Share via

Mpi Class

Manages Message Passing Interface (MPI) settings for distributed training jobs.

DEPRECATED. Use the MpiConfiguration class.

MPI can be specified for a job with the distributed_training parameter of preconfigured estimators Chainer, PyTorch, and TensorFlow, or with a generic Estimator.

A class for managing MPI settings for jobs.

Constructor

Mpi(process_count_per_node=1)

Parameters

Name Description
process_count_per_node
int

The number of processes (or "workers") to run on each node.

Default value: 1
process_count_per_node
Required
int

When using MPI, number of processes per node.