---

name: vae description: "Routes legacy VAE-family MNIST scripts, backend choices, model variant lookup, and TensorFlow/PyTorch compatibility troubleshooting." disable-model-invocation: true metadata: disco-role: operating license: Unlicense

VAE

Use this sub-skill for VAE-family requests in the Generative Models repository: vanilla VAE, conditional VAE, denoising VAE, adversarial autoencoder, and adversarial variational Bayes.

Route here when

• The user asks which VAE variant or source artifact label matches a modeling goal.
• The user needs TensorFlow-vs-PyTorch guidance for a legacy VAE example.
• The user asks about MNIST inputs, generated image outputs, latent-variable assumptions, or why a VAE script fails on a modern stack.
• The user names CVAE, DVAE, AAE, or AVB.

Route elsewhere when

• GAN variants: ../gan/SKILL.md
• Binary RBM CD/PCD: ../rbm/SKILL.md
• Helmholtz Machine wake-sleep: ../helmholtz-machine/SKILL.md
• Shared catalog, provenance, and compatibility: ../../SKILL.md

Fast decisions

• Default VAE baseline: vanilla VAE.
• Class-conditioned generation: conditional VAE.
• Denoising / corrupted input reconstruction: denoising VAE.
• Adversarial latent regularization: adversarial autoencoder.
• Adversarial variational objective with a T network: adversarial variational Bayes.

Shared conventions

• These are standalone legacy MNIST training loops, not a packaged library or CLI.
• The TensorFlow branches use old placeholder/session style.
• The PyTorch branches still rely on the legacy TensorFlow MNIST loader and old scalar logging patterns.
• Sample images are written to a working-directory-local out/ directory in source checkouts.

Read next

• references/workflows.md for the full VAE variant map and framework coverage.
• references/troubleshooting.md for VAE-specific modern-stack failures.
• ../../references/model-catalog.md for the repo-wide source artifact catalog.
• ../../references/compatibility.md for the shared TensorFlow/PyTorch/NumPy compatibility matrix.
• ../../scripts/check_legacy_stack.py before claiming an unchanged VAE example can execute.