Developer Open-Sources BootGPT: Running a Neural Network Within a 512-Byte Boot Sector
Show HN: 一个驻留在引导扇区中的语言模型
A developer has open-sourced BootGPT, a project that runs a neural network directly in 16-bit real mode via a hand-crafted 512-byte Master Boot Record (MBR)—requiring no operating system or C runtime. The system loads an 11.5 KB int8 character-level model from the subsequent 23 sectors, performing forward inference to generate synthetic Linux boot logs. Its execution environment relies entirely on x86 assembly and BIOS interrupts to handle weight loading, text output, and reboot triggers.
The project showcases an engineering implementation of executing a neural network within a 512-byte boot sector, offering a reusable, practical reference for assembly-level inference and model quantization in extremely resource-constrained environments.
Source:Buzzing HN · 中文精选 · github.com