| # Recipe |
| > Update 2025/11/25: recipes have been moved to a new repository: [verl-recipe](https://github.com/verl-project/verl-recipe). |
|
|
| verl is designed to be a modular, extensible framework for post-training: SFT and RL. Recipe is expected to import verl as a library, with necessary extensions to build specific RL training pipeline. If you find verl can't meet recipe's requirements, please open an issue or PR to verl. |
|
|
| There's still some incubation recipes kept here, which is expected to be offically supported in verl in the future. |
| - fully_async_policy: fully asynchronous off-policy training with decoupled trainer and rollout. |
| - transfer_queue: high performance asynchronous streaming data management system. |
| - vla: VLA model RL training. |
| |
| # Awesome work using verl |
| |
| - [FlowRL](https://github.com/Xuekai-Zhu/FlowRL): Matching reward distributions via **flow balance** for diverse exploration and generalizable reasoning  |
| - [Logic-RL](https://github.com/Unakar/Logic-RL): a reproduction of DeepSeek R1 Zero on 2K Tiny Logic Puzzle Dataset.  |
| - [Seed-Coder](https://github.com/ByteDance-Seed/Seed-Coder): RL training of Seed-Coder boosts performance on competitive programming  |
| - [all-hands/openhands-lm-32b-v0.1](https://www.all-hands.dev/blog/introducing-openhands-lm-32b----a-strong-open-coding-agent-model): A strong, open coding agent model, trained with [multi-turn fine-tuning](https://github.com/volcengine/verl/pull/195) |
| - [s3](https://github.com/pat-jj/s3) **Efficient Yet Effective** Search Agent Training via RL  |
| - [Rec-R1](https://arxiv.org/pdf/2503.24289): Bridging Generative Large Language Models and Recommendation Systems via Reinforcement Learning |
| - [Explore RL Data Scaling](https://arxiv.org/abs/2503.22230): Exploring Data Scaling Trends and Effects in Reinforcement Learning from Human Feedback |
| - [FIRE](https://arxiv.org/abs/2410.21236): Flaming-hot initiation with regular execution sampling for large language models |
| - [DQO](https://arxiv.org/abs/2410.09302): Enhancing multi-Step reasoning abilities of language models through direct Q-function optimization |
| - [ProRL](https://arxiv.org/abs/2505.24864): Prolonged Reinforcement Learning Expands Reasoning Boundaries in Large Language Models |
| - [cognition-engineering](https://github.com/gair-nlp/cognition-engineering): Test time scaling drives cognition engineering.  |
| - [Trust Region Preference Approximation](https://github.com/XueruiSu/Trust-Region-Preference-Approximation): A simple and stable **reinforcement learning algorithm** for LLM reasoning.  |
| - [AdaRFT](https://github.com/uscnlp-lime/verl): Efficient Reinforcement Finetuning via **Adaptive Curriculum Learning**  |
| - [critic-rl](https://github.com/HKUNLP/critic-rl): LLM critics for code generation  |
| - [self-rewarding-reasoning-LLM](https://arxiv.org/pdf/2502.19613): self-rewarding and correction with **generative reward models**  |
| - [DeepEnlighten](https://github.com/DolbyUUU/DeepEnlighten): Reproduce R1 with **social reasoning** tasks and analyze key findings  |
| - [MetaSpatial](https://github.com/PzySeere/MetaSpatial): Reinforcing **3D Spatial Reasoning** in **VLMs** for the **Metaverse**  |
| - [PURE](https://github.com/CJReinforce/PURE): **Credit assignment** is the key to successful reinforcement fine-tuning using **process reward model**  |
| - [cognitive-behaviors](https://github.com/kanishkg/cognitive-behaviors): Cognitive Behaviors that Enable Self-Improving Reasoners, or, Four Habits of Highly Effective STaRs  |
| - [deepscaler](https://github.com/agentica-project/rllm/tree/deepscaler): iterative context scaling with GRPO  |
| - [DAPO](https://dapo-sia.github.io/): the fully open source SOTA RL algorithm that beats DeepSeek-R1-zero-32B  |
| - [NoisyRollout](https://github.com/NUS-TRAIL/NoisyRollout): Reinforcing Visual Reasoning with Data Augmentation  |
| |
