Accelerated Sampling from Masked Diffusion Models via Entropy Bounded Unmasking

TL;DR

This paper introduces EB-Sampler, an efficient sampling method for masked diffusion models that unmask multiple tokens simultaneously, significantly speeding up sampling without performance loss across various language and reasoning tasks.

Contribution

The paper proposes EB-Sampler, a novel adaptive unmasking algorithm that accelerates sampling from masked diffusion models by 2-3 times while maintaining accuracy.

Findings

Speeds up sampling by 2-3x on coding and math benchmarks.

Effective on reasoning tasks like maze navigation and Sudoku.

Maintains performance despite increased sampling speed.

Abstract

Recent masked diffusion models (MDMs) have shown competitive performance compared to autoregressive models (ARMs) for language modeling. While most literature has focused on performance enhancing sampling procedures, efficient sampling from MDMs has been scarcely explored. We make the observation that often a given sequence of partially masked tokens determines the values of multiple unknown tokens deterministically, meaning that a single prediction of a masked model holds additional information unused by standard sampling procedures. Based on this observation, we introduce EB-Sampler, a simple drop-in replacement for existing samplers, utilizing an Entropy Bounded unmasking procedure that dynamically unmasks multiple tokens in one function evaluation with predefined approximate error tolerance. We formulate the EB-Sampler as part of a broad family of adaptive samplers for which we provide an error analysis that motivates our algorithmic choices. EB-Sampler accelerates sampling from current state of the art MDMs by roughly 2-3x on standard coding and math reasoning benchmarks without loss in performance. We also validate the same procedure works well on smaller reasoning tasks including maze navigation and Sudoku, tasks ARMs often struggle with.