docs

Moodist API Documentation

Welcome to the Moodist API documentation. Moodist is a PyTorch extension library that implements a high-performance process group for distributed computing, built on RDMA.

API Reference

Core Functions

• compile_op - Compile custom collective operations for arbitrary data movement patterns between ranks
• Queue - Distributed queue for inter-rank communication of tensors and objects

Infrastructure

• TcpStore - Distributed, decentralized key-value store for process group coordination

Coming Soon

Documentation for additional Moodist features:

• MoodistProcessGroup - Process group management
• serialize / deserialize - Object serialization for distributed communication
• enable_profiling, enable_cuda_allocator, enable_cpu_allocator - Performance tuning
• cuda_copy - CUDA tensor operations

Getting Started

For installation instructions and basic usage, see the main README.

Examples

All API documentation pages include practical examples. For a comprehensive guide to compile_op usage patterns, see the compile_op examples section.