Commands¶
ReFedEz provides a command-line interface (CLI) with four main commands to manage your federated learning deployments. All commands operate on the refedez.yaml configuration file in the current directory by default.
start¶
Starts the federated learning deployment by building environments, provisioning NVIDIA FLARE, and launching services on all configured machines.
Usage:
What it does:
- Builds reproducible environments using Nix for each machine
- Provisions Nvidia Flare configurations
- Patches provisioned files with the built environments
- Uploads and starts services on local and remote machines via SSH
- Updates project state to running
Requirements: Project must not already be running. Requires refedez.yaml and project.yml in the current directory.
status¶
Displays the current status of the ReFedEz project, including running machines and their configurations.
Usage:
Output:
- For running projects: A table showing machine names, types (local/remote), file folders, IPs, and users
- For clean projects: Confirmation that the project is clean
- For dirty projects: Warning about old builds and unnecessary files
Requirements: Valid refedez.yaml configuration file.
stop¶
Stops all running federated learning services by signaling shutdown to each machine.
Usage:
What it does:
- Creates shutdown signal files on all machines (local and remote)
- NVIDIA FLARE services detect the shutdown file and terminate gracefully
- Updates project state to dirty (ready for cleanup)
Requirements: Project must be in a running state.
clean¶
Removes temporary build files and cleans up the project workspace.
Usage:
What it does:
- Deletes the entire
.refedezfolder and all temporary files - Resets the project to a clean state
Requirements: Project must be in a dirty state (after stopping). This command is destructive and cannot be undone.