PyTorchConfiguration Class

  • Reference

Represents configuration information for distributed PyTorch jobs.

This class is used in the RunConfiguration class.

Class PyTorchConfiguration constructor.

Inheritance
azureml._base_sdk_common.abstract_run_config_element._AbstractRunConfigElement
PyTorchConfiguration

Constructor

PyTorchConfiguration(communication_backend='Nccl', process_count=None, node_count=1)

Parameters

communication_backend
str
default value: Nccl

Communication backend to use for PyTorch distributed training. Supported backends are "Nccl" and "Gloo". Default is "Nccl".

process_count
int
default value: None

The total number of processes to launch for the job. By default the value will be set to the node_count.

node_count
int
default value: 1

The number of nodes to use for the job.