The new Apple AI will work without internet and will be more powerful than many of its competitors.

Apple has announced OpenELM, a set of small open-source artificial intelligence language models designed to run directly on devices like iPhones and Macs. These models are small enough to run locally on the device, eliminating the dependency on cloud servers for AI tasks.

iOS 17 DOWNLOAD

OpenELM (“Open-source Efficient Language Models”) uses a layer-wise scaling strategy to efficiently allocate parameters within each layer of the model, resulting in higher accuracy. The model family consists of eight variants with parameter sizes ranging from 270 million to 3 billion, all trained on public datasets.

Compared to other similar open source models, such as OLMo, OpenELM has demonstrated superior performance, despite needing twice the amount of training data. These models have been trained using CoreNet, an open source library, along with other models that enable efficient inference and fine-tuning on Apple devices.

Apple has taken a transparent approach by not only publishing the weights of the model and the inference code, but also the complete framework for training and evaluating the language model on public datasets, as well as tools for inference and fine-tuning on Apple devices.

The launch of OpenELM comes before the Apple Worldwide Developers Conference (WWDC), suggesting an interest from Apple in integrating artificial intelligence into their next version of the mobile operating system, iOS 18.

This movement also reflects a general trend in the technology industry, where companies like Microsoft and Apple are developing smaller and more efficient language models. The fact that Apple has decided to make OpenELM open source marks a significant change in its approach, as it used to keep its development methods more restrictive and reserved.

iOS 17 DOWNLOAD