27B Large Model Fits into iPhone! Apple Focuses on AI Compression Tech: Volume Reduced to 1/14, Speed Increased 8 Times

Tech media The Information reported that Apple is in talks with AI startup PrismML to evaluate the feasibility of running larger AI models directly on iPhones. PrismML's core breakthrough is its native 1-bit model compression technology, which can compress model size to about 1/14 and reduce memory usage by over 90%. This move could enable large-scale AI models to run on mobile devices, achieving a breakthrough in edge AI.