Apple has been very secretive about its generative AI plans. There is a general consensus that the company is working on bringing on-device AI abilities to the next iPhone, but we didn’t know what the features could be and how they’ll function, until now. The company’s researchers have released OpenELM, a language model that can run locally on iPhones.
What Is OpenELM?
OpenELM, by Apple, is a family of open-sourced efficient language models. Unlike some popular large language models, OpenELM consists of four small language models that perform efficiently on phones and laptops.
Traditional Transform Models Vs. OpenELM
The official Hugging Face library page mentions that it uses “a layer-wise scaling strategy to efficiently allocate parameters within each layer of the transformer model, leading to enhanced accuracy.”
For those catching up, layers are the fundamental building blocks of a language model. They are stacked on top of each other to process information step-by-step. Each layer can handle slightly more complex aspects of the language than the previous one.
Layer-Wise Parameter Allocation Improves Accuracy And Efficiency
As a user enters text-based input, it passes through these layers before the language model can make sense of it and return a suitable response. The layer-wise scaling strategy divides parameters (variables that a model understands) unevenly across the different layers.
This results in increased accuracy as the model allocates more crucial parameters on layers. Further, the strategy improved efficiency by assigning fewer parameters to less critical layers.
OpenELM Language Models Come In Four Parameter Sizes
As part of OpenELM, Apple researchers have released four very small language models, which are available in the Hugging Face library. The models are available in four parameter sizes: 270 million, 450 million, 1.1 billion, and 3 billion. In comparison, Google’s Gemma offers a 2 billion parameter version, while Microsoft’s Phi-3 model accepts 3.8 billion parameters.
Apple Can Implement The AI Models To Perform These Tasks
The new AI models can easily perform tasks like text generation, summarization, email writing, list creation, etc. Like Samsung, Apple can integrate these features across iOS. For instance, those who use Apple Notes can benefit from the text generation and summarization features. Further, the company can also improve its keyboard by allowing it to access the language models.
Benefits Of Having On-Device Processing
Since the small language models run locally on the device, they offer two benefits. Given that the data never leaves the device, OpenELM addresses several privacy concerns related to cloud-based servers. Further, on-device processing eliminates the delays caused by a poor internet connection.
The release of OpenELM comes after a Bloomberg report about Apple developing its own large language models to power on-device AI features. While the company could ship its upcoming iPhones with the OpenELM language models, it might make a few more changes before it finalizes the use on its smartphones.
You can follow Smartprix on Twitter, Facebook, Instagram, and Google News. Visit smartprix.com for the most recent tech and auto news, reviews, and guides.
