Self-Speculative Masked Diffusions

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Campbell, Andrew, De Bortoli, Valentin, Shi, Jiaxin, Doucet, Arnaud
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917317006327808
author Campbell, Andrew
De Bortoli, Valentin
Shi, Jiaxin
Doucet, Arnaud
author_facet Campbell, Andrew
De Bortoli, Valentin
Shi, Jiaxin
Doucet, Arnaud
contents We present self-speculative masked diffusions, a new class of masked diffusion generative models for discrete data that require significantly fewer function evaluations to generate samples. Standard masked diffusion models predict factorized logits over currently masked positions. A number of masked positions are then sampled, however, the factorization approximation means that sampling too many positions in one go leads to poor sample quality. As a result, many simulation steps and therefore neural network function evaluations are required to generate high-quality data. We reduce the computational burden by generating non-factorized predictions over masked positions. This is achieved by modifying the final transformer attention mask from non-causal to causal, enabling draft token generation and parallel validation via a novel, model-integrated speculative sampling mechanism. This results in a non-factorized predictive distribution over masked positions in a single forward pass. We apply our method to GPT2 scale text modelling and protein sequence generation, finding that we can achieve a ~2x reduction in the required number of network forward passes relative to standard masked diffusion models.
format Preprint
id arxiv_https___arxiv_org_abs_2510_03929
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Self-Speculative Masked Diffusions
Campbell, Andrew
De Bortoli, Valentin
Shi, Jiaxin
Doucet, Arnaud
Machine Learning
We present self-speculative masked diffusions, a new class of masked diffusion generative models for discrete data that require significantly fewer function evaluations to generate samples. Standard masked diffusion models predict factorized logits over currently masked positions. A number of masked positions are then sampled, however, the factorization approximation means that sampling too many positions in one go leads to poor sample quality. As a result, many simulation steps and therefore neural network function evaluations are required to generate high-quality data. We reduce the computational burden by generating non-factorized predictions over masked positions. This is achieved by modifying the final transformer attention mask from non-causal to causal, enabling draft token generation and parallel validation via a novel, model-integrated speculative sampling mechanism. This results in a non-factorized predictive distribution over masked positions in a single forward pass. We apply our method to GPT2 scale text modelling and protein sequence generation, finding that we can achieve a ~2x reduction in the required number of network forward passes relative to standard masked diffusion models.
title Self-Speculative Masked Diffusions
topic Machine Learning
url https://arxiv.org/abs/2510.03929