VQ-VAE-Style CIFAR Example

Trains a compact vector reconstruction model with a narrow tanh bottleneck, paired with the VQ-VAE spec/theory modules. The theorem-facing codebook objective lives in NN.Spec.Models.VqVae; this runtime example is the executable reconstruction path.

def NN.Examples.Models.Generative.VqVae.exeName : String

CLI subcommand name used in terminal banners and error messages.

def NN.Examples.Models.Generative.VqVae.defaultLogJson : System.FilePath

Default JSON loss-curve path for this command.

def NN.Examples.Models.Generative.VqVae.cfg : TorchLean.nn.models.VectorGenerativeConfig

Shared vector-image configuration.

The VQ-VAE runtime path uses the same compact flattened-CIFAR boundary as the autoencoder and VAE commands, so the model comparison changes the bottleneck while keeping data handling fixed.

@[reducible, inline]

abbrev NN.Examples.Models.Generative.VqVae.σ : Shape

Input shape: a batch of flattened CIFAR image vectors.

@[reducible, inline]

abbrev NN.Examples.Models.Generative.VqVae.τ : Shape

Target shape: reconstructed flattened CIFAR image vectors.

def NN.Examples.Models.Generative.VqVae.model : TorchLean.nn.M (TorchLean.nn.Sequential σ τ)

Trainable VQ-VAE-style vector model.

The codebook-facing objective is handled in the imported spec/theory modules; this command exercises the executable reconstruction path with a narrow quantization-style bottleneck.

def NN.Examples.Models.Generative.VqVae.data (flags : RealData.CifarModelTrainFlags) : TorchLean.Trainer.Dataset σ τ

Public singleton dataset for compact CIFAR reconstruction.

def NN.Examples.Models.Generative.VqVae.train (opts : TorchLean.Options) (flags : RealData.CifarModelTrainFlags) : IO (TorchLean.Trainer.TrainResult σ τ)

Train the compact VQ-VAE-style model with the public Trainer surface.

def NN.Examples.Models.Generative.VqVae.main (args : List String) : IO UInt32

Executable entrypoint for the compact VQ-VAE-style run.

The command loads a real CIFAR minibatch, trains the reconstruction objective, and records the same summary/log artifact format as the other public trainer commands.