ViT-Style Real-Data Example

Runnable torchlean vit example. It trains a compact ViT-style image classifier on a prepared CIFAR-10 minibatch: patch embedding by convolution, token reshape, transformer block, and linear head.

The reusable model wiring lives in NN.API.Models.Vit (nn.models.vit1). This file is the runnable wrapper (CIFAR loader construction + multi-epoch training loop).

python3 scripts/datasets/download_example_data.py --cifar10
lake build -R -K cuda=true && lake exe torchlean vit --cuda --n-total 200 --steps 1

Tip: the defaults are set for a quick sanity run. For a longer run, bump --steps and --n-total, and enable CUDA fused kernels:

lake build -R -K cuda=true
lake exe torchlean vit --cuda --fast-kernels --n-total 2000 --steps 50

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

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

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

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

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

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

def NN.Examples.Models.Vision.Vit.patchH : ℕ

def NN.Examples.Models.Vision.Vit.patchW : ℕ

def NN.Examples.Models.Vision.Vit.stride : ℕ

def NN.Examples.Models.Vision.Vit.padding : ℕ

def NN.Examples.Models.Vision.Vit.dModel : ℕ

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

def NN.Examples.Models.Vision.Vit.numHeads : ℕ

def NN.Examples.Models.Vision.Vit.headDim : ℕ

def NN.Examples.Models.Vision.Vit.ffnHidden : ℕ

def NN.Examples.Models.Vision.Vit.cfg : API.nn.models.VitConfig

@[reducible, inline]

abbrev NN.Examples.Models.Vision.Vit.σ : Shape

@[reducible, inline]

abbrev NN.Examples.Models.Vision.Vit.τ : Shape

def NN.Examples.Models.Vision.Vit.mkModel : API.nn.M (API.nn.Sequential σ τ)

def NN.Examples.Models.Vision.Vit.loadCifarLoader {α : Type} [API.Semantics.Scalar α] [API.Runtime.Scalar α] (xPath yPath : System.FilePath) (nRows seed : ℕ) : IO (API.Data.BatchLoader α batch RealData.CifarImage RealData.CifarTarget)

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