how_to_use_pytorch_plugin.md

Pytorch Plugin User Guide

Introduction

Pytorch plugin is designed to optimize the user experience when running pytorch jobs, it not only allows users to write less yaml, but also ensures the normal operation of Pytorch jobs.

How the Pytorch Plugin Works

The Pytorch Plugin will do three things:

• Open ports used by Pytorch for all containers of the job
• Force open svc plugins
• Add some envs such like MASTER_ADDR, MASTER_PORT, WORLD_SIZE, RANK which pytorch distributed training needed to containers automatically

Parameters of the Pytorch Plugin

Arguments

ID	Name	Type	Default Value	Required	Description	Example
1	master	string	master	No	Name of Pytorch master	--master=master
2	worker	string	worker	No	Name of Pytorch worker	--worker=worker
3	port	string	23456	No	The port to open for the container	--port=23456

Examples

apiVersion: batch.volcano.sh/v1alpha1
kind: Job
metadata:
  name: pytorch-job
spec:
  minAvailable: 1
  schedulerName: volcano
  plugins:
    pytorch: ["--master=master","--worker=worker","--port=23456"] # Pytorch plugin register
  tasks:
    - replicas: 1
      name: master
      policies:
        - event: TaskCompleted
          action: CompleteJob
      template:
        spec:
          containers:
            - image: gcr.io/kubeflow-ci/pytorch-dist-sendrecv-test:1.0
              imagePullPolicy: IfNotPresent
              name: master
          restartPolicy: OnFailure
    - replicas: 2
      name: worker
      template:
        spec:
          containers:
            - image: gcr.io/kubeflow-ci/pytorch-dist-sendrecv-test:1.0
              imagePullPolicy: IfNotPresent
              name: worker
              workingDir: /home
          restartPolicy: OnFailure