暫時應該係一般電腦內 (local) 可運行的最強大模型了

[mtr]

Google 昨晚開源的 Gemma3 大模型。

Ollama 點安裝網上有不少教學，可自行找找。Mac Mini M4 最平最低配既版本已可運行 12B，如一般寫個小程式，做撮要，翻譯 (google LLM 一向最強項，Gemma3 話自己識得 140 種文字)，補數，寫範文，12B 已夠用。

27B 估計要新版 Mac Studio 至行得順。

用圖文輸入問佢野，佢話係無問題既，但小弟重未測試過。

https://developers.googleblog.com/en/in ... ng-gemma3/

" ... Gemma can use images and videos as inputs, allowing it to analyze images, answer questions about an image, compare images, identify objects, and even reply about text within an image. Although the model was originally created to work with images of 896x896 pixels, a new adaptive window algorithm is used to segment input images, allowing Gemma 3 to work with high resolution and non-square images. "


https://x.com/ollama/status/1899742981676007791

1B (text-only):
ollama run gemma3:1b

4B:
ollama run gemma3:4b

12B:
ollama run gemma3:12b

27B:
ollama run gemma3:27b

[streamboy]

一人一AI

[joe]