow. It includes an experiment manager, MLOps/LLMOps, data management, and model serving capabilities. ClearML is open-source and offers a free tier hosting option. It supports various ML/DL frameworks and integrates with Jupyter Notebook and PyCharm. ClearML provides extensive logging capabilities, including source control info, execution environment, hyper-parameters, and experiment outputs. It also offers automation features, such as remote job execution and pipeline creation. ClearML is designed to be easy to integrate, requiring only two lines of code to add to existing scripts. It aims to improve collaboration, visibility, and data transparency within ML teams.

ClicShopping
ClicShopping AI™ is an open-source Ecommerce platform powered by Generative AI, designed for B2B, B2C, and B2B-B2C businesses. It offers seamless shopping experiences, advanced AI integration, modular architecture for customization, and responsive design across devices. With features like GPT API integration, RAG-powered Business Intelligence Agent, multi-model AI support, and security compliance, ClicShopping AI™ is a comprehensive solution for online businesses. It also provides internationalization support, performance analytics, server performance optimization, content management, API connections, shipping and payment options, and a marketplace for additional modules and apps.

PhiCookBook
Phi Cookbook is a repository containing hands-on examples with Microsoft's Phi models, which are a series of open source AI models developed by Microsoft. Phi is currently the most powerful and cost-effective small language model with benchmarks in various scenarios like multi-language, reasoning, text/chat generation, coding, images, audio, and more. Users can deploy Phi to the cloud or edge devices to build generative AI applications with limited computing power.

workflows-py
LlamaIndex Workflows is a framework for orchestrating and chaining together complex systems of steps and events. It shines in orchestrating complex, multi-step processes involving AI models, APIs, and decision-making. The async-first, event-driven architecture allows building workflows that can route between different capabilities, implement parallel processing patterns, loop over complex sequences, and maintain state across multiple steps. Key features include async-first design, event-driven structure, state management, and observability through tools like Arize Phoenix and OpenTelemetry.

lobe-chat
Lobe Chat is an open-source, modern-design ChatGPT/LLMs UI/Framework. Supports speech-synthesis, multi-modal, and extensible ([function call][docs-functionc-call]) plugin system. One-click **FREE** deployment of your private OpenAI ChatGPT/Claude/Gemini/Groq/Ollama chat application.

LLM-Zero-to-Hundred
LLM-Zero-to-Hundred is a repository showcasing various applications of LLM chatbots and providing insights into training and fine-tuning Language Models. It includes projects like WebGPT, RAG-GPT, WebRAGQuery, LLM Full Finetuning, RAG-Master LLamaindex vs Langchain, open-source-RAG-GEMMA, and HUMAIN: Advanced Multimodal, Multitask Chatbot. The projects cover features like ChatGPT-like interaction, RAG capabilities, image generation and understanding, DuckDuckGo integration, summarization, text and voice interaction, and memory access. Tutorials include LLM Function Calling and Visualizing Text Vectorization. The projects have a general structure with folders for README, HELPER, .env, configs, data, src, images, and utils.

openvino
OpenVINO™ is an open-source toolkit for optimizing and deploying AI inference. It provides a common API to deliver inference solutions on various platforms, including CPU, GPU, NPU, and heterogeneous devices. OpenVINO™ supports pre-trained models from Open Model Zoo and popular frameworks like TensorFlow, PyTorch, and ONNX. Key components of OpenVINO™ include the OpenVINO™ Runtime, plugins for different hardware devices, frontends for reading models from native framework formats, and the OpenVINO Model Converter (OVC) for adjusting models for optimal execution on target devices.
For similar tasks

RLinf
RLinf is a flexible and scalable open-source infrastructure designed for post-training foundation models via reinforcement learning. It provides a robust backbone for next-generation training, supporting open-ended learning, continuous generalization, and limitless possibilities in intelligence development. The tool offers unique features like Macro-to-Micro Flow, flexible execution modes, auto-scheduling strategy, embodied agent support, and fast adaptation for mainstream VLA models. RLinf is fast with hybrid mode and automatic online scaling strategy, achieving significant throughput improvement and efficiency. It is also flexible and easy to use with multiple backend integrations, adaptive communication, and built-in support for popular RL methods. The roadmap includes system-level enhancements and application-level extensions to support various training scenarios and models. Users can get started with complete documentation, quickstart guides, key design principles, example gallery, advanced features, and guidelines for extending the framework. Contributions are welcome, and users are encouraged to cite the GitHub repository and acknowledge the broader open-source community.

spear
SPEAR (Simulator for Photorealistic Embodied AI Research) is a powerful tool for training embodied agents. It features 300 unique virtual indoor environments with 2,566 unique rooms and 17,234 unique objects that can be manipulated individually. Each environment is designed by a professional artist and features detailed geometry, photorealistic materials, and a unique floor plan and object layout. SPEAR is implemented as Unreal Engine assets and provides an OpenAI Gym interface for interacting with the environments via Python.

Awesome-LLM-3D
This repository is a curated list of papers related to 3D tasks empowered by Large Language Models (LLMs). It covers tasks such as 3D understanding, reasoning, generation, and embodied agents. The repository also includes other Foundation Models like CLIP and SAM to provide a comprehensive view of the area. It is actively maintained and updated to showcase the latest advances in the field. Users can find a variety of research papers and projects related to 3D tasks and LLMs in this repository.
For similar jobs

RLinf
RLinf is a flexible and scalable open-source infrastructure designed for post-training foundation models via reinforcement learning. It provides a robust backbone for next-generation training, supporting open-ended learning, continuous generalization, and limitless possibilities in intelligence development. The tool offers unique features like Macro-to-Micro Flow, flexible execution modes, auto-scheduling strategy, embodied agent support, and fast adaptation for mainstream VLA models. RLinf is fast with hybrid mode and automatic online scaling strategy, achieving significant throughput improvement and efficiency. It is also flexible and easy to use with multiple backend integrations, adaptive communication, and built-in support for popular RL methods. The roadmap includes system-level enhancements and application-level extensions to support various training scenarios and models. Users can get started with complete documentation, quickstart guides, key design principles, example gallery, advanced features, and guidelines for extending the framework. Contributions are welcome, and users are encouraged to cite the GitHub repository and acknowledge the broader open-source community.

FlagPerf
FlagPerf is an integrated AI hardware evaluation engine jointly built by the Institute of Intelligence and AI hardware manufacturers. It aims to establish an industry-oriented metric system to evaluate the actual capabilities of AI hardware under software stack combinations (model + framework + compiler). FlagPerf features a multidimensional evaluation metric system that goes beyond just measuring 'whether the chip can support specific model training.' It covers various scenarios and tasks, including computer vision, natural language processing, speech, multimodal, with support for multiple training frameworks and inference engines to connect AI hardware with software ecosystems. It also supports various testing environments to comprehensively assess the performance of domestic AI chips in different scenarios.

AI-System-School
AI System School is a curated list of research in machine learning systems, focusing on ML/DL infra, LLM infra, domain-specific infra, ML/LLM conferences, and general resources. It provides resources such as data processing, training systems, video systems, autoML systems, and more. The repository aims to help users navigate the landscape of AI systems and machine learning infrastructure, offering insights into conferences, surveys, books, videos, courses, and blogs related to the field.

multi-agent-orchestrator
Multi-Agent Orchestrator is a flexible and powerful framework for managing multiple AI agents and handling complex conversations. It intelligently routes queries to the most suitable agent based on context and content, supports dual language implementation in Python and TypeScript, offers flexible agent responses, context management across agents, extensible architecture for customization, universal deployment options, and pre-built agents and classifiers. It is suitable for various applications, from simple chatbots to sophisticated AI systems, accommodating diverse requirements and scaling efficiently.

AIXP
The AI-Exchange Protocol (AIXP) is a communication standard designed to facilitate information and result exchange between artificial intelligence agents. It aims to enhance interoperability and collaboration among various AI systems by establishing a common framework for communication. AIXP includes components for communication, loop prevention, and task finalization, ensuring secure and efficient collaboration while avoiding infinite communication loops. The protocol defines access points, data formats, authentication, authorization, versioning, loop detection, status codes, error messages, and task completion verification. AIXP enables AI agents to collaborate seamlessly and complete tasks effectively, contributing to the overall efficiency and reliability of AI systems.

AIInfra
AIInfra is an open-source project focused on AI infrastructure, specifically targeting large models in distributed clusters, distributed architecture, distributed training, and algorithms related to large models. The project aims to explore and study system design in artificial intelligence and deep learning, with a focus on the hardware and software stack for building AI large model systems. It provides a comprehensive curriculum covering topics such as AI chip principles, communication and storage, AI clusters, large model training, and inference, as well as algorithms for large models. The course is designed for undergraduate and graduate students, as well as professionals working with AI large model systems, to gain a deep understanding of AI computer system architecture and design.

eino
Eino is an ultimate LLM application development framework in Golang, emphasizing simplicity, scalability, reliability, and effectiveness. It provides a curated list of component abstractions, a powerful composition framework, meticulously designed APIs, best practices, and tools covering the entire development cycle. Eino standardizes and improves efficiency in AI application development by offering rich components, powerful orchestration, complete stream processing, highly extensible aspects, and a comprehensive framework structure.

zeta
Zeta is a tool designed to build state-of-the-art AI models faster by providing modular, high-performance, and scalable building blocks. It addresses the common issues faced while working with neural nets, such as chaotic codebases, lack of modularity, and low performance modules. Zeta emphasizes usability, modularity, and performance, and is currently used in hundreds of models across various GitHub repositories. It enables users to prototype, train, optimize, and deploy the latest SOTA neural nets into production. The tool offers various modules like FlashAttention, SwiGLUStacked, RelativePositionBias, FeedForward, BitLinear, PalmE, Unet, VisionEmbeddings, niva, FusedDenseGELUDense, FusedDropoutLayerNorm, MambaBlock, Film, hyper_optimize, DPO, and ZetaCloud for different tasks in AI model development.