Apple recently released its large-scale multi-modal basic model MM1, with a parameter size of 30 billion and a hybrid expert (MoE) architecture. It is worth noting that more than half of the members of the model’s research and development team are Chinese, which reflects the important contribution of Chinese scientists in the field of artificial intelligence. MM1 performed well in multi-modal benchmark tests, indicating that Apple has made a major breakthrough in the field of generative artificial intelligence and may lay a solid foundation for future product development. It uses a MoE variant and leads in pre-training indicators and multiple multi-modal benchmark tests. Researchers have also verified the importance of model architecture and pre-training data selection through multiple ablation experiments.
Apple released a large-scale multi-modal basic model called MM1 with 30 billion parameters, using MoE architecture, and more than half of the authors are Chinese. This model is of great significance in the multi-modal field and may herald the launch of related products by Apple in the future. Apple has increased its investment in the field of generative artificial intelligence, demonstrating its determination to make significant progress. MM1 adopts the MoE variant and leads in pre-training indicators and multiple multi-modal benchmark tests. The researchers explored the importance of model architecture and pre-training data selection through multiple ablation experiments.The release of MM1 marks an important step for Apple in the field of artificial intelligence. Its powerful performance and the contribution of the Chinese team are worthy of attention. In the future, we can expect Apple to launch more innovative products and services based on MM1 to further promote the development and application of multi-modal artificial intelligence technology.