Saved in:
Bibliographic Details
Main Authors: Goyal, Satyam, Patel, Kushal, Mittal, Tanush, Laxman, Arjun
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2604.05250
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913009400545280
author Goyal, Satyam
Patel, Kushal
Mittal, Tanush
Laxman, Arjun
author_facet Goyal, Satyam
Patel, Kushal
Mittal, Tanush
Laxman, Arjun
contents Masked Diffusion Models (MDMs) offer a promising alternative to autoregressive language models by enabling parallel token generation and bidirectional context modeling. However, their inference speed is significantly limited by the inability to cache key-value pairs due to bidirectional attention, requiring $O(N^2)$ computations at each generation step. While recent methods like FastDLLM and DkvCache improve inference speed through attention approximations and caching strategies, they achieve speedups at the cost of generation quality. We propose DualDiffusion, a speculative decoding framework for MDMs that combines fast drafter models (using efficient approximations) with slower, more accurate verifier models. By running multiple steps of a lightweight drafter followed by a single verification step, DualDiffusion achieves a superior Pareto frontier between generation steps and accuracy compared to existing approaches. We evaluate our method on MMLU and GSM8K, demonstrating that DualDiffusion maintains high accuracy while reducing the number of generation steps required, effectively pushing the quality-efficiency trade-off curve for masked diffusion language models.
format Preprint
id arxiv_https___arxiv_org_abs_2604_05250
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle DualDiffusion: A Speculative Decoding Strategy for Masked Diffusion Models
Goyal, Satyam
Patel, Kushal
Mittal, Tanush
Laxman, Arjun
Machine Learning
Computation and Language
Masked Diffusion Models (MDMs) offer a promising alternative to autoregressive language models by enabling parallel token generation and bidirectional context modeling. However, their inference speed is significantly limited by the inability to cache key-value pairs due to bidirectional attention, requiring $O(N^2)$ computations at each generation step. While recent methods like FastDLLM and DkvCache improve inference speed through attention approximations and caching strategies, they achieve speedups at the cost of generation quality. We propose DualDiffusion, a speculative decoding framework for MDMs that combines fast drafter models (using efficient approximations) with slower, more accurate verifier models. By running multiple steps of a lightweight drafter followed by a single verification step, DualDiffusion achieves a superior Pareto frontier between generation steps and accuracy compared to existing approaches. We evaluate our method on MMLU and GSM8K, demonstrating that DualDiffusion maintains high accuracy while reducing the number of generation steps required, effectively pushing the quality-efficiency trade-off curve for masked diffusion language models.
title DualDiffusion: A Speculative Decoding Strategy for Masked Diffusion Models
topic Machine Learning
Computation and Language
url https://arxiv.org/abs/2604.05250