苹果公司的研究团队近日宣布推出一款名为 MM1 的多模态大模型，该模型拥有高达 300 亿参数，并采用了混合专家（MoE）架构，展示了苹果在人工智能领域的最新突破。这一研究成果在一篇由多位作者共同署名的论文《MM1: Methods, Analysis & Insights from Multimodal LLM Pre-training》中得到详细阐述。MM1 模型不仅在预训练阶段实现了当前最优（SOTA）的性能，而且在多个已有的多模态基准上经过监督微调后，也展现了强大的竞争力。这标志着苹果在人工智能技术研究上的又一重要进展，为未来的智能应用和交互开辟了新的可能性。
英文标题：Apple Releases 30B Parameter Multimodal Model MM1, Pushing the Boundaries of AI
英文关键词：Apple AI Research, Multimodal Models, Mixture of Experts
英文新闻内容：Apple’s research team has unveiled MM1, a multimodal model with up to 30 billion parameters and a Mixture of Experts (MoE) architecture, marking a significant step forward in AI research. The model’s capabilities are detailed in a paper titled “MM1: Methods, Analysis & Insights from Multimodal LLM Pre-training”, authored by a team of researchers. MM1 has demonstrated state-of-the-art performance in pre-training and competitive results on several multimodal benchmarks after fine-tuning, signaling new possibilities for intelligent applications and user interaction.
【来源】https://mp.weixin.qq.com/s/i9bx6M32uk4Jq2KSRhv4ng
Views: 1