lep raycluster

Manage Ray clusters on DGX Cloud Lepton.

Manage Ray clusters on DGX Cloud Lepton.

Usage

lep raycluster [OPTIONS] COMMAND [ARGS]...

Options

  • --help : Show this message and exit.

Commands

  • create : Creates a Ray cluster from either a spec file (spec only) or...
  • get : Shows Ray cluster detail and optionally saves its spec JSON.
  • list : Lists all Ray clusters in the current workspace.
  • remove : Removes a Ray cluster.
  • submit : Submits a job to a Ray cluster.
  • update : Updates a Ray cluster worker group's min replicas.

lep raycluster list

Lists all Ray clusters in the current workspace.

Usage

lep raycluster list [OPTIONS]

Options

  • -n, --name TEXT : Filter rayclusters by name (case-insensitive substring). Can be specified multiple times.
  • --help : Show this message and exit.

lep raycluster create

Creates a Ray cluster from either a spec file (spec only) or CLI flags. If both are provided, CLI flags override values from the file.

Usage

lep raycluster create [OPTIONS]

Options

  • -n, --name TEXT : Name of the Ray cluster. [required]
  • -f, --file FILE : If specified, load the Ray cluster spec from this JSON file before applying additional CLI overrides. The file should contain a LeptonRayClusterUserSpec JSON produced by lep raycluster get -p (spec only).
  • --image TEXT : Ray cluster container image. Default: ray:2.48.0-py312-gpu
  • --image-pull-secrets TEXT : Secrets to use for pulling images.
  • --ray-version TEXT : Ray version to use.
  • --head-resource-shape TEXT : Resource shape for the head node group.
  • --head-shared-memory-size INTEGER : Shared memory size for the head node group, in MiB.
  • --head-mount TEXT : Persistent storage to mount to the head group, as FROM_PATH:MOUNT_PATH:VOLUME (split on the first two colons, so VOLUME may itself contain a colon). VOLUME is node-local, or node-<type>:<storage_name> for a named volume (e.g. node-nfs:my-nfs). Examples: /data:/mnt/data:node-local or /hf-cache:/root/.cache/huggingface:node-nfs:my-nfs.
  • -he, --head-env TEXT : Environment variables for the head group, in the format NAME=VALUE.
  • -hs, --head-secret TEXT : Secrets for the head group, in the format NAME=SECRET_NAME. If secret name equals the environment variable name, you can just pass SECRET_NAME.
  • --head-node-group TEXT : Dedicated node group(s) for the head node group (affinity). Only the first may be applied depending on backend support.
  • --head-allowed-nodes TEXT : Comma-separated node names within the specified head dedicated node group (affinity).
  • --head-reservation TEXT : Reservation ID for the head node group.
  • --head-allow-burst-to-other-reservation BOOLEAN : Allow the head node group to burst to other reservations.
  • --worker-group-name TEXT : Name of the worker group (if specifying via flags).
  • --worker-resource-shape TEXT : Resource shape for the worker node group.
  • --worker-shared-memory-size INTEGER : Shared memory size for the worker node group, in MiB.
  • --worker-mount TEXT : Persistent storage to mount to the worker group, as FROM_PATH:MOUNT_PATH:VOLUME (split on the first two colons, so VOLUME may itself contain a colon). VOLUME is node-local, or node-<type>:<storage_name> for a named volume (e.g. node-nfs:my-nfs). Examples: /data:/mnt/data:node-local or /hf-cache:/root/.cache/huggingface:node-nfs:my-nfs.
  • -we, --worker-env TEXT : Environment variables for the worker group, in the format NAME=VALUE.
  • -ws, --worker-secret TEXT : Secrets for the worker group, in the format NAME=SECRET_NAME. If secret name equals the environment variable name, you can just pass SECRET_NAME.
  • --worker-node-group TEXT : Dedicated node group(s) for the worker node group (affinity). Only the first may be applied depending on backend support.
  • --worker-allowed-nodes TEXT : Comma-separated node names within the specified worker dedicated node group (affinity).
  • --worker-reservation TEXT : Reservation ID for the worker node group.
  • --worker-allow-burst-to-other-reservation BOOLEAN : Allow the worker node group to burst to other reservations.
  • --worker-min-replicas INTEGER : Minimum replicas for the worker node group. Default: 1
  • --worker-max-replicas INTEGER : Maximum replicas for the worker node group.
  • --enable-autoscaler BOOLEAN : Enable the Ray autoscaler.
  • --autoscaler-worker-idle-timeout INTEGER : Timeout for worker idle timeout in seconds.
  • --visibility TEXT : Visibility of the Ray cluster. Can be 'public' or 'private'. If private, the cluster will only be viewable by the creator and workspace admin.
  • --help : Show this message and exit.

lep raycluster remove

Removes a Ray cluster.

Usage

lep raycluster remove [OPTIONS]

Options

  • -n, --name TEXT : The raycluster name to remove. [required]
  • --help : Show this message and exit.

lep raycluster get

Shows Ray cluster detail and optionally saves its spec JSON.

Usage

lep raycluster get [OPTIONS]

Options

  • -n, --name TEXT : The raycluster name to get. [required]
  • -d, --detail : Show the full raycluster detail.
  • -p, --path PATH : Optional local path to save the raycluster spec JSON. Directory or full filename accepted. If a directory is provided, the file will be saved as raycluster-spec-[name].json.
  • --help : Show this message and exit.

lep raycluster update

Updates a Ray cluster worker group's min replicas. The worker group name is inferred from the existing cluster when there is exactly one worker group.

Usage

lep raycluster update [OPTIONS]

Options

  • -n, --name TEXT : The raycluster name to update. [required]
  • --min-replicas INTEGER : New minimum replicas for the worker group. [required]
  • --help : Show this message and exit.

lep raycluster submit

Submits a job to a Ray cluster.

Usage: lep raycluster submit -n [cluster] -- [entrypoint command] Everything after "--" is treated as the entrypoint command, just like native Ray.

Usage

lep raycluster submit [OPTIONS] [ENTRYPOINT]...

Options

  • -n, --name TEXT : The raycluster name to submit a job to. [required]
  • --submission-id TEXT : Submission ID to specify for the job.
  • --runtime-env FILE : Path to a YAML file containing a runtime_env definition.
  • --runtime-env-json TEXT : JSON-serialized runtime_env dictionary.
  • --working-dir TEXT : Directory or remote URI (.zip) for working_dir. If specified, overrides the option in --runtime-env.
  • --metadata-json TEXT : JSON-serialized dictionary of metadata to attach to the job.
  • --entrypoint-num-cpus FLOAT : CPU cores to reserve for entrypoint.
  • --entrypoint-num-gpus FLOAT : GPUs to reserve for entrypoint.
  • --entrypoint-memory INTEGER : Memory (bytes) to reserve for entrypoint.
  • --entrypoint-resources TEXT : JSON-serialized dict of custom resources to reserve for entrypoint.
  • --no-wait : Do not stream logs or wait for job completion.
  • --help : Show this message and exit.

Corporate Info

NVIDIA Developer

Resources

Copyright @ 2026, NVIDIA Corporation.