OpenELM: New local LLMs introduced by Apple
Apple researchers have presented another new language model family. OpenELM was placed at HuggingFace and has a few interesting features.
iPhone with packaging: Can a chatbot fit on it?
(Bild: BadPixma / Shutterstock)
Apple has been particularly active in AI research recently. Siri is supposed to understand more context, a new multimodal model is particularly efficient and the AI system Ferret UI simplifies app operation. Now the iPhone manufacturer has stepped up its game again: A new model family, OpenELM, has been made available on the HuggingFace platform. OpenELM stands for "Open-source Efficient Language Models" and the large language models (LLMs) are explicitly intended to run locally ("on-device"), i.e. possibly even on an iPhone. It is already clear to observers that Apple is also planning this as part of its AI initiative planned for iOS 18.
Smaller and more accurate
A total of eight OpenELM models exist, half of which have been tuned, and the rest trained using CoreNet, Apple's own library for training deep neural networks. According to Apple's OpenELM paper, a layer-wise scaling strategy was used, which was tuned for accuracy and efficiency.
The company provides not only the trained models, but also training logs and the source code on HuggingFace as part of its hub. According to the Apple research team, OpenELM requires two times fewer pre-training tokens, but still achieves 2.36 percent higher accuracy than OLMo with one billion parameters.
AI chips specifically for Apple
It is unclear whether OpenELM will also be part of iOS 18 - it could just be one of the company's many research projects. It is still unclear whether Apple will rely entirely on its own technology to incorporate generative functions into its systems or whether it will seek help from Google, for example.
However, Apple will be spending a lot of money to train its AI system itself. The purchase of corresponding servers has already been decided or implemented, but there will also be its own SoCs, which will be manufactured specifically for the application just for Apple. This is currently being reported by a leaker on the Chinese short message service Weibo. According to the information, Apple will use a 3 nm process from TSMC, but mass production is not planned until the second half of 2025. The leaker claims to have worked in the chip business for 25 years and has been correct in the past, including with the iPhone 14 and its SoC layout.
Empfohlener redaktioneller Inhalt
Mit Ihrer Zustimmmung wird hier ein externer Preisvergleich (heise Preisvergleich) geladen.
Ich bin damit einverstanden, dass mir externe Inhalte angezeigt werden. Damit können personenbezogene Daten an Drittplattformen (heise Preisvergleich) übermittelt werden. Mehr dazu in unserer Datenschutzerklärung.
(bsc)