Hannes Vasyura-Bathke

Hi again, cool that you are still around ;) . You are right. The intention behind that is, if your forward model takes a long time, you want to rather...

might be related to theano amd linked library

Please check your bounds for the hyper-parameters. Are they still very wide? Then this can happen. Please reduce them significantly to like -1 to 3.

Ah ok saw that you posted everything. The last points make no sense why they dont work- I have not the slightest clue. The workaround for now would be to...

Sure as this will try to create 13 processes which computer doesn't have threads. I was asking for the --oversubscribe flag. Did you try that? For example: mpiexec --oversubscribe -n...

Great so thats the flag then we have to include in the execution string. We will have to check if it is available across versions- I will do that.

This seems to have deeper roots connected to issues with MPI distributions. Collecting here a bunch of relevant threads to further investigate: https://github.com/h5py/h5py/pull/1117 https://github.com/radiasoft/sirepo/issues/1311 https://github.com/radiasoft/sirepo/issues/1323

A fast workaround for that issue could be to ensure to use older MPI versions (

need to cleanup master. ..

OpenMPI version 3.3.2 works with mpi4py earlier versions in 3 have the init issue: https://bitbucket.org/mpi4py/mpi4py/issues/95/mpi4py-openmpi-300-breaks-subprocess mpi4py does not support OpenMPI4