speed up inference just by torch

MDIL-SNU / SevenNet

SevenNet - a graph neural network interatomic potential package supporting efficient multi-GPU parallel molecular dynamics simulations.

https://pubs.acs.org/doi/10.1021/acs.jctc.4c00190

GNU General Public License v3.0

126 stars 15 forks source link

speed up inference just by torch #55

Closed heyfavour closed 3 months ago

heyfavour commented 3 months ago

can speed up inference just in pytorch without lammps? i have my model , and i just want to speed up it in python , i want to split gnn message in multi-gpu,but i can't get grad

YutackPark commented 3 months ago

Sorry, but it is not feasible. You may try using LAMMPS with the Python interface.

If we try the same approach as I implemented for LAMMPS with the multi-GPU version of SevenNet, we would have to implement the fundamentals of the parallel algorithm in LAMMPS to Python, which is non-trivial.