Can I read CLEAR: Conv-Like Linearization Revs Pre-Trained Diffusion Transformers Up on EtoBox?

CLEAR: Conv-Like Linearization Revs Pre-Trained Diffusion Transformers Up by Liu, Songhua; Tan, Zhenxiong; Wang, Xinchao is a scholarly article available to read on EtoBox.

What is CLEAR: Conv-Like Linearization Revs Pre-Trained Diffusion Transformers Up about?

Diffusion Transformers (DiT) have become a leading architecture in image generation. However, the quadratic complexity of attention mechanisms, which are responsible for modeling token-wise relationships, results in significant latency when generating high-resolution images. To address this issue, we aim at a linear attention mechanism in this paper that reduces the complexity of pre-trained DiTs to linear. We begin our exploration with a comprehensive summary of existing efficient attention mechanisms and identify four key factors crucial for successful linearization of pre-trained DiTs: locality, formulation consistency, high-rank attention maps, and feature integrity. Based on these insights, we introduce a convolution-like local attention strategy termed CLEAR, which limits feature interactions to a local window around each query token, and thus achieves linear complexity. Our experiments indicate that, by fine-tuning the attention layer on merely 10K self-generated samples for 10K iterations, we can effectively transfer knowledge from a pre-trained DiT to a student model with linear complexity, yielding results comparable to the teacher model. Simultaneously, it reduces attent

Author

Liu, Songhua; Tan, Zhenxiong; Wang, Xinchao

Published

2024

Language

EN