The tests failed because allegro/nequip is taking too long. Ideally each test input should run in less than 10 seconds. This is something we also noticed on the dashboard.

Could you speed them up somehow? Is see that you are already doing just a single MD step. So maybe we need to make the system smaller, e.g. use only 16 water molecules.

Is the fix I added in #3434 erroneous?

Is the fix I added in #3434 erroneous?

I checked stress with 1 and 4 cores (both Allegro and NequIP) and I get the same result, so I think all is good there 😊

Could you speed them up somehow? Is see that you are already doing just a single MD step. So maybe we need to make the system smaller, e.g. use only 16 water molecules.

I have reduced all the models to 1 layer along with reducing few other training parameters, which makes the models not suitable for production runs but speeds up the tests. From regtest-allegro/ I have removed water-gra-film-sp.inp and left only water-bulk-sp.inp since they are using the same model, but the later one is faster. From regtest-nequip/ I removed water-bulk-dp.inp energy test case and left only stress test case since they share the input. This has improved the timings overall for both sdbg and pdbg. As for the system size, it took 14sec for a single water molecule and 19sec for 32 waters, so halving the system size, unfortunately, does not make it much faster. Thank you for the suggestions 😊