[ DeepSeek R1本地部署 ]

发表评论
74 views

DeepSeek R1部署到debian,ubuntu，的无页面系统中

ollama官网

curl -fsSL https://ollama.com/install.sh | sh

使用ollama拉取并且运行(根据自己配置选择)

# 下载 1.5B 版本（适用于 CPU 及低配 GPU）
ollama run deepseek-r1:1.5b

# 下载 8B 版本
ollama run deepseek-r1:8b

# 下载 14B 版本
ollama run deepseek-r1:14b

# 下载 32B 版本
ollama run deepseek-r1:32b

# 下载 70B 版本（最高性能）
ollama run deepseek-r1:70b

查看状态

root@CT200:~# systemctl status ollama
* ollama.service - Ollama Service
     Loaded: loaded (/etc/systemd/system/ollama.service; enabled; vendor preset: enabled)
     Active: active (running) since Sat 2025-02-08 06:24:08 UTC; 1h 9min ago
   Main PID: 192270 (ollama)
      Tasks: 17 (limit: 76773)
     Memory: 2.4G
        CPU: 3min 37.230s
     CGroup: /system.slice/ollama.service
             |-192270 /usr/local/bin/ollama serve
             `-203229 /usr/local/lib/ollama/runners/cpu_avx2/ollama_llama_server runner --model /usr/share/ollama/.ollama/models/blobs/sha256-aabd4debf0c8f0888>

Feb 08 07:29:18 CT200 ollama[192270]: llm_load_print_meta: FIM PAD token    = 151662 '<|fim_pad|>'
Feb 08 07:29:18 CT200 ollama[192270]: llm_load_print_meta: FIM REP token    = 151663 '<|repo_name|>'
Feb 08 07:29:18 CT200 ollama[192270]: llm_load_print_meta: FIM SEP token    = 151664 '<|file_sep|>'
Feb 08 07:29:18 CT200 ollama[192270]: llm_load_print_meta: EOG token        = 151643 '<<EF><BD><9C>end<E2><96><81>of<E2><96><81>sentence<EF><BD><9C>>'
Feb 08 07:29:18 CT200 ollama[192270]: llm_load_print_meta: EOG token        = 151662 '<|fim_pad|>'
Feb 08 07:29:18 CT200 ollama[192270]: llm_load_print_meta: EOG token        = 151663 '<|repo_name|>'
Feb 08 07:29:18 CT200 ollama[192270]: llm_load_print_meta: EOG token        = 151664 '<|file_sep|>'
Feb 08 07:29:18 CT200 ollama[192270]: llm_load_print_meta: max token length = 256
Feb 08 07:29:18 CT200 ollama[192270]: llama_model_load: vocab only - skipping tensors
Feb 08 07:29:18 CT200 ollama[192270]: [GIN] 2025/02/08 - 07:29:18 | 200 |  4.278917061s |       127.0.0.1 | POST     "/api/generate"

python启动个web聊天

import gradio as gr
import requests

def query_ollama(message):
    response = requests.post('http://localhost:11434/api/generate',
        json={
            "model": "deepseek-r1:1.5b", # 这里名字一定要对上自己安装的版本
            "prompt": message,
            "stream": False
        })
    code = response.status_code
    if code != 200:
        return f"请求失败，错误代码：{code}"
    data = response.json()
    print(data)
    return response.json()['response']

# 创建 Gradio 界面
iface = gr.Interface(
    fn=query_ollama,
    inputs=gr.Textbox(label="输入您的问题"),
    outputs=gr.Textbox(label="AI 回答"),
    title="Deepseek-r1 聊天界面",
    description="一个简单的 Deepseek-r1 模型交互界面"
)

# 启动服务
iface.launch(server_name="0.0.0.0", server_port=7860)