An ODE-based Sampler for Discrete Diffusion Models

Samson Gourevitch, Yazid Janati, Eric Moulines, Eric P. Xing, Alain Durmus et al.|October 29, 2026arXiv

Key Takeaway

Discrete diffusion models can now use deterministic ODE-based sampling (like continuous diffusion models do), enabling faster inference and better distillation without sacrificing generation quality.

Summary

This paper develops a deterministic sampling method for discrete diffusion models by introducing exponential residual flows (ERFs), which use ODEs to guide token generation.

efficiency training architecture

Key Terms

diffusion-models probability-flow-ode consistency-distillation discrete-diffusion