SmolCluster: Distributed Training Library for Heterogeneous Hardware Using Raw Sockets

retr0jirachi · x · 2026-09-01

SmolCluster is a distributed deep learning library designed for heterogeneous hardware, built to help users learn distributed training and inference from the ground up. It runs across Mac minis, Raspberry Pis, MacBooks, and Windows machines using only Python's socket library, avoiding dependencies like NCCL or MPI. The library supports various algorithms including FSDP, Classic Data Parallelism, and Model Parallelism. It is explicitly educational, prioritizing code readability over production speed.

Original post →

More from Infra

Infra channel →