The latest articles on DEV Community by Robert Reilly (@rdreilly58). https://dev.to/rdreilly58 https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F2843694%2Ff68f60a1-6974-48ce-86d2-1cdd119bb739.jpeg https://dev.to/rdreilly58 en Robert Reilly Thu, 19 Mar 2026 08:42:26 +0000 https://dev.to/rdreilly58/intrducing-momo-kiji-cuda-for-apple-neural-engine-10gn https://dev.to/rdreilly58/intrducing-momo-kiji-cuda-for-apple-neural-engine-10gn <h1> Introducing momo-kiji: CUDA for Apple Neural Engine </h1> <p><em>Cross-posted to Medium and Hashnode</em></p> <h2> The Problem </h2> <p>Every Apple device has an Apple Neural Engine. Every single one. Billions of them.</p> <p>Yet most ML developers ignore it completely.</p> <p>Why? Because there's no good way to target it. CoreML is locked down. You can't bring your own models. You're stuck in Apple's walled garden.</p> <p>Meanwhile, that ANE sits there doing nothing most of the time—a 10x efficiency boost, completely untapped.</p> <h2> Introducing momo-kiji </h2> <p>Today, we're releasing <strong>momo-kiji</strong>—an open-source CUDA-like SDK for Apple Neural Engine.</p> <p>It's simple: compile your model once, target ANE directly, and get 10x better efficiency without rewriting anything.</p> <div class="highlight js-code-highlight"> <pre class="highlight plaintext"><code> python import momo_kiji as mk # Load any model model = mk.load("model.onnx") # Compile for ANE compiled = model.compile(target="ane") # Deploy compiled.save("model_ane.mlmodel") </code></pre> </div> machinelearning ai opensource