Joint Branch-Space Transform Coding for Diffusion Activation Quantization with Classifier-Free Guidance
Abstract
Post-training quantization for diffusion models increasingly exploits timestep, feature, and layer structure. While recent work has begun incorporating CFG structure into diffusion quantization, activation quantization still operates independently across conditional and unconditional coordinates, leaving cross-activation structure unexploited. We show that matched CFG activations form a strongly correlated two-dimensional source and that, under a fixed bit budget, the choice of branch coding basis materially affects quantization fidelity. Motivated by this observation, we introduce branch-space transform coding, which rotates matched CFG branches via an offline derived 2x2 orthogonal matrix, requiring minimal modifications to model parameters or the quantization pipeline. We further derive the Guidance-Correlation Branch Transform (GCBT), which jointly incorporates the CFG guidance direction and cross-branch second moments. Under an equal-rate quantization-noise surrogate, GCBT admits a closed-form per-layer solution without gradient optimization or angle search. Applied on top of existing diffusion PTQ methods, GCBT yields statistically significant fidelity gains in most evaluated comparisons with no statistically significant degradation, while leaving the underlying host quantization pipeline unchanged.
Related
- PathGuide: Dynamic Classifier-Free Guidance via On-Policy Transport Alignment
- Steepest Guidance: A Practical and Principled Approach to Inference-Time Alignment of Flow and Diffusion-based Models
- Internal Guidance as a Sampling Method for Tilted Diffusion Models
- VGAS: Variance-Reduced Guidance and Adaptive Selection for Training-Free Reward Alignment in Discrete Diffusion
- Towards Digital Halftoning on Closed Manifolds--An Error Diffusion Scheme for the 2D Torus based on Sigma-Delta Quantization along the Rank-one Lattice
BibTeX
@misc{jiang2026joint,
title = {{Joint Branch-Space Transform Coding for Diffusion Activation Quantization with Classifier-Free Guidance}},
author = {Mingrun Jiang and Yuejia Liu and Zishan Shao and Ting Jiang and Qinsi Wang and Hancheng Ye and Yixiao Wang and Rui-Feng Wang and Kangning Cui and Yixuan Chen and Fan Yang and Xiang Cheng and Hai Li and Yiran Chen},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.00930},
eprint = {2610.00930},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.00930},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Joint Branch-Space Transform Coding for Diffusion Activation Quantization with Classifier-Free Guidance
AU - Mingrun Jiang
AU - Yuejia Liu
AU - Zishan Shao
AU - Ting Jiang
AU - Qinsi Wang
AU - Hancheng Ye
AU - Yixiao Wang
AU - Rui-Feng Wang
AU - Kangning Cui
AU - Yixuan Chen
AU - Fan Yang
AU - Xiang Cheng
AU - Hai Li
AU - Yiran Chen
PY - 2026
DA - 2026/10/01
PB - arXiv
DO - 10.48550/arxiv.2610.00930
UR - https://arxiv.org/abs/2610.00930
AB - Post-training quantization for diffusion models increasingly exploits timestep, feature, and layer structure. While recent work has begun incorporating CFG structure into diffusion quantization, activation quantization still operates independently across conditional and unconditional coordinates, leaving cross-activation structure unexploited. We show that matched CFG activations form a strongly correlated two-dimensional source and that, under a fixed bit budget, the choice of branch coding basis materially affects quantization fidelity. Motivated by this observation, we introduce branch-space transform coding, which rotates matched CFG branches via an offline derived 2x2 orthogonal matrix, requiring minimal modifications to model parameters or the quantization pipeline. We further derive the Guidance-Correlation Branch Transform (GCBT), which jointly incorporates the CFG guidance direction and cross-branch second moments. Under an equal-rate quantization-noise surrogate, GCBT admits a closed-form per-layer solution without gradient optimization or angle search. Applied on top of existing diffusion PTQ methods, GCBT yields statistically significant fidelity gains in most evaluated comparisons with no statistically significant degradation, while leaving the underlying host quantization pipeline unchanged.
ER -