toy-help.txt

Usage: toy_example.py [OPTIONS] COMMAND [ARGS]...

  2D toy example from the paper "Guiding a Diffusion Model with a Bad Version
  of Itself".

  Examples:

  # Visualize sampling distributions using autoguidance.
  python toy_example.py plot

  # Same, but save the plot as PNG instead of displaying it.
  python toy_example.py plot --save=out.png

  # Same, but specify the models explicitly.
  python toy_example.py plot \
      --net=https://nvlabs-fi-cdn.nvidia.com/edm2/toy-example/clsA-layers04-dim64/iter4096.pkl \
      --gnet=https://nvlabs-fi-cdn.nvidia.com/edm2/toy-example/clsA-layers04-dim32/iter0512.pkl \
      --guidance=3

  # Same, but using classifier-free guidance.
  python toy_example.py plot \
      --net=https://nvlabs-fi-cdn.nvidia.com/edm2/toy-example/clsA-layers04-dim64/iter4096.pkl \
      --gnet=https://nvlabs-fi-cdn.nvidia.com/edm2/toy-example/clsAB-layers04-dim32/iter0512.pkl \
      --guidance=4

  # Retrain the main model and visualize progress.
  python toy_example.py train

  # Retrain the main model and save snapshots.
  python toy_example.py train \
      --outdir=toy-example/clsA-layers04-dim64 \
      --cls=A --layers=4 --dim=64 --viz=false

Options:
  --help  Show this message and exit.

Commands:
  plot   Visualize sampling distributions with and without guidance.
  train  Train a 2D toy model with the given parameters.