FedML Parrot Examples

Simulation with a Single Process (Standalone)

• sp_fedavg_mnist_lr_example: Simulating FL using a single process in your personal laptop or server. This is helpful when researchers hope to try a quick algorithmic idea in small synthetic datasets (MNIST, shakespeare, etc.) and small models (ResNet-18, Logistic Regression, etc.).

Simulation with Message Passing Interface (MPI)

• mpi_torch_fedavg_mnist_lr_example: MPI-based Federated Learning for cross-GPU/CPU servers.

Simulation with NCCL-based MPI (the fastest training)

• If your cross-GPU bandwidth is high (e.g., InfiniBand, NVLink, EFA, etc.), we suggest using this NCCL-based MPI FL simulator to accelerate your development.