Training a 210M text-to-image DiT from scratch on a single GPU.

Core summary

One sentence to understand this update

A researcher successfully trained a 210M-parameter text-to-image Diffusion Transformer (DiT) model from scratch on a single RTX PRO 6000 GPU over 3.5 days, processing 4.2M images.

Impact & opportunity

What this could mean

This demonstrates the feasibility of training substantial text-to-image models on accessible hardware, providing a valuable blueprint for developers with limited resources to experiment with custom generative AI.