Heterogeneous Decentralized Diffusion Models

TL;DR

This paper introduces a resource-efficient decentralized training framework for diffusion models that supports heterogeneous objectives, significantly reducing compute and data requirements while improving quality and diversity.

Contribution

It presents a novel heterogeneous decentralized training paradigm, checkpoint conversion method, and an efficient architecture, enabling flexible, scalable, and resource-efficient diffusion model training.

Findings

Reduced compute from 1176 to 72 GPU-days (16x)

Achieved better FID and diversity metrics than homogeneous baselines

Enabled mixed objectives without synchronization, lowering infrastructure needs

Abstract

Training frontier-scale diffusion models often requires substantial computational resources concentrated in tightly coupled clusters, limiting participation to well-resourced institutions. While Decentralized Diffusion Models (DDM) enable training multiple experts in isolation, existing approaches require 1176 GPU-days and homogeneous training objectives across all experts. We present an efficient framework that reduces resource requirements while supporting heterogeneous training objectives. Our approach combines three contributions: (1) a heterogeneous decentralized training paradigm that allows experts to use different objectives (DDPM and Flow Matching), unified at inference time via a deterministic schedule-aware conversion into a common velocity space without retraining; (2) pretrained checkpoint conversion from ImageNet-DDPM to Flow Matching objectives, accelerating convergence and enabling initialization without objective-specific pretraining; and (3) PixArt-alpha's efficient AdaLN-Single architecture, reducing parameters while maintaining quality. Experiments on LAION-Aesthetics show that, relative to the training scale reported for prior DDM work, our approach reduces compute from 1176 to 72 GPU-days (16x) and data from 158M to 11M (14x). Under aligned inference settings, our heterogeneous 2DDPM:6FM configuration achieves better FID (11.88 vs. 12.45) and higher intra-prompt diversity (LPIPS 0.631 vs. 0.617) than the homogeneous 8FM baseline. By eliminating synchronization requirements and enabling mixed DDPM/FM objectives, our framework lowers infrastructure requirements for decentralized generative model training.