A Microsoft engineer has successfully ported the BitNet large language model to iOS, achieving 45 tokens per second on an iPhone 14 Pro Max using the 0.7B parameter model, which requires approximately 200MB of memory.
The achievement, detailed in a social media post, demonstrates the potential for running substantial AI models directly on mobile devices. BitNet’s functionality, as described by the engineer, centers on efficient on-device processing.
The iPhone 14 Pro Max, released in 2022, features Apple’s A16 Bionic chip, a 6-core CPU, and a 5-core GPU. According to Apple’s technical specifications, the device supports 5G connectivity, Gigabit LTE, VoLTE, and Wi-Fi calling where available. The phone also includes a professional-grade camera system with a 48MP main camera and advanced image stabilization.
Apple’s iPhone 14 and iPhone 14 Pro models are equipped with crash detection capabilities, automatically contacting emergency services when a severe car crash is detected, provided a mobile connection or configured Wi-Fi calling is available. This feature relies on a data plan.
The iPhone 14 Pro Max boasts a 6.7-inch Super Retina XDR display with a resolution of 2796 x 1290 pixels at 460 pixels per inch. The display incorporates features like Dynamic Island, ProMotion technology with adaptive refresh rates up to 120Hz, and HDR support. It is protected by Ceramic Shield on the front and features a textured matte glass back with a surgical-grade stainless steel frame.
Available storage options for the iPhone 14 Pro Max include 128GB, 256GB, 512GB, and 1TB. The device is rated IP68 for dust and water resistance, capable of withstanding submersion in up to 6 meters of water for 30 minutes according to IEC 60529 standards.
