normflows: A PyTorch Package for Normalizing Flows

Normalizing flows model probability distributions through an expressive tractable density. They transform a simple base distribution, such as Gaussian, sequence of invertible functions, which are referred to layers. These layers typically use neural networks become very expressive. Flows ubiquitous in machine learning and have been applied image generation, text modeling, variational inference, approximating Boltzmann distributions, many other problems. Here, we present normflows, Python package for normalizing flows. It allows build flow models from suite layers, networks. The is implemented the popular deep framework PyTorch, simplifies integration larger or pipelines. supports most common architectures, Real NVP, Glow, Masked Autoregressive Flows, Neural Spline Residual more. can be easily installed via pip code publicly available on GitHub.