Hi,
I have a case (with containerization) where I would love to run an evaluation of non-MPI simulations in parallel on a single node. Each simulation uses one GPU exclusively and my node has 4 GPUs.
The example here seems to suggest one can change the libEnsemble executor:
https://github.com/optimas-org/optimas/blob/v0.8.1/examples/dummy_mt/run_example.py#L60-L62
But from what I see in practice I think optimas only implements the MPIExecutor of libEnsemble, not the "base executor", too?
(And it defines a hard dependency on mpi4py?)
Hi,
I have a case (with containerization) where I would love to run an evaluation of non-MPI simulations in parallel on a single node. Each simulation uses one GPU exclusively and my node has 4 GPUs.
The example here seems to suggest one can change the libEnsemble executor:
https://github.com/optimas-org/optimas/blob/v0.8.1/examples/dummy_mt/run_example.py#L60-L62
But from what I see in practice I think optimas only implements the
MPIExecutorof libEnsemble, not the "base executor", too?(And it defines a hard dependency on mpi4py?)