Today, Apple released several open-source large language model LLMs that are specifically designed to run on devices directly rather than using the cloud server.
Apple introduced the small source available AI language model called Open ELM which is tiny enough that it could be easily run on small devices like smartphones.
Apple has made 8 different AI language models that are open source and anyone can use them. Under the Sample Code License, Apple’s New AI model is currently available on Hugging Face Hub. Although there are some restrictions in the license that don’t make this completely open source, somehow the source code for Open ELM is available.
Different Open EML models of Apple are so much smaller, ranging from 270 million to 3 billion parameters in eight different models.
While recently the Meta Llama 3 family contains 70 billion and Chat GPT’s Open AI contains 175 billion parameters. A model with more parameters can handle more detailed and nuanced tasks, but it also requires more computing power to run.
LLMs is the Large Language Model that has a big set of data. The bigger the LLM, the more data it holds. Generally, the language model data is stored in a cloud server, and users can get online access easily, but they are making a small LMS that can easily fit into small devices like phones and tablets.
According to Apple, the release of the Open ELM models “to empower and enrich the open research community”. The sharing of open-source models gives researchers a path to investigate the data and risks associated with them.
This way, Apple will get the top engineers, scientists, and experts, and scientists. This is because before due to Apple’s secretive policies they could not publish research papers, so they could not get the opportunity to do this.
It is expected that these AI capabilities will be added as new AI features in iOS 18. Rumors say that Apple is considering its large language model on devices for privacy reasons.
Apple explains this in its paper, the eight latest AI models. Four will be trained on the big library of data called Core Net, while the other four will be trained to respond to certain commands and queries.
In the era of AI, where everyday tech giants like Google, Microsoft, and Meta are competing with each other in AI and releasing the latest updates, Apple is also seriously considering AI. It was revealed that Apple wants to make chips for its AI servers.
Generally, Apple does not share their software with anyone, but now they state that they want to help other people in research. As they are sharing these Open ELMS with everyone in the AI community.
This way, the iPhone will do most of the AI work on its device directly instead of using the cloud server or internet. And their data will not go on cloud servers. It is the era of AI where people are more worried about privacy with AI.
It is believed that Apple will share these new models with developers, so they can find the problem in it and fix it to make a better version of the software. This will be an essential step if Apple wants to use these AI modes in future versions of iOS and macOS.