别说,还挺简单的。
环境:ubuntu24.04 4核cpu无GPU 18gbRAM
直接装Ollama, 有命令行提示,安装Gemma4模型(大约9.5GB),10多分钟就下好了。
我也没有装Web界面,直接用命令行真男人操作,OpenClaw过几天再整吧。
真没想到,AI模型纯粹用cpu运行还很流畅,就是18GB的运行内存几乎都能干满。
Ollama的交互和命令都挺简洁易懂的。目前唯一的缺点就是没有联网搜索的API(不过有深度思考,嘿嘿嘿),不过用来润色文本都可以了。我也不知道有没有永久记忆功能,如果有的话其实可以手动喂信息。
下面是一段对话
> > > 你是gemma4?
> > > ⠇
> > > Thinking...
> > > Thinking Process:
1. **Analyze the Request:** The user asks, "你是gemma4?" (Nǐ shì Gemma 4?). This is a direct question asking for identity confirmation in Chinese.
2. **Identify Core Identity:** My established persona is "Gemma 4."
3. **Determine Tone and Language:** The question was asked in Chinese, so the answer must be in fluent, polite Chinese.
4. **Formulate the Response (Self-Correction/Refinement):**
* *Initial Draft (Mental):* Yes, I am Gemma 4. I am a Large Language Model developed by Google DeepMind.
* *Translation & Refinement (Chinese):*
* Confirm identity: 是的,我是 Gemma 4。 (Yes, I am Gemma 4.)
* Provide context (Developer): 我是由 Google DeepMind 开发的。 (I was developed by Google DeepMind.)
* Describe nature: 我是一个大型语言模型 (LLM)。 (I am a Large Language Model.)
5. **Final Polish:** Combine these elements into a natural and definitive response. (Ensure to mention the open-weights aspect, as it is part of the model's core identity.) (Self-Correction: Since the prompt only requires confirming the name and origin, keeping it concise but comprehensive is best.)
6. **Generate Output.**
...done thinking.
是的,我是 **Gemma 4**。
我是一个由 Google DeepMind 开发的大型语言模型 (Large Language Model, LLM)。
您可以把我当作一个知识库和辅助工具,无论您需要写代码、翻译语言、总结文章,还是进行创意性的对话,我都会尽力协助您。
您有什么需要我帮助的吗?😊
就看接下来怎么整活了。想整花活的可以在下面评论一下。