Transformers Trained and Executed on Apple’s Neural Engine via Espresso

A new open‑source project called Espresso targets Apple hardware. It provides tools to train transformer‑based models on the device. The same framework also supports running inference on the Apple Neural Engine. By leveraging the Neural Engine, developers can keep model workloads on iOS devices. The approach reduces reliance on external GPUs or cloud services. Espresso’s code is hosted on GitHub for community contributions. The project aims to simplify on‑device AI development for Apple platforms. Early adopters can experiment with the repository to explore performance gains.