Related papers: DiTAS: Quantizing Diffusion Transformers via Enhanced Activation Smoothing

DiTAS: Quantizing Diffusion Transformers via Enhanced Activation Smoothing

URL: http://arxiv.org/abs/2409.07756v2
Date: Mon, 25 Nov 2024 01:36:31 GMT
Title: DiTAS: Quantizing Diffusion Transformers via Enhanced Activation Smoothing
Authors: Zhenyuan Dong, Sai Qian Zhang,
Abstract summary: We propose a data-free post-training quantization (PTQ) method for efficient Diffusion Transformers (DiTs) DiTAS relies on the proposed temporal-aggregated smoothing techniques to mitigate the impact of the channel-wise outliers within the input activations. We show that our approach enables 4-bit weight, 8-bit activation (W4A8) quantization for DiTs while maintaining comparable performance as the full-precision model.
Score: 5.174900115018253
License:
Abstract: Diffusion Transformers (DiTs) have recently attracted significant interest from both industry and academia due to their enhanced capabilities in visual generation, surpassing the performance of traditional diffusion models that employ U-Net. However, the improved performance of DiTs comes at the expense of higher parameter counts and implementation costs, which significantly limits their deployment on resource-constrained devices like mobile phones. We propose DiTAS, a data-free post-training quantization (PTQ) method for efficient DiT inference. DiTAS relies on the proposed temporal-aggregated smoothing techniques to mitigate the impact of the channel-wise outliers within the input activations, leading to much lower quantization error under extremely low bitwidth. To further enhance the performance of the quantized DiT, we adopt the layer-wise grid search strategy to optimize the smoothing factor. Moreover, we integrate a training-free LoRA module for weight quantization, leveraging alternating optimization to minimize quantization errors without additional fine-tuning. Experimental results demonstrate that our approach enables 4-bit weight, 8-bit activation (W4A8) quantization for DiTs while maintaining comparable performance as the full-precision model.