Generative Flow Networks - GFlowNet

A modular, easy to extend GFlowNet library

GFlowNet library specialized for graph & molecular data

A deep reinforcement learning framework for generating formulaic alpha factors for quantitative investment, powered by GFlowNet, implemented in Python&PyTorch.

Environments and Algorithms for Generative Flow Networks in JAX

DynGFN: Bayesian Dynamic Causal Discovery using Generative Flow Networks

A PyTorch implementation of a Generative Flow Network (GFlowNet) proposed by Bengio et al. (2021)

Code for GFlowNet-EM, a novel algorithm for fitting latent variable models with compositional latents and an intractable true posterior.

Repository for "Generative Flow Networks as Entropy-Regularized RL" (AISTATS-2024, Oral)

Synthesis-oriented GFlowNets on a large action space: "Generative Flows on Synthetic Pathway for Drug Design" (ICLR 2025)

The source code for the paper "Optimizing Backward Policies in GFlowNets via Trajectory Likelihood Maximization" (ICLR 2025)

Official Implementation of Nabla-GFlowNet (ICLR 2025)

Repository for "Revisiting Non-Acyclic GFlowNets in Discrete Environments" (ICML 2025)

A collection of deep reinforcement learning-based & GFlowNet drug molecule generators focused on generation of molecules using Graphs/SELFIES guided by modern retrosynthesis tools to increase synthetic accessibility of de-novo designed drugs.

Investigating Generalization Behaviours of Generative Flow Networks

A simple implementation of a GFlowNet using PyTorch

GFlowNet-based therapeutic peptide design. Generates diverse, high-quality candidates (5.4x better diversity vs PPO/GRPO) to reduce correlated wet-lab failures. Published method with pre-trained models. Validated on FLIP & Propedia benchmarks.

LUCIDE : Latent Unified Causal Inference through Dynamic Equilibrium

Code for implementing GFlowOut for medical images, to improve uncertainty predictions.

🎯 Build and explore Generative Flow Networks easily with gfnx, a JAX-native toolkit offering environments and baseline agents for rapid experimentation.