Contrary to perceptions of Apple being slow in AI development, the company has been actively working on it for years. Known for its secretive nature, Apple recently unveiled OpenELM, a generative AI model.
OpenELM is an open-source language model family optimized for Apple devices like iPhones and Macs. It employs a layer-wise scaling strategy, enhancing accuracy by efficiently allocating parameters. It includes eight models with parameter sizes ranging from 270M to 3B, all trained on public datasets.
OpenELM is designed for on-device use, reducing reliance on cloud servers. Despite using half the training data of similar models, OpenELM reportedly outperforms them. It's trained on CoreNet, an open-source library, along with other models for efficient inference on Apple devices.
"Diverging from prior practices that only provide model weights and inference code... our release includes the complete framework for training and evaluation... This comprehensive release aims to empower and strengthen the open research community," shared Apple in a research paper.
The release of OpenELM precedes WWDC in June, where Apple is anticipated to introduce iOS 18 with new AI features. OpenELM offers insights into Apple's AI advancements ahead of the event.
Apple's move to make OpenELM open-source marks a departure from its traditionally secretive practices. This change aligns with the trend of tech giants embracing small model AI solutions.
