Minified generative models

Bare-bones, minified versions of some common (and not-so-common) generative models, for pedagogical purposes.

Installation

First, install JAX following these instructions. For CPU-only, this is as simple as:

pip install "jax[cpu]"

Additional libraries:

pip install flax optax diffrax tensorflow_probability scikit-learn tqdm matplotlib

β-VAEs: Variational autoencoders and basic rate-distortion theory.
Diffusion models: Diffusion models, covering likelihood-based and score-matching interpretations.
Normalizing flows (WiP annotations): Normalizing flows, specifically RealNVP.
Continuous normalizing flows: Continuous-time normalizing flows from e.g., Grathwohl et al 2018.
Consistency models (WiP annotations): Consistency models from Song et al 2023.
Flow matching (WiP annotations): From Lipman et al 2022; see also Albergo et al 2023.
Diffusion distillation (WiP): Progressive (Salimans et al 2022) and consistency (Song et al 2023) distillation.
Discrete walk-jump sampling (WiP): From Frey et al 2023.

Name		Name	Last commit message	Last commit date
Latest commit History 43 Commits
assets		assets
.gitignore		.gitignore
01_beta_vae.ipynb		01_beta_vae.ipynb
02_diffusion.ipynb		02_diffusion.ipynb
03_normalizing_flows.ipynb		03_normalizing_flows.ipynb
04_continuous_normalizing_flows.ipynb		04_continuous_normalizing_flows.ipynb
05_consistency_models.ipynb		05_consistency_models.ipynb
06_flow_matching.ipynb		06_flow_matching.ipynb
07_diffusion_distillation.ipynb		07_diffusion_distillation.ipynb
08_discrete_walk_jump_sampling.ipynb		08_discrete_walk_jump_sampling.ipynb
LICENSE.md		LICENSE.md
README.md		README.md
requirements.txt		requirements.txt