Trying parallel training for cpu version of deepmd-kit

[AnuragKr]

I was trying to do parallel training for the CPU version of deepmd-kit to compare the performance b/w CPU single thread,multi-thread, and multi-node.

Command --
mpirun --hostfile hostfile -np 4 dp train --mpi-log=workers input.json

hostfile Content --
localhost slots=6

Output --

I want to know why it is not running in parallel or if this feature is limited to the GPU version of deepmd-kit.. Usually, we are able to launch multiple processes for regular code via mpirun. If I want to train to CPU cluster with multi-cores, how should I go about it?

[njzjz]

Did you install horovod?

[AnuragKr]

No, I didn't install horovod, Do I need to install horovod to make mpirun work?

[njzjz]

Yes.

[AnuragKr]

Thanks, @njzjz. It worked after installing horovod.