统一连续生成模型

Unified Continuous Generative Models

May 12, 2025

作者: Peng Sun, Yi Jiang, Tao Lin

cs.AI

摘要

近期，连续生成模型领域取得了显著进展，其中包括多步方法（如扩散模型和流匹配，通常需要8至1000个采样步骤）以及少步方法（如一致性模型，通常仅需1至8步），这些方法展现了卓越的生成性能。然而，现有研究往往将这些方法视为独立范式，导致训练和采样方法各自为政。我们提出了一种统一的框架，用于训练、采样和分析这些模型。我们的实现——统一连续生成模型训练与采样器（UCGM-{T,S}），达到了业界领先（SOTA）的性能。例如，在ImageNet 256x256数据集上，使用675M参数的扩散变换器，UCGM-T训练的多步模型在20步内实现了1.30的FID分数，而少步模型仅需2步便达到了1.42的FID。此外，将UCGM-S应用于一个预训练模型（此前在250步时FID为1.26），仅用40步就将性能提升至1.06的FID。代码已发布于：https://github.com/LINs-lab/UCGM。

English

Recent advances in continuous generative models, including multi-step approaches like diffusion and flow-matching (typically requiring 8-1000 sampling steps) and few-step methods such as consistency models (typically 1-8 steps), have demonstrated impressive generative performance. However, existing work often treats these approaches as distinct paradigms, resulting in separate training and sampling methodologies. We introduce a unified framework for training, sampling, and analyzing these models. Our implementation, the Unified Continuous Generative Models Trainer and Sampler (UCGM-{T,S}), achieves state-of-the-art (SOTA) performance. For example, on ImageNet 256x256 using a 675M diffusion transformer, UCGM-T trains a multi-step model achieving 1.30 FID in 20 steps and a few-step model reaching 1.42 FID in just 2 steps. Additionally, applying UCGM-S to a pre-trained model (previously 1.26 FID at 250 steps) improves performance to 1.06 FID in only 40 steps. Code is available at: https://github.com/LINs-lab/UCGM.