Cupertino, California – Technology giant Apple has recently unveiled a groundbreaking development in the field of artificial intelligence, releasing a series of advanced open source language models designed to operate directly on device rather than relying on cloud servers. These models, known as OpenELM (Open-source Efficient Language Models), can be accessed through the Hugging Face Hub, a platform dedicated to sharing AI code within the community.
According to a white paper detailing the project, Apple has introduced a total of eight OpenELM models, with four pre-trained using the CoreNet library and four finely-tuned instruction models. The company has implemented a layer-wise scaling strategy to enhance the accuracy and efficiency of these models, a move aimed at propelling advancements in the natural language AI sector.
In a departure from traditional practices that only offer model weights and inference code, Apple has made available the full code, training logs, and multiple versions of the models. The hope is that this comprehensive approach will accelerate progress and yield more reliable outcomes in natural language processing tasks.
By sharing these state-of-the-art language models with the research community, Apple aims to foster collaboration and innovation in the field. This open source initiative not only allows researchers to explore potential risks and biases in data and models but also provides developers and companies the opportunity to utilize the models as they are or customize them for specific applications.
Though Apple has not yet integrated these advanced AI capabilities into its products, speculations surround the upcoming iOS 18 release, which is anticipated to incorporate a range of new AI features. Rumors suggest that Apple is considering running its large language models on-device for privacy reasons, a move that could have significant implications for the future of AI technology.