Flow Matching (FM) generative models offer efficient simulation-free training and deterministic sampling, but their practical deployment is challenged by high-precision parameter requirements. We adapt optimal transport (OT)-based post-training quantization to FM models, minimizing the 2-Wasserstein distance between quantized and original weights, and systematically compare its effectiveness against uniform, piecewise, and logarithmic quantization schemes. Our theoretical analysis provides upper bounds on generative degradation under quantization, and empirical results across five benchmark datasets of varying complexity show that OT-based quantization preserves both visual generation quality and latent space stability down to 2-3 bits per parameter, where alternative methods fail. This establishes OT-based quantization as a principled, effective approach to compress FM generative models for edge and embedded AI applications.

翻译：流匹配（Flow Matching，FM）生成模型提供了高效的无模拟训练和确定性采样，但其实际部署受到高精度参数需求的挑战。我们将基于最优传输（Optimal Transport，OT）的训练后量化方法应用于FM模型，通过最小化量化权重与原始权重之间的2-Wasserstein距离，并系统性地比较其与均匀量化、分段量化及对数量化方案的有效性。我们的理论分析提供了量化下生成性能退化的上界，并在五个不同复杂度的基准数据集上的实验结果表明，基于OT的量化在每参数2-3比特的低比特率下仍能保持视觉生成质量和潜在空间稳定性，而其他方法在此条件下失效。这确立了基于OT的量化作为一种原理性、有效的方法，可用于压缩FM生成模型，适用于边缘和嵌入式人工智能应用。