PyTorch to Core AI in Xcode (Fashion-MNIST)

On a Mac, train a small clothing model on Fashion-MNIST. Use MPS if you have it, or CPU if you do not. Convert it to .aimodel, then show the top label in a SwiftUI app with Core AI.

This course is for you if you can write Python and Swift, and you are new to Core AI.

Lessons

Key concepts

Train ClothingCNN on Fashion-MNIST on a Mac with MPS or CPU. Never CUDA.
Save clothing.pt, export, then convert with coreai-torch to clothing.aimodel.
Fixed contract: input image is NCHW [1, 1, 28, 28] float32. Output logits is [1, 10] float32.
Xcode 27 loads function main. Prepare once. Softmax is for display only.

Takeaways

You can train, convert, and show a clothing label in a SwiftUI Core AI app.
You can profile load, specialisation, and inference on a real device.
You can check held-out accuracy, Mac versus Core AI labels, and a yes or no top-label rule.

Agent briefs

FAQ

Can I train on a Mac then run in an iOS app?

Yes. Train in PyTorch on MPS/CPU, convert with coreai-torch to a .aimodel, then load in Xcode on iOS/macOS 27+.

What is coreai-torch?

Apple's converter that turns a trained PyTorch model into a Core AI .aimodel for on-device inference.

Is Fashion-MNIST production-ready?

No — it is the teaching classifier used to prove the train→convert→ship loop before you swap in your own data.

Do I need MLX for this path?

No. This course uses PyTorch + coreai-torch. MLX is a separate train/experiment stack (see Core AI vs Core ML vs MLX).

Where does LiteRT fit?

LiteRT is the Android sibling path — see PyTorch to Android with LiteRT.