Apple has launched the language model OpenELM, with an emphasis on open-source and efficiency. The model is expected to be available on the iPhone.
Compared to other tech companies, Apple's progress in AI has been rather unclear. The full name of OpenELM is "Open-source Efficient Language Models." Apple has highlighted its open-source nature and emphasis on efficiency.
Apple has launched four versions of the model with different parameter scales: 270M, 450M, 1.1B, and 3B. In comparison, Phi-3, the "small language model" recently released by Microsoft, has a parameter scale of 3.8B, while Google's lightweight model Gemma has versions with 2B and 7B parameters.
It is anticipated that Apple's generative AI strategy is to integrate AI into products such as the iPhone, iPad, and Mac. It also aims to enable AI computation on the device side.
In addition to self-developed models, Bloomberg reported that Apple is also in talks with companies such as Google and OpenAI to implement models such as Gemini and GPT on the iPhone. Chinese media outlets reported that Apple has approached Baidu to use Ernie Bot in China.
Over the past year, Apple has acquired two AI startups, WaveOne and Datakalab, both known for their expertise in image compression and visual computing technology. This year's WWDC is set to take place from June 10 to 14, and Apple's senior VP in Marketing Greg Joswiak has hinted on X (formerly Twitter) that AI will be the main theme of this year's event.
