MimiQ: Low-Bit Data-Free Quantization of Vision Transformers with Encouraging Inter-Head Attention Similarity
Kanghyun Choi, Hyeyoon Lee, Dain Kwon, SunJong Park, Kyuyeun Kim, Noseong Park, Jonghyun Choi, Jinho Lee
Abstract
Data-free quantization (DFQ) is a technique that creates a lightweight network from its full-precision counterpart without the original training data, often through a synthetic dataset. Although several DFQ methods have been proposed for vision transformer (ViT) architectures, they fail to achieve efficacy in low-bit settings. Examining the existing methods, we observe that their synthetic data produce misaligned attention maps, while those of the real samples are highly aligned. From this observation, we find that aligning attention maps of synthetic data helps improve the overall performance of quantized ViTs. Motivated by this finding, we devise MimiQ, a novel DFQ method designed for ViTs that enhances inter-head attention similarity. First, we generate synthetic data by aligning head-wise attention outputs from each spatial query patch. Then, we align the attention maps of the quantized network to those of the full-precision teacher by applying head-wise structural attention distillation. The experimental results show that the proposed method significantly outperforms baselines, setting a new state-of-the-art for ViT-DFQ.
BibTeX
@article{Choi_Lee_Kwon_Park_Kim_Park_Choi_Lee_2025, title={MimiQ: Low-Bit Data-Free Quantization of Vision Transformers with Encouraging Inter-Head Attention Similarity}, volume={39}, url={https://ojs.aaai.org/index.php/AAAI/article/view/33761}, DOI={10.1609/aaai.v39i15.33761}, abstractNote={Data-free quantization (DFQ) is a technique that creates a lightweight network from its full-precision counterpart without the original training data, often through a synthetic dataset. Although several DFQ methods have been proposed for vision transformer (ViT) architectures, they fail to achieve efficacy in low-bit settings. Examining the existing methods, we observe that their synthetic data produce misaligned attention maps, while those of the real samples are highly aligned. From this observation, we find that aligning attention maps of synthetic data helps improve the overall performance of quantized ViTs. Motivated by this finding, we devise MimiQ, a novel DFQ method designed for ViTs that enhances inter-head attention similarity. First, we generate synthetic data by aligning head-wise attention outputs from each spatial query patch. Then, we align the attention maps of the quantized network to those of the full-precision teacher by applying head-wise structural attention distillation. The experimental results show that the proposed method significantly outperforms baselines, setting a new state-of-the-art for ViT-DFQ.}, number={15}, journal={Proceedings of the AAAI Conference on Artificial Intelligence}, author={Choi, Kanghyun and Lee, Hyeyoon and Kwon, Dain and Park, SunJong and Kim, Kyuyeun and Park, Noseong and Choi, Jonghyun and Lee, Jinho}, year={2025}, month={Apr.}, pages={16037-16045} }