Single Node Training

Interactive Access

See the documentation about Jupyter notebooks on how to configure and create a notebook to access one or more GPUs on a single node.

Submission-based

With a Job resource you can start your container with one or multiple GPUs.

apiVersion: batch/v1
kind: Job
metadata:
  name: job
spec:
  template:
    metadata:
      annotations:
        sidecar.istio.io/inject: "false"
      labels:
        mount-eos: "true"
        inject-oauth2-token: "true"
    spec:
        containers:
        - name: job
            image: REPLACE_WITH_YOUR_IMAGE
            command: 
              - REPLACE_WITH_COMMAND # e.g. python train.py
            resources:
              limits:
                nvidia.com/gpu: REPLACE_WITH_NUMBER_OF_GPUS # e.g. 2
        restartPolicy: Never