Apple is taking steps to integrate its language learning model, known as GPT (Generative Pre-trained Transformer), into iPhone and iPad devices. The introduction of this move aims to strengthen Siri’s capabilities and provide users with more accessible shortcut integration. While the tech giant invests heavily in training its language model, one of the key aspects that sets Apple apart is its commitment to on-device processing instead of relying solely on cloud processing.
To achieve on-device functionality for GPT, Apple researchers have developed a method that efficiently uses large language models with limited memory. Their approach involves constructing an inference cost model that aligns with flash memory behavior, optimizing the volume of data transferred from flash memory, and enabling reading of larger, coherent chunks. This is made possible through two innovative technologies implemented by Apple:
1. Windowing: This technique selectively loads parameters for a few recent tokens, utilizing activations from previously computed tokens. By employing a sliding window approach, the number of input/output (IO) requests needed to load weights is significantly reduced.
2. Row-column bundling: Apple stores combined row and column projections to read larger, coherent chunks from flash memory. This increases throughput, enabling reading of larger blocks of data.
Implementing these methods can result in significant performance improvements. CPUs can experience 4-5 times speed improvement, while graphics cards (GPUs) can be up to 20-25 times faster. As a result, Apple’s AI models may potentially have twice the size of the iPhone’s memory. In practice, this means that Siri will become more capable of real-time translation, photo and video analysis, and gaining insights into user behavior and preferences.
By prioritizing on-device processing, Apple emphasizes its commitment to user privacy, aligning with the vision of privacy as a “fundamental human right.” While the exact timeline for integrating GPT into Apple devices remains uncertain, it is expected that the company will introduce this technology for iPhone and iPad users during the iOS 18 cycle, possibly by the end of 2024. As Apple continues to push the boundaries of AI technology, users eagerly anticipate the introduction of this groundbreaking enhancement.
Frequently Asked Questions (FAQ)
Q: When will Apple introduce GPT integration in iPhone and iPad devices?
A: The exact timeline for integration remains uncertain, but it is expected that Apple will introduce this technology for iPhone and iPad users by the end of 2024, during the iOS 18 cycle.
Q: What benefits will the implementation of windowing and row-column bundling techniques bring?
A: Implementing these techniques can significantly increase performance. CPUs can experience 4-5 times speed improvement, and GPUs can be up to 20-25 times faster. This means that Apple’s AI models can have twice the size of the iPhone’s memory.
Definitions
– GPT (Generative Pre-trained Transformer): A language model based on the Transformer architecture, trained through self-supervision on large datasets. GPT was developed by OpenAI and has broad applications in natural language processing.
– Siri: A virtual voice assistant developed by Apple. Siri responds to user queries, executes commands, and provides various services, integrating with various applications on Apple devices.
Source: www.apple.com
The source of the article is from the blog be3.sk