OmniFed Federated Learning Framework

OmniFed is a federated learning framework built on Ray and Hydra for distributed training experiments.

Core Capabilities

• 11 built-in algorithms - FedAvg, SCAFFOLD, MOON, FedProx, DiLoCo, and others with support for custom implementations
• Multiple topologies - Centralized and hierarchical structures with support for custom patterns
• Communication backends - gRPC and TorchDist implementations with support for custom protocols
• Scalable deployment - Single machine to multi-datacenter execution
• PyTorch integration - Works with any PyTorch neural network

Installation

git clone https://github.com/sahiltyagi4/OmniFed.git
cd OmniFed

pip install -r requirements.txt
./main.sh --config-name test_fedavg_centralized_torchdist

System Components

• Topologies - Network structure definitions with support for custom patterns
• Nodes - Individual federated learning participants with local state
• Communicators - gRPC and TorchDist implementations with support for custom protocols
• Algorithms - 11 built-in implementations with support for custom algorithms
• Models - Any PyTorch module through backbone and head architecture
• DataModules - Containers for PyTorch DataLoaders with configuration examples