Skip to content

Extension Guides

Complete guides for extending the discrete diffusion library with custom components.

Available Guides

Core Components

  1. Custom Algorithm: Implement a new discrete diffusion method

    • Inherit from base classes
    • Override key methods
    • Register and configure

  2. Custom Forward Process: Create custom noise patterns

    • Implement forward diffusion $q(x_t|x_0)$
    • Integrate with noise schedules
    • Support custom sampling

  3. Custom Noise Schedule: Define custom time-dependent noise

    • Implement $\alpha(t)$ and $\alpha'(t)$
    • Handle boundary conditions
    • Integrate with algorithms

  4. Custom Model: Add new backbone architectures

    • Define model interface
    • Support time conditioning
    • Register and configure

Next Steps

Choose the guide for the component you want to extend and follow along with the complete example. All guides assume you're working from the repository root.