Krlntn builds neural network that boots PC from 512‑byte MBR
The project is a neural network that runs in 16‑bit real mode on a 386, loaded from a 512‑byte MBR. The author, krlntn, writes hand‑crafted assembly that loads an 11.5‑KB int8 model from the next 23 sectors. The model is a single‑hidden‑layer char‑level net trained on fake dmesg logs. The boot sector prints a simulated Linux boot log, including timestamps, PCI bars, ACPI events, and a command…
Key points
- krlntn loads an 11.5‑KB int8 model from 23 sectors and runs it in 16‑bit real mode on a 386.
- The boot sector prints a simulated Linux boot log with timestamps, PCI bars, ACPI events, and a command line.
- Scripts include corpusgen.py for fake logs, train.py for training, boot.asm for runtime, and record.py to capture a demo GIF from QEMU.
The author provides scripts: corpusgen.py creates 3 MB of fake logs, train.py trains and quantizes the model, boot.asm contains the runtime, and record.py captures a demo GIF from QEMU. The project references earlier boot‑sector AI demos such as sectorllm, sectorlisp‑nn, and bootgpt, and cites Oscar Toledo’s 512‑byte games. The demonstration shows that a tiny neural net can run on bare hardware without an OS, highlighting the limits of inference on constrained devices.
Show HN: A language model that lives in a boot sector
github.com · 31 October 2026Loading the full article…
This text was published by github.com. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
Coverage and discussion
1source- Hacker News discussion · 43 pointsnews.ycombinator.com
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Generative AI & Models
All →- NVIDIA releases MotionBricks AI for Real-Time game and robot movement · 1 src
- Anthropic removes Opus 4.8 fast mode on November 30 · 2 src
- Mistral releases Mistral Large 4 (Le Chonk) 1-trillion-parameter multimodal model on Hugging Face · 12 src
- Google announces Gemini 4 Carbon, its latest frontier AI model for cybersecurity · 7 src
- LiquidAI releases browser demo for LFM3-Omni voice AI · 1 src
Comments
via GitHub Discussions