https://github.com/typoverflow/flow-rl

Flow RL is a high-performance RL library with flow and diffusion models.
https://github.com/typoverflow/flow-rl

Last synced: 12 months ago
JSON representation

Flow RL is a high-performance RL library with flow and diffusion models.

• Host: GitHub
• URL: https://github.com/typoverflow/flow-rl
• Owner: typoverflow
• License: mit
• Created: 2025-04-11T05:27:14.000Z (about 1 year ago)
• Default Branch: master
• Last Pushed: 2025-05-29T13:00:47.000Z (about 1 year ago)
• Last Synced: 2025-06-28T06:25:23.421Z (12 months ago)
• Language: Python
• Homepage:
• Size: 148 KB
• Stars: 6
• Watchers: 2
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# Flow RL

[![PyPI version](https://img.shields.io/pypi/v/flowrl.svg)](https://pypi.org/project/flowrl) [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](LICENSE) [![Python 3.11+](https://img.shields.io/badge/python-3.11%2B-green.svg)](https://www.python.org/) [![Python 3.8+](https://static.pepy.tech/badge/flowrl)](https://pepy.tech/projects/flowrl)

Flow RL is a high-performance reinforcement learning library, combining modern deep RL algorithms with flow and diffusion models for advanced policy parameterization, planning ability or dynamics modeling. It features:

- **State-of-the-Art Algorithms and Efficiency**: We provide JAX implementations of SOTA algorithms, such FQL, BDPO, DAC and etc;

- **Flexible Flow Architectures**: We provide built-in support various types of flow and diffusion models, such as CNFs and DDPM;

- **Comprehensive Evaluations**: We test the algorithms on commonly adopted benchmark and provide the results.

## 🚀 Installation & Usage

Currently FlowRL is hosted on PyPI and therefore can be installed via `pip install flowrl`. However, we recommend to clone and install the library using the following commands:

```bash

git clone https://github.com/typoverflow/flow-rl.git

cd flow-rl

pip install -e .

```

The entry files are presented in `examples/`. Please refer to the scripts in `scripts/` for how to execute the algorithms.

## 📊 Supported Algorithms

Offline RL:

|Algorithm|Location|WandB Report|

|:---:|:---:|:---:|

|IQL|`flowrl/agent/iql.py`|[[Performance]](https://wandb.ai/lamda-rl/flow-rl?nw=urvdu9rz7b&panelDisplayName=eval%2Fmean&panelSectionName=eval) [[Full Log]](https://wandb.ai/lamda-rl/flow-rl?nw=urvdu9rz7b)|

|IVR|`flowrl/agent/ivr.py`|[[Performance]](https://wandb.ai/lamda-rl/flow-rl/panel/nz7r4sj4n?nw=oslzekjlr1q) [[Full Log]](https://wandb.ai/lamda-rl/flow-rl?nw=oslzekjlr1q)|

|FQL|`flowrl/agent/fql/fql.py`|[[Performance]](https://wandb.ai/lamda-rl/flow-rl?nw=u9y84ki7rdi&panelDisplayName=eval%2Fmean&panelSectionName=eval) [[Full Log]](https://wandb.ai/lamda-rl/flow-rl?nw=u9y84ki7rdi)|

|DAC|`flowrl/agent/dac.py`|[[Performance]](https://wandb.ai/lamda-rl/flow-rl/panel/nz7r4sj4n?nw=uqr7jg46c5) [[Full Log]](https://wandb.ai/lamda-rl/flow-rl?nw=uqr7jg46c5)|

|BDPO|`flowrl/agent/bdpo/bdpo.py`|[[Performance]](https://wandb.ai/lamda-rl/flow-rl/panel/nz7r4sj4n?nw=2q8v54gusia) [[Full Log]](https://wandb.ai/lamda-rl/flow-rl?nw=2q8v54gusia)|

## 📝 Citing Flow RL

If you use Flow RL in your research, please cite:

```bibtex

@software{flow_rl,

  author       = {Chen-Xiao Gao and Mingjun Cao},

  title        = {Flow RL: Flow-based Reinforcement Learning Algorithms},

  year         = 2025,

  version      = {v0.0.1},

  url          = {https://github.com/typoverflow/flow-rl}

}

```

## 💎 Acknowledgements

Inspired by foundational work from

- [Jax-CORL](https://github.com/nissymori/JAX-CORL)

- [DAC](https://github.com/Fang-Lin93/DAC)