sparkrun provides a unified CLI for managing LLM inference workloads on DGX Spark systems.
Option Description -v / --verboseIncrease verbosity (stackable: -v, -vv, -vvv) -q / --quietSuppress all output except errors (for scripting) --versionShow version and exit --helpShow help for any command
Flag Level Output (default) PROGRESS Phase and step output only -vINFO Adds detail lines -vvVERBOSE Adds timestamps and logger names -vvvDEBUG Full SSH command output, script diagnostics -qWARNING Errors and warnings only (for scripting)
Command Description sparkrun run <recipe>Launch an inference workload sparkrun build <recipe>Prepare assets without inference; advanced command sparkrun stop [target]Stop a running workload (recipe name or workload ID) sparkrun logs <target>Re-attach to workload logs (recipe name or workload ID) sparkrun statusShow running sparkrun containers (alias for cluster status) sparkrun benchmark <recipe>Benchmark an inference recipe (auto: run → benchmark → stop) sparkrun benchmark <recipe> --arenaBenchmark and submit results to Spark Arena sparkrun updateUpdate core, compatible managed plugins, and registries; --no-update-plugins skips plugin fetching
Command Description sparkrun list [query]List available recipes (alias) sparkrun show <recipe>Show recipe details + VRAM estimate (alias) sparkrun search <query>Search recipes (alias) sparkrun recipe list [query]List available recipes from all registries sparkrun recipe show <recipe>Show detailed recipe information sparkrun recipe search <query>Search for recipes by name, model, or description sparkrun recipe validate <recipe>Validate a recipe file sparkrun recipe vram <recipe>Estimate VRAM usage for a recipe sparkrun registry update [NAME]Refresh all enabled recipe registries, or one named registry, from git
Tip
Recipes can be local names, file paths, URLs, or Spark Arena shortcuts (@spark-arena/<recipe-id>). For example: sparkrun run @spark-arena/abc123 or sparkrun show @spark-arena/abc123.
Command Description sparkrun registry listList configured recipe registries sparkrun registry add <url>Add a registry from a repo manifest sparkrun registry remove <name>Remove a recipe registry sparkrun registry enable <name>Enable a recipe registry sparkrun registry disable <name>Disable a recipe registry sparkrun registry update [name]Update registries from git sparkrun registry list-benchmark-profilesList available benchmark profiles sparkrun registry show-benchmark-profile <name>Show benchmark profile details sparkrun registry trust <name>Trust a registry’s recipes (skip hook confirmation) sparkrun registry untrust <name>Revoke a registry’s trust sparkrun registry export-metadataExport registry metadata sparkrun registry revert-to-defaultsRestore the built-in default registry list
Spark Arena is included as a default registry. See Registries for details.
Command Description sparkrun cluster create <name>Create a new named cluster sparkrun cluster update <name>Update hosts, description, or user sparkrun cluster listList all saved clusters sparkrun cluster show <name>Show details of a saved cluster sparkrun cluster delete <name>Delete a saved cluster sparkrun cluster set-default <name>Set the default cluster sparkrun cluster unset-defaultRemove the default cluster setting sparkrun cluster defaultShow the current default cluster sparkrun cluster statusShow sparkrun containers running on cluster hosts sparkrun cluster check-job <target>Check if a job is running (exit 0 = running, 1 = not) sparkrun cluster monitorLive host metrics with sensor history and interactive workload logs with search and export sparkrun cluster inspect [name]Show effective cluster config and check remote cache dirs sparkrun cluster import svd <path>Import a spark-vllm-docker .env file into a cluster (aliased as eugr)
Command Description sparkrun export recipe <recipe>Export a normalized recipe to stdout or file sparkrun export recipe <recipe> --realizeResolve hardware-specific settings and pin image/model references sparkrun export running-recipe <target>Export the effective recipe from a running workload sparkrun export systemd <target>Generate (and optionally install) a systemd service for persistent inference
See export for details.
Command Description sparkrun tune sglang <recipe>Tune SGLang fused MoE or dense block-FP8 kernels sparkrun tune vllm <recipe>Tune vLLM fused MoE or dense FP8 kernels
See tune for details.
Command Description sparkrun proxy startStart the selected gateway (LiteLLM by default) sparkrun proxy stopStop the running proxy sparkrun proxy statusShow proxy status and registered models sparkrun proxy systemd installInstall and enable a Linux proxy service sparkrun proxy systemd statusInspect the service and recent journal output sparkrun proxy systemd uninstallStop, disable, and remove the service sparkrun proxy modelsList models registered with the proxy sparkrun proxy load <recipe>Load a model via sparkrun run and register with proxy sparkrun proxy unload <recipe>Stop a model and remove from proxy sparkrun proxy alias add <name> <model>Add a model alias sparkrun proxy alias remove <name>Remove a model alias sparkrun proxy alias listList configured aliases
See proxy for details.
These commands manage Spark Arena authentication.
Command Description sparkrun arena loginAuthenticate with Spark Arena via OAuth sparkrun arena logoutRemove stored Spark Arena credentials sparkrun arena statusShow Spark Arena login status
Use sparkrun benchmark <recipe> --arena to run and submit benchmarks. See the
Spark Arena tutorial for the submission workflow and
Spark Arena for the leaderboard.
Command Description sparkrun setup installInstall sparkrun as a uv tool + tab-completion + registry sync sparkrun setup completionInstall shell tab-completion (bash/zsh/fish) sparkrun setup updateUpdate core, plugins, and registries; supports --no-update-plugins and --no-update-registries sparkrun setup sshSet up passwordless SSH mesh across hosts sparkrun setup cx7Detect and configure ConnectX-7 NICs across hosts sparkrun setup fix-permissionsFix root-owned HF cache files on cluster hosts sparkrun setup clear-cacheDrop Linux page cache on cluster hosts to free memory sparkrun setup earlyoomInstall and configure earlyoom OOM protection on cluster hosts sparkrun setup wizardGuided setup wizard (cluster, SSH, CX7, Docker group, sudoers, earlyoom) sparkrun setup checkNon-destructive readiness probe of a cluster’s hosts sparkrun setup diagnoseCollect hardware/firmware/network/Docker diagnostics sparkrun setup docker-groupAdd the SSH user to the docker group on cluster hosts (usermod -aG docker) sparkrun setup fe-system-updateFull system update on DGX Spark Founders Edition hosts (apt + firmware + reboot) sparkrun setup telemetryView or change anonymous telemetry settings sparkrun setup featuresManage feature flags (list/enable/disable/reset) sparkrun setup uninstall [cluster]Reverse the wizard’s changes for a cluster, in reverse order
sparkrun setup tailscale (publish an inference endpoint on a tailnet) ships
gated behind a feature flag and is hidden until enabled. See
Feature Flags .
Show advanced help with SPARKRUN_ADVANCED=1. Feature gates remain separate from
visibility. See What’s New in 0.4.0 for release highlights.