Difficult to reproduce results listed in the paper

[ltz0120]

Thank you for your fantastic work!

However, I got a problem reproducing the results in your paper. Based on your released code and the method listed in your paper, I got much inferior results.

Could you please share more details on your implementation for molecule generation? Thank you so much.

[zdhNarsil]

Hi, not sure what is your case, but there is some numerical stability issue in the original implementation of graph neural network in molecule experiments. I have a slightly cleaned up version here https://github.com/zdhNarsil/Distributional-GFlowNets