APPLE DEVELOPS BREAKTHROUGH METHOD FOR RUNNING LLMS ON IPHONES
Apple researchers have made a key breakthrough in deploying large language models on Apple devices with limited memory by using an innovative flash memory utilization technique. The method uses windowing and row-column bundling to minimize data transfer and maximize flash memory throughput. It allows AI models to run up to five times faster on standard processors and up to 25 times faster on graphics processors. The technology sets the stage for iPhones to run complex AI assistants on-device.