Step 3: First Experiment

Deploy federated learning with notebook workflow in 5 minutes

Run a complete MNIST federated learning experiment using the interactive notebook approach.

Open the Interactive Notebook

Navigate to the MNIST example and open the interactive notebook:

# Go to the example directory
cd examples/fl_pytorch_mnist

# Start Jupyter notebook
jupyter notebook swarm.ipynb

The notebook opens in your browser with a complete federated learning workflow.

Execute the Notebook Workflow

The notebook guides you through:

Step 1: Data Preparation - Run the data partitioning cell - MNIST dataset automatically split across nodes

Step 2: Swarm Deployment - Execute the swarm deployment cell - Uses your configured credentials automatically

Step 3: Training Monitoring - Real-time progress shown in notebook cells - Training metrics and graphs update automatically

Step 4: Results Visualization - Final accuracy and loss curves displayed - Model performance across all nodes shown

Expected Output in Notebook:

🚀 Starting federated learning experiment...
📊 Data prepared: 2 node partitions
🔄 Round 1/5: Average accuracy = 0.75
🔄 Round 2/5: Average accuracy = 0.85
🔄 Round 3/5: Average accuracy = 0.91
🔄 Round 4/5: Average accuracy = 0.93
🔄 Round 5/5: Average accuracy = 0.94
✅ Training completed! Final accuracy: 94%

Stop Nodes When Finished

After your experiment completes, clean up the running nodes:

# Stop all running nodes
manta_node stop --all

Expected Output:

🛑 Stopping all running nodes...
✅ Node 0: Stopped
✅ Node 1: Stopped
🎉 All nodes stopped successfully

Congratulations!

🎉 You’ve successfully completed your first Manta experiment!

In just 10 minutes, you’ve:

✅ Created platform account and got credentials ✅ Configured cluster with connected nodes ✅ Deployed federated learning experiment using notebook workflow ✅ Trained distributed model across multiple nodes ✅ Achieved 94% accuracy through federated averaging

What’s Next?

Now you’re ready to: