Muon Optimizer Trains nanoGPT in Just 1.23 Minutes

dianarycai · x · 2026-08-19

Researchers at the Flatiron Institute developed 'Polar Express,' a key component of the hyper-efficient AI training tool Muon. This optimization drastically improves training speed, allowing the latest Muon version to train nanoGPT in just 1.23 minutes. This is a massive leap from 4 days in 2019 (GPT predecessor) and 45 minutes in May 2024. Such efficiency gains are critical for reducing the energy and computational costs of the AI industry.

Original post →

More from Infra

Infra channel →