search

gcorso / DiffDock

Implementation of DiffDock: Diffusion Steps, Twists, and Turns for Molecular Docking
https://arxiv.org/abs/2210.01776
MIT License
1.1k stars 263 forks source link

Upgrade `e3nn` to fix batch normalization runtime issue with `final_conv` module #155

Closed amorehead closed 6 months ago

amorehead commented 1 year ago

Starting with version 0.5.1 of e3nn, the authors of the library are now guarding against attempting to aggregate scalar means if no scalar irreps are given as input: https://github.com/e3nn/e3nn/blob/bb71b3c75c38c6d633cfebe9988b1fb1563432c0/e3nn/nn/_batchnorm.py#L191

(This PR also adds wandb to the installation instructions for new Conda environments)

It's possible version 0.5.0 of e3nn was pinned accidentally, even though the source code will only run with 0.5.1 of e3nn. I hope this helps!

jsilter commented 6 months ago

@amorehead Have you encountered this e3nn bug while running DiffDock?

amorehead commented 6 months ago

Hello. I haven't tested to see if DiffDock-L's codebase now raises this error, but for DiffDock 1 it did.