Apple has released a number of open source AI models designed to run on hardware rather than through cloud servers.
The company calls its models “Open Source Efficient Language Models,” or OpenELM for short, and makes them available through the Hugging Face Hub platform, which enables code sharing for AI systems.
Apple offers eight different OpenELM models, and the company says it uses a “layered scaling strategy” to improve the accuracy and efficiency of these models with between 270 million and 3 billion parameters, making it a relatively small model to consider.
Apple does not provide the final training model, but rather all accessories for the new model, including code, training logs, and more, as the company's researchers want to accelerate the development of artificial intelligence systems.
Apple announced the launch of OpenELM templates to “enhance and enrich the open research community” with modern language models, where developers and companies can use the templates as is or modify them.
Recent reports indicate that Apple can use custom artificial intelligence models to run on its various upcoming devices without relying on servers to achieve certain benefits.
On the other hand, companies working in the field of artificial intelligence are working to develop small models that are faster, consume fewer resources than large linguistic models, and are characterized by the ability to work directly in the device.
There are many examples of these models, including the Phi-3 model from Microsoft, the Gemini Nano model from Google, and other similar models.