我在《把离线AI智能体装进口袋里》（The Dawn of Offline AI Agents in Your Pocket）一文中对此进行了详细阐述。但文章中的示例更像是Demo，而非生产解决方案。像 Gemma 3n 这样的模型虽然能够很好地处理函数调用，但它们体积过大：无法集成到应用程序包中，需要单独下载，即使在旗舰机型上推理速度也很慢。在低端设备上，它们根本无法运行。而较小的型号则经常出现故障，难以记住工具。
the 4730, 4731, 4736, 4737, and 4738. These various models were introduced from
。同城约会对此有专业解读
In 2025, their data also showed a slight increase in labelling fraud, such as olive oil labelled falsely as being extra virgin, or non-organic crops marked as organic.
我的中文學習旅程則是用截然不同的方式開始的。
AI safety and wide distribution of…