job.yml

# Training job submission via AML CLI v2

$schema: https://azuremlschemas.azureedge.net/latest/commandJob.schema.json

command: pip list && python cifar10_deepspeed.py --deepspeed --deepspeed_config ds_config.json --deepspeed_mpi --global_rank $RANK 

experiment_name: cifar10-ds-nebulaml
environment: < your environment name > # Should replace your environment
environment_variables:
  AZUREML_COMPUTE_USE_COMMON_RUNTIME: 'True'
  AZUREML_COMMON_RUNTIME_USE_INTERACTIVE_CAPABILITY: 'True'
code: src
outputs:
  output:
    type: uri_folder
    mode: rw_mount
    path: azureml://datastores/workspaceblobstore/paths/outputs
compute: < your compute name > # Should replace your compute
distribution:
  type: pytorch 
  process_count_per_instance: 4
resources:
  instance_count: 4