Skip to content
Spike Documentation

Launch workloads with the Run:ai CLI

The Run:ai CLI is useful when a workload definition is stable enough to repeat from a script or terminal. Start with the portal while learning the fields, then translate the approved project, environment, compute resource, and data source into CLI flags.

The CLI is available only after account creation. Follow Install and verify the command-line tools before using this page.

Sign in and set the project

Section titled “Sign in and set the project”
Terminal window
runai login
runai project list
runai project set PROJECT_NAME
runai whoami

Setting the project reduces repetition, but keeping --project PROJECT_NAME in saved commands makes their target explicit. Run runai project set again when you intentionally switch projects.

Inspect the command before submitting

Section titled “Inspect the command before submitting”

The installed CLI matches the Spike Run:ai environment and can change after a platform update. Check its built-in help before adding flags:

Terminal window
runai version
runai training standard submit --help

NVIDIA maintains the complete Run:ai CLI command reference and a set of CLI examples.

Submit a standard training workload

Section titled “Submit a standard training workload”

Replace every uppercase placeholder with a value visible in your Run:ai project:

Terminal window
runai training standard submit cli-smoke-test \
  --project PROJECT_NAME \
  --environment ENVIRONMENT_NAME \
  --compute COMPUTE_RESOURCE_NAME \
  --datasource type=pvc,name=DATA_SOURCE_NAME \
  --command -- python /workspace/train.py --epochs 1

This example deliberately reuses Run:ai assets instead of rebuilding their settings as raw flags:

  • --environment selects the approved container environment.
  • --compute selects an existing compute-resource preset.
  • --datasource attaches the project data source.
  • --command -- replaces the image entrypoint with everything that follows.

Omit --datasource when the test does not need project data. Omit --command -- ... when the image’s own entrypoint is already correct. See the official runai training standard submit reference for the flags supported by the installed release.

Run:ai compute-resource selection showing synthetic preset names and neutral capacity values
The CLI's --compute value refers to the same approved compute-resource asset shown in the portal. Do not copy the synthetic names from this screenshot.

Inspect the run

Section titled “Inspect the run”
Terminal window
runai training standard list --project PROJECT_NAME
runai training standard describe cli-smoke-test --project PROJECT_NAME
runai training standard logs cli-smoke-test \
  --project PROJECT_NAME \
  --follow \
  --timestamps

Use describe to inspect the selected image, compute, pods, and events. Stop following logs with Ctrl+C; that does not stop the workload.

Suspend, resume, and remove

Section titled “Suspend, resume, and remove”
Terminal window
runai training standard suspend cli-smoke-test --project PROJECT_NAME
runai training standard resume cli-smoke-test --project PROJECT_NAME
runai training standard delete cli-smoke-test --project PROJECT_NAME

Delete only after outputs and diagnostic logs are safe. Suspending does not make ephemeral files persistent. Use the portal when you are unsure which workload a command targets.