Normalizing Flows in PyTorch

This repository contains implementations of Normalizing Flows in PyTorch. The implementations are heavily inspired by karpathy/pytorch-normalizing-flows.

Getting Started

Run the following command to install the required dependencies:

uv sync
source .venv/bin/activate

NICE [Dinh+ 2014]

Train the NICE model on the specified dataset for 10,000 steps by default:

python train_nice.py --dataset moons

Moons Dataset

Gaussian Mixture Dataset

RealNVP [Dinh+ 2016]

Train the RealNVP model on the specified dataset for 10,000 steps by default:

python train_real_nvp.py --dataset moons

Moons Dataset

Gaussian Mixture Dataset

References

• karpathy/pytorch-normalizing-flows
• NICE: Non-linear Independent Components Estimation
• RealNVP: Improved Variational Inference with Inverted Autoregressive Flow