Deepseek本地部署指南：在linux服务器部署，在mac远程web-ui访问

1. 在Linux服务器上部署DeepSeek模型

要在 Linux 上通过 Ollama 安装和使用模型，您可以按照以下步骤进行操作：

步骤 1：安装 Ollama

安装 Ollama： 使用以下命令安装 Ollama：

curl -sSfL ollama /install.sh | sh

验证安装： 安装完成后，您可以通过以下命令验证 Ollama 是否安装成功：

ollama --version

步骤 2：下载模型 ollama run deepseek-r1:32b

这将下载并启动DeepSeek R1 32B模型。

DeepSeek R1 蒸馏模型列表

模型名称参数量基础架构适用场景DeepSeek-R1-Distill-Qwen-1.5B1.5BQwen2.5适合移动设备或资源受限的终端DeepSeek-R1-Distill-Qwen-7B7BQwen2.5适合普通文本生成工具DeepSeek-R1-Distill-Llama-8B8BLlama3.1适合小型企业日常文本处理DeepSeek-R1-Distill-Qwen-14B14BQwen2.5适合桌面级应用DeepSeek-R1-Distill-Qwen-32B32BQwen2.5适合专业领域知识问答系统DeepSeek-R1-Distill-Llama-70B70BLlama3.3适合科研、学术研究等高要求场景

RTX 4090 显卡显存为 24GB，32B 模型在 4-bit 量化下约需 22GB 显存，适合该硬件。32B 模型在推理基准测试中表现优异，接近 70B 模型的推理能力，但对硬件资源需求更低。

步骤 3：运行模型 ollama run deepseek-r1:32b

通过上面的步骤，已经可以直接在 Linux服务器通过命令行的形式使用Deepseek了。但是不够友好，下面介绍更方便的形式。

2. 在linux服务器配置Ollama服务

1. 设置Ollama服务配置 设置OLLAMA_HOST=0.0.0.0环境变量，这使得Ollama服务能够监听所有网络接口，从而允许远程访问。

sudo vi /etc/systemd/system/ollama.service [Unit] Description=Ollama Service After=network-online.target [Service] ExecStart=/usr/local/bin/ollama serve User=ollama Group=ollama Restart=always RestartSec=3 Environment="OLLAMA_HOST=0.0.0.0" Environment="PATH=/usr/local/cuda/bin:/home/bytedance/miniconda3/bin:/home/bytedance/miniconda3/condabin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/games:/usr/local/games:/snap/bin" [Install] WantedBy=default.target

2. 重新加载并重启Ollama服务

sudo systemctl daemon-reload sudo systemctl restart ollama 验证Ollama服务是否正常运行 运行以下命令，确保Ollama服务正在监听所有网络接口： sudo netstat -tulpn | grep ollama

您应该看到类似以下的输出，表明Ollama服务正在监听所有网络接口（0.0.0.0）：

tcp 0 0 0.0.0.0:11434 0.0.0.0:* LISTEN - ollama

4. 配置防火墙以允许远程访问

为了确保您的Linux服务器允许从外部访问Ollama服务，您需要配置防火墙以允许通过端口11434的流量。

sudo ufw allow 11434/tcp sudo ufw reload

5. 验证防火墙规则 确保防火墙规则已正确添加，并且端口11434已开放。您可以使用以下命令检查防火墙状态：

sudo ufw status 状态： 激活 至 动作 来自 - -- -- 22/tcp ALLOW Anywhere 11434/tcp ALLOW Anywhere 22/tcp (v6) ALLOW Anywhere (v6) 11434/tcp (v6) ALLOW Anywhere (v6)

6. 测试远程访问 在完成上述配置后，您可以通过远程设备（如Mac）测试对Ollama服务的访问。 在远程设备上测试连接： 在Mac上打开终端，运行以下命令以测试对Ollama服务的连接：

curl http://10.37.96.186:11434/api/version

显示

{"version":"0.5.7"}

测试问答

curl -X POST http://10.37.96.186:11434/api/generate \ -H "Content-Type: application/json" \ -d '{"model": "deepseek-r1:32b", "prompt": "你是谁?"}'

显示

{"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:15.118616168Z","response":"\u003cthink\u003e","done":false} {"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:15.150938966Z","response":"\n\n","done":false} {"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:15.175255854Z","response":"\u003c/think\u003e","done":false} {"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:15.199509353Z","response":"\n\n","done":false} {"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:15.223657359Z","response":"您好","done":false} {"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:15.24788375Z","response":"！","done":false} {"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:15.272068174Z","response":"我是","done":false} {"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:15.296163417Z","response":"由","done":false} {"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:15.320515728Z","response":"中国的","done":false} {"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:15.344646528Z","response":"深度","done":false} {"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:15.36880216Z","response":"求","done":false} {"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:15.393006489Z","response":"索","done":false} {"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:15.417115966Z","response":"（","done":false} {"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:15.441321254Z","response":"Deep","done":false} {"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:15.465439117Z","response":"Seek","done":false} {"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:15.489619415Z","response":"）","done":false} {"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:15.51381827Z","response":"公司","done":false} {"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:15.538012781Z","response":"开发","done":false} {"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:15.562186246Z","response":"的","done":false} {"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:15.586331325Z","response":"智能","done":false} {"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:15.610539651Z","response":"助手","done":false} {"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:15.634769989Z","response":"Deep","done":false} {"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:15.659134003Z","response":"Seek","done":false} {"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:15.683523205Z","response":"-R","done":false} {"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:15.70761762Z","response":"1","done":false} {"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:15.731953604Z","response":"。","done":false} {"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:15.756135462Z","response":"如","done":false} {"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:15.783480232Z","response":"您","done":false} {"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:15.807766337Z","response":"有任何","done":false} {"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:15.831964079Z","response":"任何","done":false} {"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:15.856229156Z","response":"问题","done":false} {"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:15.880487159Z","response":"，","done":false} {"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:15.904710537Z","response":"我会","done":false} {"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:15.929026993Z","response":"尽","done":false} {"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:15.953239249Z","response":"我","done":false} {"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:15.977496819Z","response":"所能","done":false} {"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:16.001763128Z","response":"为您提供","done":false} {"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:16.026068523Z","response":"帮助","done":false} {"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:16.050242581Z","response":"。","done":false} {"model":"deepseek-r1:32b","created_at":"2025-02-06T00:47:16.074454593Z","response":"","done":true,"done_reason":"stop","context":[151644,105043,100165,30,151645,151648,271,151649,198,198,111308,6313,104198,67071,105538,102217,30918,50984,9909,33464,39350,7552,73218,100013,9370,100168,110498,33464,39350,12,49,16,1773,29524,87026,110117,99885,86119,3837,105351,99739,35946,111079,113445,100364,1773],"total_duration":3872978599,"load_duration":2811407308,"prompt_eval_count":6,"prompt_eval_duration":102000000,"eval_count":40,"eval_duration":958000000}

通过上述步骤，已经成功在Linux服务器上配置了Ollama服务，并通过Mac远程访问了DeepSeek模型。接下来，将介绍如何在Mac上安装Web UI，以便更方便地与模型进行交互。

3. 在Mac上安装Web UI

为了更方便地与远程Linux服务器上的DeepSeek模型进行交互，可以在Mac上安装一个Web UI工具。这里我们推荐使用 Open Web UI，它是一个基于Web的界面，支持多种AI模型，包括Ollama。

1. 通过conda安装open-webui 打开终端，运行以下命令创建一个新的conda环境，并指定Python版本为3.11：

conda create -n open-webui-env python=3.11 conda activate open-webui-env pip install open-webui

2. 启动open-webui

open-webui serve

3. 浏览器访问

http://localhost:8080/

使用管理员身份（第一个注册用户）登录

在Open webui界面中，依次点击“展开左侧栏”（左上角三道杠）–>“头像”（左下角）–>管理员面板–>设置（上侧）–>外部连接

在外部连接的Ollama API一栏将switch开关打开，在栏中填上http://10.37.96.186:11434（这是我的服务器地址）

点击右下角“保存”按钮

点击“新对话”（左上角），确定是否正确刷出模型列表，如果正确刷出，则设置完毕。

4. 愉快的使用本地deepseek模型