smolcluster: distributed LLM training (DDP, FSDP, ZeRO) from scratch in plain Python sockets
East-Muffin-6472 · reddit · 2026-11-21
A developer open-sourced smolcluster, an educational project implementing core distributed training/inference algorithms from scratch using only basic Python and sockets: DDP, Pipeline Parallelism, ZeRO, Expert Parallelism, FSDP and Elastic Data Parallel. The library clusters heterogeneous local hardware — Mac minis, Jetson Nanos, Raspberry Pis, phones/tablets — for local AI training and inference. Still in development, education-focused; ideal for learners who want full control over every detail of distributed training.
More from coding & agent
- AI gateways are becoming the key control layer in production AI stacks: 8 capabilities — goyalshaliniuk · 2026-11-21
- Mnemos 2.0: agent memory designed to preserve what matters for the next session — RileyRalmuto · 2026-11-21
- InferenceBench ranks agents on LLM serving speedup: Claude Opus 5.5 tops at 12.08x, agents lose to simple hyperparameter search — maksym_andr · 2026-11-21
- 4 architecture patterns to tame multi-agent system reliability, from a 27-year reliability engineer — bibryam · 2026-11-21
- AutoCut + FrameAtlas: open-source local-first video editing driven by Claude Code via MCP — moheshmohan · 2026-11-21
- InfoQ: What an agent harness is, and two ways to build one — bibryam · 2026-11-21