FunctionGemma 是 Google 最小的函数调用专用模型——2.7 亿个参数，288 MB，解码速度约为 126 tok/s。没错，它需要微调（准确率从 58% 提升到 85%），没错，它使用了一种奇怪的自定义格式，而不是 JSON。但它适用于任何手机，响应速度极快，而且确实有效。现在就可以构建带有离线 AI 代理的应用——体积小、速度快、可靠性高，足以满足生产环境的需求。无需等待模型体积更小、设备速度更快的“神奇未来”，未来已来！
Маргарита Щигарева
。91视频对此有专业解读
Third, implement quick wins on those priority pieces. Add "Last updated: [current date]" to each. Create a simple FAQ section addressing three to five common questions related to each article's topic. Add specific statistics or data points if they're currently missing. These improvements take hours rather than days but can meaningfully impact AI visibility.
Back in 2024 I learned about SDF (signed distance field) rendering of fonts. I was trying to implement outlines and shadows in a single pass instead of drawing over the text multiple times in different styles. I intended to use these fonts for two different projects, a game and a map generator. I got things working but didn’t fully understand why certain things worked or didn’t work. I wrote some notes on my site about what I tried. In the end, I stopped working on both the game’s fonts and the map generator, so I put all of this on hold.
蜜蜂一蜇，就能治病？这不是武侠小说里的情节，而是一种古老的中医疗法——蜂疗。