bootgpt

What is bootgpt?

bootgpt is an open-source project that runs a character-level neural network directly in a computer master boot record. The entire inference engine fits into 510 bytes of hand-written x86 assembly. When the system starts, it loads an 11.5 kilobyte quantized model from the following disk sectors and executes it in 16-bit real mode without an operating system or C runtime.

The model is trained specifically to predict sequences of text that resemble Linux kernel boot messages. It uses char embeddings, a single hidden layer, and softmax output, with all weights quantized to int8 using power-of-two scales. This allows the boot sector to perform forward passes using basic integer multiplication and division instructions.

Development relies on Python scripts to generate training data, train the network, and verify the assembly math. Users can build and run the project using NASM and QEMU, or write the disk image to a USB drive for physical hardware testing. The project is released under the MIT license.

bootgpt features

• Runs a neural network directly in a 512-byte master boot record
• Loads an 11.5 KB int8 quantized model from adjacent disk sectors
• Executes inference in 16-bit real mode without an operating system
• Trained to generate realistic Linux kernel boot log sequences
• Verifies assembly math against Python simulations before building
• Compatible with QEMU emulation and physical BIOS hardware

Who uses bootgpt and for what

• Experimenting with extreme model compression and sizecoding techniques
• Testing how neural networks behave in constrained 16-bit environments
• Creating custom boot screen experiences or demo scene visuals
• Studying the intersection of low-level assembly and machine learning
• Building educational examples of bare-metal AI inference

bootgpt FAQ

Comments (0)

Comments and upvotes open when this product launches on ProductPerch.