Urgent.News

What's breaking now, across hundreds of outlets.

AI

One gigabyte of Wikipedia losslessly compressed to 105.39MB with self-learning AI transformer model

Tufa Labs’ tufazip compresses 1GB of Wikipedia to 105.39MB with a 163M-parameter Transformer as its only predictor, on two Nvidia B200s.

One gigabyte of Wikipedia losslessly compressed to 105.39MB with self-learning AI transformer model

A research team from Tufa Labs has compressed 1GB of Wikipedia information down to a mere 105.39MB using an innovative self-learning AI transformer model called tufazip. This groundbreaking achievement, as reported by the wire service on November 19, 2026, is the smallest result on the Large Text Compression Benchmark that relies solely on a neural network as its predictor.

By utilizing a 163-million-parameter Transformer model, tufazip was able to compress and decompress the data without any loss, demonstrating the potential of this cutting-edge technology in data-saving applications.

Rewritten by urgent.news; may contain errors — check the original.

Read the original at tomshardware.com →

Report a problem with this story

More in AI

Building Malik MediGaurd AI: An Open-Source Healthcare & Medical Bill Audit Assistant with Python

The Problem: Healthcare Information Asymmetry in India Healthcare should heal families, not push them into debt. Yet, every day across India, millions of middle-class and rural patients overpay at…

  • Malik MediGuard AI is an open-source AI tool for healthcare expense management in India.
  • Features include prescription/bill reader, generic medicine finder, and bill overcharge audit.
  • Built with Python, Streamlit, and Google Gemini AI, released under MIT License.

More from Saturday 21 November →