📋

Quick Start Templates

⚙️

General Settings

Unique identifier for this federation run
Enable production mode (disables debug features)
Compute device for training
GPU index (e.g., '0', '1', or '0:3' for multi-GPU)
Random seed for reproducibility
Deep learning framework to use
Ray placement group strategy for resource allocation
🔗

Federated Learning Schema

Type of federated learning architecture
Network topology for client connections
Visualize the network topology
Custom adjacency matrix file (for custom topology)
Number of neighbors per client (for k_connect)
Role of clients in the federation
🧠

Model Configuration

Type of neural network model
Size of transformer model (if applicable)
Use pretrained model weights
📊

Dataset Configuration

Dataset for training
Data distribution across clients
Custom desired distribution (leave empty for null)
Dirichlet distribution parameter (lower = more heterogeneous)
Input image size after transformation
🎯

Training Parameters

Learning rate for optimization
Loss function for training
Optimization algorithm
Weight decay (L2 regularization), 0 for null/disabled
Local epochs per federated round
Training batch size
Testing batch size
🌐

Federation Settings

Total number of federated clients
Fraction of clients participating per round
Total federated learning rounds
Stop training when this accuracy is reached
🔄

Aggregation Strategy

Aggregation algorithm
FedAvg variant or enable/disable
Distance metric for model comparison
Compute distance on model parameters
Use dynamic sensitivity for aggregation
Sensitivity percentage (send X% of most important chunks)
Remove common IDs during aggregation
🎨

Clustering

Use pre-computed clustering
Enable dynamic clustering
Rounds between clustering updates
💾

Model Saving

Save models before aggregation
Save global models after aggregation
Export mean accuracy to CSV
🔒

Differential Privacy

Enable differential privacy
Privacy budget (epsilon)
Privacy parameter (delta)
Gradient clipping norm
Noise multiplier for DP
🧩

Model Chunking

Enable model chunking
Include gradients in chunks (MUST be true when chunking=true)
Number of chunks to divide model into
Use random chunk selection (false = importance-based)
📈

Shapley Values

Compute Shapley values for contribution
Shapley value computation method
📡

Ray Dashboard

Enable Ray dashboard for monitoring
Port for Ray dashboard
📊

Metrics Configuration

Track round-level metrics
Track memory usage metrics
Track performance metrics
Track communication metrics
Track system metrics
Track convergence metrics
Track throughput metrics
Track availability metrics