The parallelization for glocal alignment via openmpi is giving issues. According to the docs the number of mpi processes should always be one larger than the number of gpu's, but this is failing currently. Also, it fails when setting multiple processes without gpu's. Something is clearly broken.
Testing the functionality can be done easily by going to the E2ETests and changes the number of openmpi processes in the different tests.
The parallelization for glocal alignment via openmpi is giving issues. According to the docs the number of mpi processes should always be one larger than the number of gpu's, but this is failing currently. Also, it fails when setting multiple processes without gpu's. Something is clearly broken.
Testing the functionality can be done easily by going to the E2ETests and changes the number of openmpi processes in the different tests.