HarmoniCa: Harmonizing Training and Inference for Better Feature Caching in Diffusion Transformer Acceleration

TL;DR

HarmoniCa introduces a novel training-inference harmonization framework for diffusion transformers, significantly improving feature caching efficiency and reducing inference latency while maintaining high image quality.

Contribution

It proposes Step-Wise Denoising Training and an Image Error Proxy-Guided Objective to align training with inference, enhancing caching effectiveness and model performance.

Findings

Achieves over 40% latency reduction and 2.07x speedup.

Reduces training time by 25% with an image-free approach.

Demonstrates superior performance across multiple models and resolutions.

Abstract

Diffusion Transformers (DiTs) excel in generative tasks but face practical deployment challenges due to high inference costs. Feature caching, which stores and retrieves redundant computations, offers the potential for acceleration. Existing learning-based caching, though adaptive, overlooks the impact of the prior timestep. It also suffers from misaligned objectives--aligned predicted noise vs. high-quality images--between training and inference. These two discrepancies compromise both performance and efficiency. To this end, we harmonize training and inference with a novel learning-based caching framework dubbed HarmoniCa. It first incorporates Step-Wise Denoising Training (SDT) to ensure the continuity of the denoising process, where prior steps can be leveraged. In addition, an Image Error Proxy-Guided Objective (IEPO) is applied to balance image quality against cache utilization through an efficient proxy to approximate the image error. Extensive experiments across $8$ models, $4$ samplers, and resolutions from $256\times256$ to $2K$ demonstrate superior performance and speedup of our framework. For instance, it achieves over $40\%$ latency reduction (i.e., $2.07\times$ theoretical speedup) and improved performance on PixArt-$\alpha$. Remarkably, our image-free approach reduces training time by $25\%$ compared with the previous method. Our code is available at https://github.com/ModelTC/HarmoniCa.