Apple has unveiled a series of open-source AI large language models (LLMs), designed to run locally on devices, rather than being processed via cloud servers.
These advanced models, known as OpenELM (Open-source Efficient Language Models), are now available on the Hugging Face Hub, a community dedicated to sharing AI code.
According to a detailed white paper [PDF], there are a total of eight OpenELM models. Of these, four are pre-trained using the CoreNet library, while the remaining four are run using guided tuning. Apple is implementing a targeted, layer-based scaling strategy to increase accuracy and efficiency.
Read more: Apple reduces Vision Pro shipments due to low demand
Apple did not limit itself to providing the final trained model, but provided code, training logs, and various versions of it, in the hope that this will accelerate progress and foster "more reliable results" in the field of artificial intelligence applied to natural language.
Apple today announced the release of OpenELM models, aiming to empower and enrich the open research community with cutting-edge language models. Providing open-source models gives researchers the ability to examine the risks, data, and biases associated with the models. At the same time, developers and businesses have the ability to use these models as is or modify them to suit their needs.
Open information sharing is a critical tool for Apple in identifying and hiring the most talented engineers, scientists, and skilled professionals. This is because it encourages the development and publication of research work that, under Apple’s usually strict confidentiality guidelines, would not be possible.
See more: Apple: Releases the third beta version of iOS 17.5 and iPadOS 17.5
Although Apple has yet to integrate AI capabilities into its devices, it is expected that the upcoming iOS 18 will be enriched with a number of advanced AI features. Rumors have it that Apple plans to run its advanced language models directly on devices , thus strengthening user privacy protection
Source: macrumors
