# clementchadebec/benchmark_VAE

Unifying Variational Autoencoder (VAE) implementations in Pytorch (NeurIPS 2022)

Repository: https://github.com/clementchadebec/benchmark_VAE
Canonical: https://ross.abutalabs.com/products/benchmark_vae
Language: Python
License: Apache-2.0
License Family: permissive
Topics: vae, benchmarking, beta-vae, comparison, normalizing-flows, pixel-cnn, pytorch, reproducibility, reproducible-research, vae-gan, vae-implementation, vae-pytorch, variational-autoencoder, vq-vae, wasserstein-autoencoder
Last push: 2024-07-31T12:13:28+00:00

## Health v2 (maintenance only)
Score: 23/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 0, release rhythm 8, longevity 100
- inputs: {"age_days": 1796, "days_push": 763, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1995, forks 178 (observed 2026-08-28T04:06:03.460349+00:00)

## What it is
Pythae is a PyTorch library that unifies implementations of many Variational Autoencoder (VAE) variants under a common interface, enabling fair benchmarking with identical encoder/decoder architectures. It supports custom autoencoder architectures, distributed training, experiment tracking via wandb/mlflow/comet-ml, and model sharing through the HuggingFace Hub.

## Use cases
- benchmark VAE variants under the same neural network architecture
- train a beta-VAE or VQ-VAE on my own dataset with custom encoder and decoder
- reproduce published VAE research results in PyTorch
- compare generative autoencoder models like VAE-GAN and Wasserstein autoencoder
- share and load trained VAE models on the HuggingFace Hub
- run distributed training of a variational autoencoder on larger datasets
- generate new samples from a trained VAE with built-in samplers

## When to choose
- you need a unified, reproducible implementation of many VAE variants for fair comparison
- you want to train VAEs with your own encoder/decoder networks and track experiments with wandb, mlflow, or comet-ml
- you want to share or reuse trained VAE models via the HuggingFace Hub
- you need distributed (DDP) training for VAEs on large datasets

## When to avoid
- you need general-purpose deep learning tooling beyond autoencoder models
- you want diffusion models or transformer-based generative models rather than VAE-family methods
- you need a production inference server rather than a research and benchmarking library

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, deep-learning, benchmarking
- domain: machine-learning, deep-learning, artificial-intelligence
- platform: python
- tags: variational-autoencoder, pytorch, generative-models, reproducible-research, huggingface-hub, experiment-tracking, beta-vae, vq-vae, normalizing-flows, model-training

## Member repositories
- clementchadebec/benchmark_VAE (main) score 23

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:06:03.460349+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T03:02:30.793412+00:00, confidence not recorded.
  - readme: https://github.com/clementchadebec/benchmark_VAE (fetched 2026-08-28T04:06:03.460349+00:00, sha dcdae2715888)
- Data as of 2026-08-30T08:39:29.467469+00:00.