用单次推理实现高质量生成，效率提升50%
P-Guide: Parameter-Efficient Prior Steering for Single-Pass CFG Inference
- 仅调整初始潜在状态实现条件引导，避免双次前向计算
- 实验显示延迟降低约50%，生成质量与标准方法相当
- 支持异方差先验，增强对数据不确定性的鲁棒性
Classifier-Free Guidance (CFG) 在流匹配中对高质量条件生成至关重要，但每次采样步骤需两次前向传播，带来显著计算开销。本文提出 P-Guide 框架，通过仅调节初始潜在状态，在单次推理中实现高质量引导。我们证明，在一阶近似下，P-Guide 等价于 CFG，可通过引导先验空间实现生成，无需采样时显式外推速度场。研究考虑同方差和异方差先验，发现联合建模均值与方差可实现自适应损失衰减，提升对数据不确定性的鲁棒性。大量实验表明，P-Guide 将推理延迟降低约50%，同时保持与标准双次前向传递的 CFG 基线相当的保真度和提示对齐能力。
原文摘要 · Abstract (English)
Classifier-Free Guidance (CFG) is essential for high-fidelity conditional generation in flow matching, yet it imposes significant computational overhead by requiring dual forward passes at each sampling step. In this work, we address this bottleneck by introducing \textbf{P-Guide}, a framework that achieves high-quality guidance through a single inference pass by modulating only the initial latent state. We further show that, under a first-order approximation, P-Guide is equivalent to CFG in the sense that it steers generation from the prior space, without requiring explicit velocity field extrapolation during sampling. We consider both homoscedastic and \textbf{heteroscedastic} priors, and find that jointly modeling the mean and variance enables adaptive loss attenuation and improved robustness to data uncertainty. Extensive experiments demonstrate that P-Guide reduces inference latency by approximately 50\% while maintaining fidelity and prompt alignment competitive with standard dual-pass CFG baselines.
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。