Apple debuts its latest Open Source AI Models designed to operate directly on devices, bypassing the need for cloud services

Apple has entered the AI arena with the launch of OpenELM (Open-source Efficient Language Models), its new suite of open-source large language models (LLMs). Designed to operate directly on devices rather than relying on cloud services, OpenELM models can now be accessed via the Hugging Face Hub, a widely-used platform for AI code sharing.

The suite comprises eight language models, including four pre-trained with the CoreNet library and four instruction-tuned models. Apple has implemented a layer-wise scaling strategy across these models, prioritizing both accuracy and efficiency. This approach is detailed in the release white paper PDF accompanying the launch.

In a move to distinguish its LLM from competitors merely offering pre-trained models, Apple has opted to release the entire framework, encompassing code, training logs, and multiple versions. This decision underscores Apple’s commitment to advancing the research community by providing access to state-of-the-art language models through OpenELM.

By making OpenELM models open source, Apple seeks to empower and enrich the research community. According to the company, sharing these models enables researchers not only to utilize them but also to explore their inner workings, fostering faster progress and yielding “more trustworthy results” in the realm of natural language AI.

Researchers, developers, and companies have the flexibility to utilize Apple’s OpenELM models as they are or tailor them to specific requirements. This openness marks a departure from previous practices, where companies typically provided only model weights and inference code, withholding access to underlying training data or configurations.

The advantages of Apple’s on-device AI processing are two-fold: privacy and efficiency. By keeping data and processing local, OpenELM addresses growing concerns about user privacy and potential breaches from cloud servers. Moreover, on-device processing eliminates the need for internet connectivity, enabling AI functionalities even in offline scenarios. Apple underscores this advantage, noting that OpenELM achieves “enhanced accuracy” while demanding fewer resources compared to similar models.

While open-sourcing benefits researchers, it also holds strategic advantages for Apple. The open sharing of information facilitates collaboration within the research community, allowing others to contribute to and refine OpenELM. Additionally, this open environment attracts top talent, including engineers, scientists, and experts, to the company. According to Apple, OpenELM serves as a springboard for further AI advancements, benefiting not only Apple but the entire AI landscape.

Though Apple has yet to introduce these AI capabilities to its devices, the imminent release of iOS 18 and swirling rumors suggest Apple’s plan to incorporate on-device AI features with the new OS. With the launch of its LLM, it’s evident that Apple is laying the groundwork for upgrading its devices, including iPhones, iPads, and Macs, with AI capabilities. Incorporating large language models into its devices could enable more personalized and efficient user experiences. This shift towards on-device processing could also facilitate user privacy while providing developers with readily available, efficient AI tools.