verl is a flexible, efficient and production-ready RL training library for large language models (LLMs).
verl is the open-source version of HybridFlow: A Flexible and Efficient RLHF Framework paper.
verl is flexible and easy to use with:
-
Easy extension of diverse RL algorithms: The hybrid-controller programming model enables flexible representation and efficient execution of complex Post-Training dataflows. Build RL dataflows such as GRPO, PPO in a few lines of code.
-
Seamless integration of existing LLM infra with modular APIs: Decouples computation and data dependencies, enabling seamless integration with existing LLM frameworks, such as FSDP, Megatron-LM, vLLM, SGLang, etc
-
Flexible device mapping: Supports various placement of models onto different sets of GPUs for efficient resource utilization and scalability across different cluster sizes.
-
Ready integration with popular HuggingFace models
verl is fast with:
-
State-of-the-art throughput: SOTA LLM training and inference engine integrations and SOTA RL throughput.
-
Efficient actor model resharding with 3D-HybridEngine: Eliminates memory redundancy and significantly reduces communication overhead during transitions between training and generation phases.
- [2025/03] verl v0.3.0.post1 is released! See release note for details.
- [2025/03] DAPO is the open-sourced SOTA RL algorithm that achieves 50 points on AIME 2024 based on the Qwen2.5-32B pre-trained model, surpassing the previous SOTA achieved by DeepSeek's GRPO (DeepSeek-R1-Zero-Qwen-32B). DAPO's training is fully powered by verl and the reproduction code is publicly available now.
- [2025/03] We will present verl(HybridFlow) at EuroSys 2025. See you in Rotterdam!
- [2025/03] We introduced the programming model of verl at the vLLM Beijing Meetup and verl intro and updates at the LMSys Meetup in Sunnyvale mid March.
- [2025/02] verl v0.2.0.post2 is released!
- [2025/01] Doubao-1.5-pro is released with SOTA-level performance on LLM & VLM. The RL scaling preview model is trained using verl, reaching OpenAI O1-level performance on math benchmarks (70.0 pass@1 on AIME).
more...
- [2025/02] We presented verl in the Bytedance/NVIDIA/Anyscale Ray Meetup. See you in San Jose!
- [2024/12] verl is presented at Ray Forward 2024. Slides available here
- [2024/10] verl is presented at Ray Summit. Youtube video available.
- [2024/12] The team presented Post-training LLMs: From Algorithms to Infrastructure at NeurIPS 2024. Slides and video available.
- [2024/08] HybridFlow (verl) is accepted to EuroSys 2025.
- FSDP and Megatron-LM for training.
- vLLM, SGLang(experimental) and HF Transformers for rollout generation.
- Compatible with Hugging Face Transformers and Modelscope Hub: Qwen-2.5, Llama3.1, Gemma2, DeepSeek-LLM, etc
- Supervised fine-tuning.
- Reinforcement learning with PPO, GRPO, ReMax, REINFORCE++, RLOO, PRIME, etc.
- Support model-based reward and function-based reward (verifiable reward)
- Support vision-language models (VLMs) and multi-modal RL
- Flash attention 2, sequence packing, sequence parallelism support via DeepSpeed Ulysses, LoRA, Liger-kernel.
- Scales up to 70B models and hundreds of GPUs.
- Experiment tracking with wandb, swanlab, mlflow and tensorboard.
- Roadmap volcengine#710
- DeepSeek 671b optimizations with Megatron v0.11 volcengine#708
- Multi-turn rollout optimizations
- Environment interactions
Quickstart:
Running a PPO example step-by-step:
- Data and Reward Preparation
- Understanding the PPO Example
Reproducible algorithm baselines:
For code explanation and advance usage (extension):
- PPO Trainer and Workers
- Advance Usage and Extension
Blogs from the community
The performance is essential for on-policy RL algorithm. We have written a detailed performance tuning guide to help you optimize performance.
veRL now supports vLLM>=0.8.2 when using FSDP as the training backend. Please refer to this document for installation guide and more information. Please avoid vllm 0.7.x which contains bugs that may lead to OOMs and unexpected errors.
If you find the project helpful, please cite:
- HybridFlow: A Flexible and Efficient RLHF Framework
- A Framework for Training Large Language Models for Code Generation via Proximal Policy Optimization
@article{sheng2024hybridflow,
title = {HybridFlow: A Flexible and Efficient RLHF Framework},
author = {Guangming Sheng and Chi Zhang and Zilingfeng Ye and Xibin Wu and Wang Zhang and Ru Zhang and Yanghua Peng and Haibin Lin and Chuan Wu},
year = {2024},
journal = {arXiv preprint ar
8000
Xiv: 2409.19256}
}verl is inspired by the design of Nemo-Aligner, Deepspeed-chat and OpenRLHF. The project is adopted and contributed by Bytedance, Anyscale, LMSys.org, Alibaba Qwen team, Shanghai AI Lab, Tsinghua University, UC Berkeley, UCLA, UIUC, University of Hong Kong, ke.com, All Hands AI, ModelBest, OpenPipe, JD AI Lab, Microsoft Research, StepFun, Amazon, Linkedin, Meituan, Camel-AI, OpenManus, Prime Intellect, NVIDIA research, Baichuan, and many more.
- TinyZero: a reproduction of DeepSeek R1 Zero recipe for reasoning tasks
- DAPO: the fully open source SOTA RL algorithm that beats DeepSeek-R1-zero-32B
- SkyThought: RL training for Sky-T1-7B by NovaSky AI team.
- simpleRL-reason: SimpleRL-Zoo: Investigating and Taming Zero Reinforcement Learning for Open Base Models in the Wild
- Easy-R1: Multi-modal RL training framework
- OpenManus-RL: LLM Agents RL tunning framework for multiple agent environments.
- deepscaler: iterative context scaling with GRPO
- PRIME: Process reinforcement through implicit rewards
- RAGEN: a general-purpose reasoning agent training framework
- Logic-RL: a reproduction of DeepSeek R1 Zero on 2K Tiny Logic Puzzle Dataset.
- Search-R1: RL with reasoning and searching (tool-call) interleaved LLMs
- ReSearch: Learning to Reason with Search for LLMs via Reinforcement Learning
- DeepRetrieval: Hacking Real Search Engines and retrievers with LLMs via RL for information retrieval
- Code-R1: Reproducing R1 for Code with Reliable Rewards
- cognitive-behaviors: Cognitive Behaviors that Enable Self-Improving Reasoners, or, Four Habits of Highly Effective STaRs
- PURE: Credit assignment is the key to successful reinforcement fine-tuning using process reward model
- MetaSpatial: Reinforcing 3D Spatial Reasoning in VLMs for the Metaverse
- DeepEnlighten: Reproduce R1 with social reasoning tasks and analyze key findings
- DeepResearcher: Scaling deep research via reinforcement learning in real-world environments
- self-rewarding-reasoning-LLM: self-rewarding and correction with generative reward models
- critic-rl: LLM critics for code generation
- VAGEN: Training VLM agents with multi-turn reinforcement learning
- DQO: Enhancing multi-Step reasoning abilities of language models through direct Q-function optimization
- FIRE: Flaming-hot initiation with regular execution sampling for large language models
- Rec-R1: Bridging Generative Large Language Models and Recommendation Systems via Reinforcement Learning
- all-hands/openhands-lm-32b-v0.1: A strong, open coding agent model, trained with multi-turn fine-tuning
Contributions from the community are welcome! Please check out our project roadmap and good first issues to see where you can contribute.
We use yapf (Google style) to enforce strict code formatting when reviewing PRs. To reformat your code locally, make sure you have installed the latest version of yapf
pip3 install yapf --upgradeThen, make sure you are at top level of verl repo and run
bash scripts/format.shWe are HIRING! Send us an email if you are interested in internship/FTE opportunities in MLSys/LLM reasoning/multimodal alignment.