Google全新開源模型Gemma 3 270M登場！

公司或者做野想用AI處理私隱文件，客戶資料，但又怕私隱，資料唔安全？

Gemma 3 呢個超輕量AI模型，唔使用AI電腦都可以部署
只需0.5GB RAM就喺你嘅手機上跑得掂！唔洗上網連線都用到💼
一般公司d爛電腦都跑到😏

速度快到飛起：用Dynamic GGUF格式，喺手機上可以跑到大約50 token/秒！⚡
慳資源：0.5GB RAM要求，連普通手機都handle到，唔使再靠雲端！🌐

點部署？

strong post leave name.
暫時未玩過手機
電腦host oss 120b 就試過

1B 都唔係好實用 懶得deploy落部手機玩

Mini多數得英文

諗唔到有太多實制既用途 summarize reddit post 求其揾個cloud llm 都得

要自己fine tune
有台仔用左黎fine tune 正體中文

>>> how many r's in word resbeerysssssrrrr
Okay, let’s count the ‘r’s in “Word Resbeerysssssrrrr”!

There are 14 ‘r’s. 😊

Let me know if you’d like me to do another counting task!

>>> how many r's in word resbeerysssssrrrr
Okay, let’s count the ‘r’s in “Word Resbeerysssssrrrr”!

There are 14 ‘r’s. 😊

Let me know if you’d like me to do another counting task!

>>> are you sure?
You are absolutely right to question me! My apologies. I made a mistake.

There are **13** ‘r’s in “Word Resbeerysssssrrrr”.

I was relying on a pattern I was trying to find and got it wrong. Thanks for catching that!

>>> are you sure?
You are absolutely right to ask! I am still under development and sometimes my responses aren’t perfect.

Let me check again...

There are **13** ‘r’s in “Word Resbeerysssssrrrr”.

I appreciate you keeping me honest!

唔知呢個model係咪印度佬train嘅
到錯嘢會keep住道歉
但係永遠唔好做返啱

呢啲細model tokenizer 一定廢
除左寫下email 都冇乜用

private llm 暫時都仲係deepseek的天下
係d傻仔先成日笑鳩deepseek網上無人用

根本人地主力唔撚係個網度