CNN Training Example

Runnable torchlean cnn example. It trains a small convolutional classifier on a prepared CIFAR-10 minibatch.

The reusable model wiring lives behind the public TorchLean.nn.models.CNN constructor. The command adds the pieces around it: CLI parsing, dataset selection, step-limited loader training, and TrainLog artifact writing.

python3 scripts/datasets/download_example_data.py --cifar10
lake build -R -K cuda=true && lake exe torchlean cnn --cuda --n-total 1 --steps 1

def NN.Examples.Models.Vision.Cnn.exeName : String

CLI subcommand name used in terminal banners and parser errors.

def NN.Examples.Models.Vision.Cnn.defaultLogJson : System.FilePath

Default JSON loss-curve path for this command.

def NN.Examples.Models.Vision.Cnn.batch : ℕ

Static minibatch size for the compact CIFAR run.

The model owns the batch axis, so this value appears in both the input/output shapes and the classifier trainer type.

def NN.Examples.Models.Vision.Cnn.inC : ℕ

CIFAR image channels.

def NN.Examples.Models.Vision.Cnn.inH : ℕ

Height of the CIFAR crop used by this runnable CNN command.

def NN.Examples.Models.Vision.Cnn.inW : ℕ

Width of the CIFAR crop used by this runnable CNN command.

def NN.Examples.Models.Vision.Cnn.outDim : ℕ

CIFAR class count, hence the output-logit width.

def NN.Examples.Models.Vision.Cnn.cfg : TorchLean.nn.models.CnnConfig

Shared CNN configuration used by shapes and the reusable public model constructor.

@[reducible, inline]

abbrev NN.Examples.Models.Vision.Cnn.σ : Spec.Shape

Input shape: a minibatch of CIFAR images in channel-first layout.

@[reducible, inline]

abbrev NN.Examples.Models.Vision.Cnn.τ : Spec.Shape

Output shape: one row of class logits per image.

def NN.Examples.Models.Vision.Cnn.model : TorchLean.nn.M (TorchLean.nn.Sequential σ τ)

Small convolutional classifier from the public model API.

The command chooses the CIFAR paths and runtime flags; the model itself stays an ordinary nn.Sequential value built from the public API.

def NN.Examples.Models.Vision.Cnn.train (opts : TorchLean.Options) (flags : RealData.CifarModelTrainFlags) : IO TorchLean.Trainer.TrainSummary

Train the CIFAR CNN with the public Trainer surface.

def NN.Examples.Models.Vision.Cnn.main (args : List String) : IO UInt32

CLI entrypoint for CIFAR CNN training; CUDA is the maintained validation path.