Saved in:
Bibliographic Details
Main Authors: Lu, Wenbo, Zheng, Shaoyi, Xia, Yuxuan, Wang, Shengjie
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2509.10918
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912739293659136
author Lu, Wenbo
Zheng, Shaoyi
Xia, Yuxuan
Wang, Shengjie
author_facet Lu, Wenbo
Zheng, Shaoyi
Xia, Yuxuan
Wang, Shengjie
contents Diffusion models excel in high-fidelity image generation but face scalability limits due to transformers' quadratic attention complexity. Plug-and-play token reduction methods like ToMeSD and ToFu reduce FLOPs by merging redundant tokens in generated images but rely on GPU-inefficient operations (e.g., sorting, scattered writes), introducing overheads that negate theoretical speedups when paired with optimized attention implementations (e.g., FlashAttention). To bridge this gap, we propose Token Merge with Attention (ToMA), an off-the-shelf method that redesigns token reduction for GPU-aligned efficiency, with three key contributions: 1) a reformulation of token merge as a submodular optimization problem to select diverse tokens; 2) merge/unmerge as an attention-like linear transformation via GPU-friendly matrix operations; and 3) exploiting latent locality and sequential redundancy (pattern reuse) to minimize overhead. ToMA reduces SDXL/Flux generation latency by 24%/23%, respectively (with DINO $Δ< 0.07$), outperforming prior methods. This work bridges the gap between theoretical and practical efficiency for transformers in diffusion. Code available at https://github.com/WenboLuu/ToMA.
format Preprint
id arxiv_https___arxiv_org_abs_2509_10918
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle ToMA: Token Merge with Attention for Diffusion Models
Lu, Wenbo
Zheng, Shaoyi
Xia, Yuxuan
Wang, Shengjie
Machine Learning
Artificial Intelligence
Diffusion models excel in high-fidelity image generation but face scalability limits due to transformers' quadratic attention complexity. Plug-and-play token reduction methods like ToMeSD and ToFu reduce FLOPs by merging redundant tokens in generated images but rely on GPU-inefficient operations (e.g., sorting, scattered writes), introducing overheads that negate theoretical speedups when paired with optimized attention implementations (e.g., FlashAttention). To bridge this gap, we propose Token Merge with Attention (ToMA), an off-the-shelf method that redesigns token reduction for GPU-aligned efficiency, with three key contributions: 1) a reformulation of token merge as a submodular optimization problem to select diverse tokens; 2) merge/unmerge as an attention-like linear transformation via GPU-friendly matrix operations; and 3) exploiting latent locality and sequential redundancy (pattern reuse) to minimize overhead. ToMA reduces SDXL/Flux generation latency by 24%/23%, respectively (with DINO $Δ< 0.07$), outperforming prior methods. This work bridges the gap between theoretical and practical efficiency for transformers in diffusion. Code available at https://github.com/WenboLuu/ToMA.
title ToMA: Token Merge with Attention for Diffusion Models
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2509.10918