Apple has released OpenELM (Open-source Efficient Language Models), a set of large-scale AI language models that can operate on devices without needing to connect to cloud servers. More information can be found on Hugging Face.
OpenELM consists of a total of 8 models, with 4 models yet to be trained and another 4 models that have been fine-tuned. The parameter sizes range from 270M to 3B.
In the research details released, Apple stated that OpenELM allocates parameters in each model layer to increase accuracy. The result is a model that operates with 2.36% more accuracy compared to OLMo while using less than half the untrained tokens.
Having AI models that can run on small devices is another direction that large-scale technology companies are striving towards. Just recently, Microsoft unveiled Phi-3 which can run on smartphones.
TLDR: Apple introduces OpenELM, a set of AI language models that can operate without cloud servers. The models have parameter sizes ranging from 270M to 3B and boast increased accuracy compared to OLMo. This development aligns with the trend of creating AI models that can run on small devices.
Leave a Comment