SDK

Load the OA narration, convert it to mono, normalize it, render a native GPU reverberation tail, then save a checked WAV output.

Create two GPU-resident matrices, add them, and verify the result after the language-appropriate completion boundary.

Record, submit, validate, and measure GPU matrix multiplication across representative shapes.

Run three SHAKE-256 digests on the GPU and compare every output byte with the independent CPU implementation.

Learn a 300-token BPE vocabulary, train a causal Transformer for 300 optimizer steps, verify that cross-entropy falls, and generate text from the trained model.

Train, evaluate, checkpoint, reload, and verify a complete image-classification pipeline.

Record one training-loss series, render the retained figure through OA's GPU compositor, save it, and optionally inspect it in Viewer.

Build and render a retained diagnostics gallery with matching C++ and Python programs.

Load the Vision cover, convert its GPU-resident RGB pixels to Rec.709 grayscale, save the semantic Gray image, and optionally inspect it in Viewer.

Open the authored ML cover in OA’s native Vulkan viewer, inspect it on the navigable canvas, and move into fullspace presentation.

Compose deterministic image augmentation views and render the result without intermediate readback.