Diffusion language models differ in parallel text generation speed
Parallelism, critical windows, and separations among diffusion language models
Machine LearningData Structures and Algorithms
Summary
Generating text can be done in different ways by AI models called diffusion language models. Some methods let the models create text much faster by doing many parts in parallel, unlike older models that generate one word at a time. The paper shows that two popular diffusion approaches can achieve faster generation depending on the complexity of the language data, while another approach (masked diffusion) takes more steps in some cases. The key difference comes from how the methods decide and narrow down which parts of the text to generate at once.
What this means in practice
- •For language model engineers: Design faster text generation algorithms by choosing diffusion models with better parallelism based on task complexity.
- •For ai infrastructure teams: Optimize hardware use by selecting appropriate diffusion sampling methods that reduce necessary computation steps.
A theory result. No direct application yet.
Authors
Sitan Chen, Liye Wang
Abstract
A popular selling point of diffusion large language models (dLLMs) is their capacity for parallelism: the ability to generate sequences of text far more efficiently than autoregressive models, which require one forward pass per token. Yet among the many competing paradigms for dLLMs, from masked to uniform to Gaussian diffusion, principled understanding of how these different proposals compare in parallelism remains limited. In this work, we initiate a fine-grained comparison of the capacity for parallelism among these three leading approaches and prove the following: - Uniform and Gaussian diffusion can sample in a number of forward passes which scales with the dual total correlation of the underlying distribution, a measure of intrinsic complexity which can be much smaller than the context length. Previously, it was only known how to achieve this using masked diffusion. - For a certain family of random empirical measures, we show that $\widetildeΘ(\sqrt{d})$ forward passes are necessary and sufficient to sample using uniform or Gaussian diffusion, yet there exist approximate score oracles for which $\widetildeΩ(d)$ forward passes are needed for masked diffusion. This establishes the first provable separation in parallelism between the three prevailing dLLM paradigms. Contrary to popular intuition that masked diffusions are harder to parallelize because they must commit to token values, the latter separation instead comes from the fact that the critical windows in masked diffusion sampling are asymptotically narrower than those in uniform and Gaussian diffusion sampling.