Newshot

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

Summary

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). Though it shares the transformer model architecture of its much bigger siblings, Tiny 2.1 is humbly described as “a tiny question-answering transformer written in Turbo C 2.0.” (Image credit: WollyMonjac ) There isn’t much room for modern software on a floppy disk, with a typical PC formatted disk in the 486 era boasting just shy of 1,475,000 bytes of data capacity. Tiny 2.1 makes the best use of this limited headroom as possible, fitting in the following: 367,000 parameters: 4 layers, 96 wide, reads and writes letter by letter 8-bit weights (one scale per row), 380 KB in all Trained on about 3,500 facts, each asked in many wordings Your creaky old PC will need about 540KB of free conventional memory to run Tiny 2.1. If you do test it, we are sure you will appreciate the clean U…

Original text(via Tom's Hardware)
Floppy containing an LM
Floppy containing an LM

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). Though it shares the transformer model architecture of its much bigger siblings, Tiny 2.1 is humbly described as “a tiny question-answering transformer written in Turbo C 2.0.”

Tiny 2.1
Tiny 2.1

There isn’t much room for modern software on a floppy disk, with a typical PC formatted disk in the 486 era boasting just shy of 1,475,000 bytes of data capacity. Tiny 2.1 makes the best use of this limited headroom as possible, fitting in the following:

  • 367,000 parameters: 4 layers, 96 wide, reads and writes letter by letter
  • 8-bit weights (one scale per row), 380 KB in all
  • Trained on about 3,500 facts, each asked in many wordings

Your creaky old PC will need about 540KB of free conventional memory to run Tiny 2.1. If you do test it, we are sure you will appreciate the clean UI with its conversation history pane, neural monitor panel, and Ask Tiny prompt window. As all DOS programs should, there’s a range of F-key shortcuts highlighted within the UI. Perhaps the retro-futuristic cherry on the cake, though, is the PC-speaker robot voice which can be used to speak the responses.

We’ve already indicated that the dev seems rather humble about this DOS program. The GitHub repository says it was “Written by Fable 5.2 and trained by Opus 5.5.” Monjac also wryly admits that outside of this language model’s training, “it makes things up confidently.” We have heard less about LLM hallucinations in recent months, but here it feels almost like a fun feature.

Should any of Tiny 2.1’s knowledge gaps irk you, WollyMonjac has included a training folder in the MIT-licensed distribution. A dedicated user could probably build up a capable question-answer transformer tailored to their favorite topic(s).

Nowadays it is pretty easy to get old software to run on your shiny new hardware, one way or another. It isn’t entirely a one-way street, though; getting modern apps or games to run on hardware from decades earlier is an exciting development trend we are seeing established. In addition to the headlining Tiny 2.1, we’ve reported on similarly incongruous new-meets-old software projects like the ChatGPT client for DOS, or Minecraft running on a Game Boy, creating images on the C64 with generative AI, or playing Doom on an Atari ST.

Google Preferred Source
Google Preferred Source

Download the Tom's Hardware app for up-to-the-minute updates on iOS and Android. Also follow Tom's Hardware on Google News, or add us as a preferred source, to get our latest news, analysis, & reviews in your feeds.

Get Tom's Hardware's best news and in-depth reviews, straight to your inbox.

Mark Tyson
Mark Tyson

Mark Tyson is a news editor at Tom's Hardware. He enjoys covering the full breadth of PC tech; from business and semiconductor design to products approaching the edge of reason.

The text above was extracted automatically from the source page and may miss images, tables or code blocks. The original is authoritative.

Published by Tom's Hardware
This site only summarises and aggregates; copyright stays with the original author. Headlines and summaries follow the source — where they differ, the original wins.
Read the original ↗