Skip to content

Guides¤

Each guide takes one part of flowMC and shows it in use, with examples of its options.

  • Training Normalising Flows — The two normalising flow architectures in flowMC, RealNVP and Masked Coupling RQSpline, trained on a 2D toy dataset: their expressiveness and cost, and how a trained flow plugs into the resource-strategy system via the TrainModel strategy.
  • Training Flow Matching Models — Flow matching as an alternative to normalising flows: training a Conditional Optimal Transport model on a 2D distribution, sampling by integrating the learned ODE, and evaluating the density on a grid.