HQ-DiT: Efficient Diffusion Transformer with FP4 Hybrid Quantization

TL;DR

This paper introduces HQ-DiT, a novel 4-bit floating-point quantization method for Diffusion Transformers, significantly reducing model size and computation with minimal performance loss, enabling deployment on resource-constrained devices.

Contribution

The paper presents the first 4-bit FP quantization approach for both weights and activations in Diffusion Transformers, improving efficiency while maintaining performance.

Findings

Achieves 4-bit quantization with only 0.12 increase in sFID on ImageNet.

Utilizes a novel clipping range selection mechanism for minimal quantization error.

Introduces a universal identity transform to reduce outlier-induced errors.

Abstract

Diffusion Transformers (DiTs) have recently gained substantial attention in both industrial and academic fields for their superior visual generation capabilities, outperforming traditional diffusion models that use U-Net. However,the enhanced performance of DiTs also comes with high parameter counts and implementation costs, seriously restricting their use on resource-limited devices such as mobile phones. To address these challenges, we introduce the Hybrid Floating-point Quantization for DiT(HQ-DiT), an efficient post-training quantization method that utilizes 4-bit floating-point (FP) precision on both weights and activations for DiT inference. Compared to fixed-point quantization (e.g., INT8), FP quantization, complemented by our proposed clipping range selection mechanism, naturally aligns with the data distribution within DiT, resulting in a minimal quantization error. Furthermore, HQ-DiT also implements a universal identity mathematical transform to mitigate the serious quantization error caused by the outliers. The experimental results demonstrate that DiT can achieve extremely low-precision quantization (i.e., 4 bits) with negligible impact on performance. Our approach marks the first instance where both weights and activations in DiTs are quantized to just 4 bits, with only a 0.12 increase in sFID on ImageNet.