In a quiet yet significant move, Apple has released its first multimodal Large Language Model (LLM) AI as an open source project, named Ferret.
Introduced in October by Apple AI researcher Zhe Gan via X/Twitter, Ferret has largely flown under the radar until now.
The Birth of Ferret
Ferret is the result of a joint development effort between Gan and his colleagues at Apple, along with researchers at Columbia University.
Also Read: Huawei’s satellite internet service hits 660 Mbps speeds
According to Gan, Ferret surpasses OpenAI’s GPT-4 in precision when it comes to understanding small image regions and describing them, while producing fewer hallucinations (errors).
Behind the Scenes: Training Ferret
Apple’s Github repository reveals some intriguing details about the training of Ferret. The company utilized 8 high-end Nvidia A100 GPUs equipped with 80GB of HBM2e RAM.
The A100 is currently the most in-demand GPU on the market, following the explosion of generative AI technology that came in the wake of the launch of OpenAI’s ChatGPT late last year.
Capable of 312 TeraFLOPS at Tensor Float 32 precision, the 80GB model used by Apple to train Ferret delivers a bandwidth of up to 2,039 GB/s. However, the company has not disclosed the subject matter it used to train the new model.
Apple’s Journey with Generative AI
While Apple is still in the early stages of its generative AI journey with Ferret, the ultimate goal is to have a model like Ferret work effectively on a smartphone.
OpenAI’s GPT4 is thought to have in excess of 1 trillion parameters, but mobile phones can currently only handle LLMs with around 10 billion parameters.
To this end, Apple researchers have recently made a breakthrough, demonstrating how to supplement smartphone RAM with onboard flash storage. This allows for larger models than might otherwise be possible to run on a device.
The Rise of Multimodal AI
Multimodal AI combines different data types like text, images, and speech within a single model. This allows for a wider range of real-world applications compared to single-modal models.
Apple’s introduction of Ferret represents its first foray into this critical AI space, one that is expected to become ubiquitous in the future across devices and platforms.
What’s Next for Apple in AI?
The launch of Ferret signals Apple’s growing ambition to be an AI leader. With strong compute hardware via its A-series chips, Apple is in a unique position to push multimodal AI to iPhones and Mac devices.
This could soon make Apple a formidable player alongside other tech giants investing heavily in AI like Google, Meta, and Microsoft. Exciting times lie ahead.
Conclusion
Apple’s introduction of Ferret marks a significant step in the company’s AI journey. By making Ferret an open-source project, Apple is not only contributing to the AI community but also paving the way for more advanced and efficient AI models on smartphones.
As the field of AI continues to evolve, it will be interesting to see how Apple and other tech giants leverage these advancements to enhance user experiences.
