Sparse-LaViDa: Sparse Multimodal Discrete Diffusion Language Models

TL;DR

Sparse-LaViDa significantly accelerates masked discrete diffusion models by dynamically truncating redundant tokens during inference, maintaining quality while achieving up to 2x speedup across multimodal tasks.

Contribution

It introduces a dynamic token truncation method and specialized attention mechanisms to improve inference speed in discrete diffusion models without sacrificing quality.

Findings

Achieves up to 2x faster inference speed.

Maintains high-quality generation across tasks.

Compatible with state-of-the-art LaViDa-O model.

Abstract

Masked Discrete Diffusion Models (MDMs) have achieved strong performance across a wide range of multimodal tasks, including image understanding, generation, and editing. However, their inference speed remains suboptimal due to the need to repeatedly process redundant masked tokens at every sampling step. In this work, we propose Sparse-LaViDa, a novel modeling framework that dynamically truncates unnecessary masked tokens at each inference step to accelerate MDM sampling. To preserve generation quality, we introduce specialized register tokens that serve as compact representations for the truncated tokens. Furthermore, to ensure consistency between training and inference, we design a specialized attention mask that faithfully matches the truncated sampling procedure during training. Built upon the state-of-the-art unified MDM LaViDa-O, Sparse-LaViDa achieves up to a 2x speedup across diverse tasks including text-to-image generation, image editing, and mathematical reasoning, while maintaining generation quality.