Tutorials

ML Tutorials

End-to-end examples with real data, verified results, and full source code. Each tutorial is a self-contained .cpp file in Tutorial/Ml/.

Image Classification — Fashion-MNIST

Train a two-layer MLP on real IDX binary data. 83.2% test accuracy. Covers OaModule, OaLinear, OaAdamW, OaFnGrad::Backward, hardware timestamp timing, and cross-device portability.

Text Generation — Character Model

Predict the next character from an 8-token context. Covers OaEmbedding, OaLinear, OaFnMatrix::Tanh, greedy generation, and OaGradMode comparison.

Build & Run

All tutorials build from the oa repo root:

cmake --preset release
ninja -C Build/Release tutorial_mnist_classifier tutorial_text_generation_rnn

./Bin/Release/Tutorial/tutorial_mnist_classifier
./Bin/Release/Tutorial/tutorial_text_generation_rnn