Executors

An execution unit is a (task, executor) pair. The task defines what to run; the executor defines where and how. NeMo-Run keeps these two concerns separate so you can swap executors without changing your task configuration.

Choose an executor

Pick the executor that matches your environment:

Executor	When to use	Setup cost
LocalExecutor	Prototyping, debugging, CI	None — works out of the box
DockerExecutor	Reproducible local runs, container-based workflows	Docker installed & running
SlurmExecutor	HPC clusters with Slurm and Pyxis	SSH access to a Slurm cluster
SkypilotExecutor	Multi-cloud: AWS, GCP, Azure, Kubernetes	pip install nemo_run[skypilot] + cloud credentials
DGXCloudExecutor	NVIDIA DGX Cloud via Run:ai	Pod access + PVC on DGX Cloud
LeptonExecutor	NVIDIA DGX Cloud Lepton (standard execution)	Lepton CLI installed & authenticated
KubeflowExecutor	Distributed training via Kubeflow Training Operator v2	kubectl + Kubeflow Training Operator v2
KubeRayExecutor	Ray workloads on Kubernetes	kubectl + KubeRay operator

Packager support matrix

The packager controls how your code is bundled and sent to the execution environment.

Executor	Packagers
LocalExecutor	run.Packager (passthrough)
DockerExecutor	run.Packager, run.GitArchivePackager, run.PatternPackager, run.HybridPackager
SlurmExecutor	run.Packager, run.GitArchivePackager, run.PatternPackager, run.HybridPackager
SkypilotExecutor	run.Packager, run.GitArchivePackager, run.PatternPackager, run.HybridPackager
DGXCloudExecutor	run.Packager, run.GitArchivePackager, run.PatternPackager, run.HybridPackager
LeptonExecutor	run.Packager, run.GitArchivePackager, run.PatternPackager, run.HybridPackager
KubeflowExecutor	run.Packager

See Execution — Packagers for a description of each packager.

Launcher support

The launcher controls how the process is started inside the executor.

Launcher	Flag	Description
Default	None	Direct subprocess — no special launcher
Torchrun	"torchrun" / run.Torchrun(...)	Distributed training via torchrun
Fault Tolerance	"ft" / run.core.execution.FaultTolerance(...)	NVIDIA fault-tolerant launcher
SlurmRay	"slurm_ray"	Ray cluster on Slurm (see ray.md)

See Execution — Launchers for details.