mirror of https://github.com/hanzo-ml/trainer.git synced 2026-09-24 00:31:34 +00:00

Files

T

Chitsing KUIandGitHub a81242a4cb Set correct ENV for PytorchJob to support torchrun (#1840 )

* fix pytorch WORLD_SIZE env inconsistent

* add correct env node_rank, nnodes for torch

* restore elastic nnproc

* use string for nproc_per_node

* add defaults in api

* add validation for two nproc_per_node, use auto for defaulter

* add ut for defaults and validation

* fix ut

* add doc for nproc_per_node

2023-07-12 17:11:07 +00:00

1.1 KiB

Raw Blame History

KubeflowOrgV1PyTorchJobSpec

PyTorchJobSpec is a desired state description of the PyTorchJob.

Properties

Name	Type	Description	Notes
elastic_policy	KubeflowOrgV1ElasticPolicy		[optional]
nproc_per_node	str	Number of workers per node; supported values: [auto, cpu, gpu, int]. For more, https://github.com/pytorch/pytorch/blob/26f7f470df64d90e092081e39507e4ac751f55d6/torch/distributed/run.py#L629-L658. Defaults to auto.	[optional]
pytorch_replica_specs	dict(str, KubeflowOrgV1ReplicaSpec)	A map of PyTorchReplicaType (type) to ReplicaSpec (value). Specifies the PyTorch cluster configuration. For example, { "Master": PyTorchReplicaSpec, "Worker": PyTorchReplicaSpec, }
run_policy	KubeflowOrgV1RunPolicy

[Back to Model list] [Back to API list] [Back to README]

1.1 KiB Raw Blame History

KubeflowOrgV1PyTorchJobSpec

Properties

1.1 KiB

Raw Blame History