Skip to main content

​
Synopsis

syft-flwr run PROJECT_DIR [OPTIONS]

​
Description

The run command executes a bootstrapped syft-flwr project in simulation mode using mock datasets. This allows you to:
  • Test your federated learning workflow locally
  • Validate model training with sample data
  • Debug issues before deploying to distributed datasites
  • Simulate multiple clients with different datasets
The simulation creates a temporary SyftBox network with all clients and the aggregator server, runs the federated learning workflow, and cleans up automatically when complete.

​
Arguments

​
PROJECT_DIR
path
required
Path to a bootstrapped syft-flwr project directory.The directory must contain:
  • pyproject.toml with syft-flwr configuration
  • main.py (created by syft-flwr bootstrap)
Both relative and absolute paths are supported. Paths with ~ are expanded.

​
Options

​
--mock-dataset-paths
string
Comma-separated list of paths to mock datasets for each client.Aliases: -mFormat: Comma-separated file system paths (e.g., ./data/client1,./data/client2)Requirements:
  • Number of paths must match the number of datasites configured in pyproject.toml
  • All paths must exist on the file system
  • Each path should contain the dataset for one client
If not provided via flag, you’ll be prompted interactively:
Enter comma-separated paths to mock datasets:
​
--help
flag
Display help information for the run command.Aliases: -h

​
Examples

​
Run with All Flags

syft-flwr run ./my-fl-project \
  --mock-dataset-paths ./data/hospital1,./data/hospital2,./data/hospital3
Output: 
Running syft_flwr project at '/path/to/my-fl-project'
Mock dataset paths: ['./data/hospital1', './data/hospital2', './data/hospital3']
[Simulation logs...]
Simulation completed successfully ✅

​
Run with Interactive Prompts

syft-flwr run ./my-fl-project
Interactive session: 
Enter comma-separated paths to mock datasets: ./data/client1,./data/client2
Running syft_flwr project at '/path/to/my-fl-project'
Mock dataset paths: ['./data/client1', './data/client2']
[Simulation logs...]
Simulation completed successfully ✅

​
Run with Short Flag

syft-flwr run ./diabetes-fl -m ~/datasets/do1,~/datasets/do2

​
Run with Absolute Paths

syft-flwr run ~/projects/fl/diabetes-prediction \
  --mock-dataset-paths /data/samples/client1,/data/samples/client2

​
Run with Current Directory

cd my-fl-project
syft-flwr run . -m ../test-data/client1,../test-data/client2

​
How It Works

​
Simulation Process

  1. Validation: Validates project structure and dataset paths
  2. Setup: Creates temporary SyftBox network directory
  3. Client Creation: Initializes mock RDS clients for each datasite
  4. Encryption Bootstrap: Sets up E2E encryption keys (if enabled)
  5. Parallel Execution: Runs server and all clients concurrently
  6. Logging: Captures output from each participant
  7. Cleanup: Removes temporary directories and configuration

​
Temporary Environment

During simulation, the following temporary directories are created: 
/tmp/
  ├── my-fl-project/           # Simulated SyftBox network
  │   ├── ds@example.com/      # Aggregator workspace
  │   ├── do1@example.com/     # Client 1 workspace  
  │   └── do2@example.com/     # Client 2 workspace
  └── .syftbox/                # Encryption keys and config
      ├── ds@example.com/
      ├── do1@example.com/
      └── do2@example.com/
All temporary files are automatically deleted when simulation completes.

​
Simulation Logs

Logs are written to PROJECT_DIR/simulation_logs/: 
my-fl-project/
  └── simulation_logs/
      ├── ds@example.com.log      # Aggregator logs
      ├── do1@example.com.log     # Client 1 logs
      └── do2@example.com.log     # Client 2 logs
These logs persist after simulation for debugging.

​
Environment Variables

The run command sets environment variables for each participant:
​
SYFTBOX_CLIENT_CONFIG_PATH
string
Path to the participant’s SyftBox client configuration file.Set automatically for each client and server process.
​
DATA_DIR
string
Path to the mock dataset for each client.Set from the corresponding --mock-dataset-paths value.
​
SYFT_FLWR_ENCRYPTION_ENABLED
boolean
default:"true"
Enable or disable E2E encryption during simulation.Set to false to disable encryption:
SYFT_FLWR_ENCRYPTION_ENABLED=false syft-flwr run ./my-project -m ./data1,./data2
​
SYFT_FLWR_SKIP_MODULE_CHECK
boolean
default:"false"
Skip module validation during project loading.Useful for testing. Not recommended for production use.

​
Validation

The command validates:
  1. Project directory exists and is a directory
  2. pyproject.toml exists with valid syft-flwr configuration
  3. main.py exists in the project directory
  4. All mock dataset paths exist on the file system
  5. Number of datasets matches number of configured datasites

​
Error Handling

​
Project Not Bootstrapped

$ syft-flwr run ./my-project
Error: main.py not found at ./my-project
Solution: Bootstrap the project first: 
syft-flwr bootstrap ./my-project --aggregator ds@example.com --datasites do1@example.com,do2@example.com

​
Dataset Path Does Not Exist

$ syft-flwr run ./my-project -m ./nonexistent,./data
Error: Mock dataset path ./nonexistent does not exist
Solution: Ensure all dataset paths exist: 
mkdir -p ./data/client1 ./data/client2
syft-flwr run ./my-project -m ./data/client1,./data/client2

​
Dataset Count Mismatch

$ syft-flwr run ./my-project -m ./data/client1
# When pyproject.toml has 2 datasites configured
Error: Expected 2 dataset paths, got 1
Solution: Provide one dataset path per datasite.

​
Simulation Failed

$ syft-flwr run ./my-project -m ./data1,./data2
Running syft_flwr project at './my-project'
Mock dataset paths: ['./data1', './data2']
[Simulation logs...]
Simulation failed ❌
Error: [error details]
Solution: Check simulation logs in ./my-project/simulation_logs/ for details.

​
Async Execution

In environments with an existing event loop (e.g., Jupyter notebooks), the run command returns an asyncio Task: 
import asyncio
from syft_flwr.run_simulation import run

# Returns a Task in Jupyter
task = run("./my-project", ["./data1", "./data2"])

# Await the task
result = await task
print(f"Success: {result}")
In regular scripts, execution is synchronous and returns a boolean.

​
Encryption

By default, the simulation uses end-to-end encryption with:
  • Automatic key generation for all participants
  • DID (Decentralized Identifier) document creation
  • Encrypted message passing between clients and server
To disable encryption for debugging: 
SYFT_FLWR_ENCRYPTION_ENABLED=false syft-flwr run ./my-project -m ./data1,./data2
You’ll see: 
⚠️ Encryption disabled - skipping key bootstrap

​
Performance

​
Parallel Execution

All clients run in parallel using asyncio, with the aggregator server coordinating the workflow. The simulation completes when the server process finishes.

​
Resource Usage

Each client runs as a separate Python subprocess. For N clients:
  • Processes: N + 1 (clients + server)
  • Memory: Depends on model size and dataset
  • Disk: Temporary files for communication and logs

​
Exit Codes

  • 0: Simulation succeeded
  • 1: Simulation failed (validation error, runtime error, etc.)

​
Best Practices

​
Dataset Preparation

Organize mock datasets clearly: 
my-fl-project/
  ├── mock-data/
  │   ├── client1/
  │   │   └── dataset.csv
  │   ├── client2/
  │   │   └── dataset.csv
  │   └── client3/
  │       └── dataset.csv
  ├── main.py
  └── pyproject.toml
Run from project root: 
syft-flwr run . -m mock-data/client1,mock-data/client2,mock-data/client3

​
Testing Before Deployment

Always run simulations before deploying to distributed datasites:
  1. Create representative mock datasets
  2. Run simulation multiple times
  3. Review logs for errors
  4. Verify model training converges
  5. Check aggregation results

​
Debugging

Enable detailed logging in your main.py: 
import logging
logging.basicConfig(level=logging.DEBUG)
Logs will appear in simulation_logs/.

​
See Also

Build docs developers (and LLMs) love

Get started for freeTalk to us