Core ML improvements that contribute to achieving that goal are big news for the community! Naturally, many developers are exploring ways to run these models efficiently on-device and creating new apps and use cases. Many people want to run these models on their hardware for a variety of reasons, including convenience, privacy, and API cost savings. On-device execution is going through a period of extraordinary interest triggered by the popularity of models such as Stable Diffusion and Large Language Models with chat interfaces. These changes make running models such as Stable Diffusion faster and with less memory use! As a taste, consider the following test I ran on my iPhone 13 back in December, compared with the current speed using 6-bit palettization: Stable Diffusion on iPhone, back in December and now with 6-bit palettizationĬore ML is a mature framework that allows machine learning models to run efficiently on-device, taking advantage of all the compute hardware in Apple devices: the CPU, the GPU, and the Neural Engine specialized in ML tasks.

This year we are particularly excited about changes in Core ML devoted to compression and optimization techniques. Like every year, WWDC week is packed with more than 200 technical sessions that dive deep inside the upcoming features across Apple operating systems and frameworks. A lot of the news focused on the Vision Pro announcement during the keynote, but there’s much more to it. WWDC’23 (Apple Worldwide Developers Conference) was held last week.

Converting and Optimizing Custom Models.Using Quantized and Optimized Stable Diffusion Models.

Download the last version for ios Core