联合分支空间变换编码用于带分类器引导的扩散激活量化
Joint Branch-Space Transform Coding for Diffusion Activation Quantization with Classifier-Free Guidance
另 2 家 · 查看机构详情
- Duke University（杜克大学）
- Carnegie Mellon University（卡内基梅隆大学）
- University of Florida（佛罗里达大学）
- Wake Forest University（维克森林大学）
- University of Oxford（牛津大学）
机构由 AI 辅助整理，请以论文原文为准。
浏览论文内容
中文总结 AI 辅助
针对扩散模型激活量化忽略跨分支相关性的问题，提出分支空间变换编码及GCBT方法，利用2x2正交矩阵旋转CFG分支，在等速率代理下获得闭式解，显著提升量化保真度且不改变宿主流程。
中文摘要 AI 辅助
扩散模型的后训练量化日益利用时间步、特征和层结构。虽然近期工作已开始将CFG结构纳入扩散量化，但激活量化仍在条件坐标和无条件坐标上独立运行，未利用跨激活结构。我们表明，匹配的CFG激活形成一个强相关的二维源，且在固定比特预算下，分支编码基的选择会实质性地影响量化保真度。受此观察启发，我们引入分支空间变换编码，通过离线推导的2x2正交矩阵旋转匹配的CFG分支，仅需对模型参数或量化流程进行最小修改。我们进一步推导出引导相关分支变换（GCBT），其联合纳入CFG引导方向和跨分支二阶矩。在等速率量化噪声代理下，GCBT无需梯度优化或角度搜索即可获得逐层的闭式解。应用于现有扩散PTQ方法之上，GCBT在大多数评估比较中产生统计上显著的保真度提升，且无统计上显著的退化，同时保持底层宿主量化流程不变。
英文摘要
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.
