Show HN: a Rust OS kernel built for LLM inference
Axiom is a Rust-based operating system kernel designed specifically for LLM inference workloads, aiming to optimize performance and efficiency for running large language models at the system level.
Background
- This is a hobbyist project (shown on Hacker News as "Show HN") that builds a tiny operating system kernel in the Rust programming language, designed specifically to run large language model (LLM) inference — the computation needed to generate text from models like GPT.
- Most LLM inference today runs on top of full operating systems (Linux, Windows) with many layers of abstraction. The idea here is to strip away everything except what the neural network needs, potentially improving performance or reducing latency.
- Rust is a systems language known for memory safety without a garbage collector; it is increasingly used for OS kernels (e.g., Redox OS) and for AI infrastructure (e.g., Hugging Face's candle, Burn).
- The project is not a production-ready platform — it is an experimental demonstration of how minimal a runtime could be for neural network computation, similar in spirit to earlier "unikernel" projects (e.g., MirageOS) but targeting AI workloads.