BlazeFL Documentation

BlazeFL is a blazing-fast, minimalist, and researcher-friendly simulation framework for Federated Learning.

• 🚀 High Performance: Optimized for single-node simulations, BlazeFL allows you to adjust the degree of parallelism for efficient resource management.

• 🧩 High Extensibility: BlazeFL focuses on core communication and parallelization interfaces, avoiding excessive abstraction to maintain flexibility.

• 🍃 Minimal Dependencies: The framework’s core relies only on PyTorch, ensuring a lightweight and straightforward setup.

• 🔄 Robust Reproducibility: Utilities for saving and restoring seed states are provided to ensure consistent results, even in multi-process and multi-threaded environments.

• 🛡️ Structured and Type-Safe by Design: By leveraging dataclasses and protocols, BlazeFL enables the creation of clear, type-safe, and self-documenting communication packages (UplinkPackage, DownlinkPackage). This design enhances code readability, maintainability, and reduces errors in FL workflows.

For a comprehensive overview, including detailed execution modes, benchmarks, and setup examples, please refer to the README.md on our GitHub repository.

• API Reference
• Contributions