What changed

A developer named WollyMonjac has created a language model called Tiny 2.1, which is designed to run on older DOS-based PCs, specifically those with 486DX and Pentium processors. The model is stored on a floppy disk and requires only 540KB of conventional memory to operate. It is described as a 'tiny question-answering transformer' with 367,000 parameters, 4 layers, and 96 width. The model is trained on approximately 3,500 facts, each presented in multiple wordings. It uses 8-bit weights and occupies 380KB of space.

What this means in practice

Tiny 2.1 is a compact AI model that can run on vintage hardware, making it accessible to retro computing enthusiasts. The model is written in Turbo C 2.0 and uses a DOS interface with a clean UI featuring a conversation history pane, a neural monitor panel, and an 'Ask Tiny' prompt window. It also includes F-key shortcuts for navigation and a PC-speaker robot voice for spoken responses. The model is MIT-licensed, and users can expand its knowledge by training it on additional data through the included training folder.

Limitations and availability

The model is not a large language model (LLM) and is limited in scope, with the developer acknowledging that it 'makes things up confidently.' It is intended for educational or nostalgic purposes rather than practical use in modern computing environments. The model is available via a GitHub repository, and while it can be run on modern hardware with emulators, it is not designed for contemporary applications.

Source facts for Developer Fits Tiny AI Model on Floppy Disk for 486DX and Pentium PCs: An inspired developer has shoehorned a language model onto a floppy disk created especially for 486DX and Pentium PC users running DOS. Please note that WollyMonjac’s project isn’t a large language model (LLM). Your creaky old PC will need about 540KB of free conventional memory to run Tiny 2.1.
Original AINewsTrend explainer based on the cited source. This graphic summarizes source statements; it is not an independent benchmark.

Sources and further reading

  1. tomshardware: Dev squeezes an AI language model onto a floppy for 486DX and Pentium PC stalwarts