README.md

This repo include the code for the hypergrid experiments shown in the "GFlowNets and Variational Inference" paper.

The files paper_configs and small_configs shows the configurations used for the main file in order to obtain the paper results.

To use

First install the gfn library:

git clone https://github.com/saleml/gfn.git
conda create -n gfn python=3.10
conda activate gfn
cd gfn
pip install -e .

Then:

conda activate gfn
pip install tqdm

and optionally pip install wandb

You should then be able to run train.py with python train.py --no_wandb --env manual --ndim 2 --height 16 --mode reverse_kl for example