Skip to content

DiT-MoE

A PyTorch diffusion transformer scaled to 16 billion parameters

Visit official site
Category AI models

What DiT-MoE is

The model is described as a PyTorch diffusion transformer that scales to 16 billion parameters and competes with dense networks while keeping optimised inference.

What you can do with it

  • Load the PyTorch model definitions
  • Run the supplied training or sampling scripts
  • Inspect experts with the routing analysis tool

Getting started

  • Clone or download the code from GitHub
  • Set up the environment using the README file
  • Use the scripts to train or sample

Pros & cons

✓ What we like

  • Training and sampling code are included
  • Large-scale parameter extension is supported

! What to watch out for

  • The implementation is tied to PyTorch
  • Inference is described as optimised rather than measured here

FAQ

What is included?

The feature list names model definitions, pretrained weights and training and sampling code.

What extra tool is offered?

The material lists an expert routing analysis tool.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review