解决量化噪声在扩散模型采样中累积问题，提升生成质量。
Q-Drift: Quantization-Aware Drift Correction for Diffusion Model Sampling
- 将量化误差视为每步的随机扰动，设计分布保持的漂移修正方法。
- 仅需5次校准运行即可估计时间步方差，最多降低4.59点FID。
- 无需修改模型，兼容主流采样器与量化方法，推理开销极小。
后训练量化（PTQ）是部署大型扩散模型的实用路径，但量化噪声会随去噪轨迹累积，降低生成质量。本文提出Q-Drift，一种基于采样器侧的原理性修正方法，将量化误差视为每步去噪中的隐式随机扰动，并推导出保持边际分布的漂移调整。Q-Drift通过校准估计时间步相关的方差统计量，实际只需5次成对的全精度/量化校准运行。该采样器修正可即插即用，兼容常见采样器、扩散模型及PTQ方法，且推理开销可忽略。在六种不同文本到图像模型（涵盖DiT与U-Net）、三种采样器（Euler、flow-matching、DPM-Solver++）及两种PTQ方法（SVDQuant、MixDQ）下，Q-Drift在多数设置中均优于对应量化基线，于PixArt-Sigma（SVDQuant W3A4）上实现最高达4.59的FID降低，同时保持CLIP分数不变。
原文摘要 · Abstract (English)
Post-training quantization (PTQ) is a practical path to deploy large diffusion models, but quantization noise can accumulate over the denoising trajectory and degrade generation quality. We propose Q-Drift, a principled sampler-side correction that treats quantization error as an implicit stochastic perturbation on each denoising step and derives a marginal-distribution-preserving drift adjustment. Q-Drift estimates a timestep-wise variance statistic from calibration, in practice requiring as few as 5 paired full-precision/quantized calibration runs. The resulting sampler correction is plug-and-play with common samplers, diffusion models, and PTQ methods, while incurring negligible overhead at inference. Across six diverse text-to-image models (spanning DiT and U-Net), three samplers (Euler, flow-matching, DPM-Solver++), and two PTQ methods (SVDQuant, MixDQ), Q-Drift improves FID over the corresponding quantized baseline in most settings, with up to 4.59 FID reduction on PixArt-Sigma (SVDQuant W3A4), while preserving CLIP scores.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。