Hello! I want to apply normalizing flow to some numerical data. Depending on the situation, the distribution I want to model can be a function of only one variable (for example a Lorentzian distribution) or of several (for example a multivariate gaussian). In the case of one number (say I have 10,000 numbers generated from a lorentzian distribution, with one variable), what is the best normalizing flow architecture I can use (I am aware that this might be an overkill, but I really need to use a normalizing flow)? The architectures I find on GitHub requires at least 2 numbers (for example Nice, or RealNVP), to be able to mix channels, but I have just one number as output, so I would need just one as input. Also, mathematically speaking, normalizing flows should work even for 1 number input/output. What should I use? Thank you!