Show HN: a Rust OS kernel built for LLM inference

Category: infrastructure

Tags: rust, kernel, llm-inference, operating-system, tensor, scheduling, no-std

Score: 7.5/10 (Innovation: 8, Technical: 9, Documentation: 7, Utility: 6)

Axiom is a bootable Rust no_std kernel designed to optimize LLM inference by replacing general-purpose OS abstractions with inference-specific primitives like tensor-native memory allocation, layer-boundary scheduling, and double-buffered weight streaming. It demonstrates significant performance improvements by reducing streaming overhead from seconds to microseconds, making it a novel exploration of kernel-level inference optimization. The project is particularly interesting for its focused, bare-metal approach to addressing memory and scheduling mismatches in transformer inference.

Target audience: systems engineers, kernel developers, ML inference engineers

Repository: https://github.com/Kanchisaw03/axiom · Rust · 4 stars

View on Hacker News