Learning Energy Decompositions for Partial Inference in GFlowNets

This is an official implementation of our paper Learning Energy Decompositions for Partial Inference in GFlowNets (LED-GFN) accepted at ICLR 2024 (oral).

Repository Overview

We provide the PyTorch implementation for LED-GFN framework here. Our implementations are based on prior studies:

• led_gfn-Bengio (molecule): The code is based on https://github.com/ling-pan/FL-GFN
• led_gfn-Shen (bag and rna sequence): The code is based on https://github.com/maxwshen/gflownet
• led_gfn-Zhang (maximum independent set): The code is based on https://github.com/zdhNarsil/GFlowNet-CombOpt

Setting up the environment and Running

One can check how to run the experiments in each directory.