Why the RectifiedFlow performs better than the original flow matching paper?

Hi, thanks for the great work! I'm just wondering why the results on CIFAR-10 are much higher than the results in work "Flow matching for generative modeling"? Is there any important thing we need to note?