CLI Usage

Main Command: torchx run

torchx run [options] <component> [component_args]

Options

| Flag | Description | |-------------------|------------------------------------------------| | -s, --scheduler | Scheduler backend (local, docker, kubernetes) | | --cfg | Scheduler configuration (key=value) | | -n, --name | Application name | | --dryrun | Print job definition without submitting |

Examples

# Local execution
torchx run -s local dist.ddp -j 1x4 --script train.py

# Kubernetes with 2 nodes × 4 procs
torchx run -s kubernetes dist.ddp -j 2x4 --script main.py

# Docker execution
torchx run -s docker dist.ddp -j 1x2 --script train.py

# Dryrun (inspect without submitting)
torchx run -s local dist.ddp -j 1x4 --script main.py --dryrun

Other Commands

torchx status <app_id>     # Check app status
torchx log <app_id>        # View logs
torchx cancel <app_id>     # Cancel running app
torchx describe <app_id>   # Detailed app info
torchx list                # List apps

Component Resolution

Components can be specified as:

# Built-in component
torchx run dist.ddp ...

# File-based component
torchx run ~/components.py:my_component ...

# Module path
torchx run my_package.components:train ...

Common Component Parameters

| Param | Description | |---------|--------------------------------------------------| | j | Job topology: {nnodes}x{nproc_per_node} | | m | Python module to run | | h | Named resource (gpu type, instance type) | | env | Environment variables | | image | Container image | | script| Script or binary to run |