Performance difference between pytorch direct and forpy coupling

DataWaveProject / MiMA-machine-learning

Model of an idealized Moist Atmosphere: Intermediate-complexity General Circulation Model with full radiation

GNU General Public License v3.0

1 stars 0 forks source link

Performance difference between pytorch direct and forpy coupling #9

Open dorchard opened 1 year ago

dorchard commented 1 year ago

Our hypothesis was that direct coupling would always be faster, but the latest benchmarks show that a forpy based coupling for PyTorch is faster (even if a bit more inconvenient). We are not sure why yet but think it might be something to do with the special support for ndarray's provided by forpy. This needs some more exploration.

SimonClifford commented 1 year ago

This is being addressed in https://github.com/Cambridge-ICCS/fortran-pytorch-lib-benchmark