架构图

在WSL2安装多个Ubuntu子系统

意在利用子系统隔离不同的AI探索项目，避免依赖冲突等问题

1. 安装Ubuntu22.04

wsl --install -d Ubuntu-22.04

2. 初始环境

sudo vim /etc/wsl.conf

[network]
hostname = 新主机名
generateHosts = false
generateResolvConf = false

[user]
default = root

sudo vi /etc/hosts

127.0.1.1 新主机名.localdomain 新主机名

sudo vi /etc/resolv.conf

nameserver 8.8.8.8
nameserver 8.8.4.4

sudo vi /etc/systemd/resolved.conf

[Resolve]
DNS=8.8.8.8

sudo systemctl restart systemd-resolved
sudo systemctl restart NetworkManager

3. 更新系统，安装依赖包

sudo apt update && sudo apt upgrade -y
sudo apt install -y net-tools network-manager zstd build-essential
sudo apt install -y cmake libcurl4-openssl-dev checkinstall git curl unzip
sudo ln -fs /bin/bash /bin/sh

4. 更新cmake 3.28

wget https://github.com/Kitware/CMake/releases/download/v3.28.3/cmake-3.28.3-linux-x86_64.sh
chmod +x cmake-3.28.3-linux-x86_64.sh
sudo ./cmake-3.28.3-linux-x86_64.sh --skip-license --prefix=/usr/local

# 备份旧版本的 cmake 链接（可选，但建议做）
sudo mv /usr/bin/cmake /usr/bin/cmake.old
 
# 创建新版本的软链接（指向 /usr/local/bin/cmake）
sudo ln -s /usr/local/bin/cmake /usr/bin/cmake
 
# 同理，更新 cpack、ctest 等相关工具（避免后续报错）
sudo mv /usr/bin/cpack /usr/bin/cpack.old
sudo ln -s /usr/local/bin/cpack /usr/bin/cpack
 
sudo mv /usr/bin/ctest /usr/bin/ctest.old
sudo ln -s /usr/local/bin/ctest /usr/bin/ctest

cmake --version

5. 导出为基础版本

#wsl --export  [子系统名] "导出目标路径"
wsl --export  Ubuntu-22.04 E:\WSL\Ubuntu-22.04.tar

6. 利用WSL导入功能创建新的子系统

wsl --import [子系统名] "子系统路径" "导入来源路径"

wsl --import Ubuntu-22.04-llamacpp "E:\WSL\Ubuntu-22.04-llamacpp" "E:\WSL\Ubuntu-22.04.tar"

7. 与宿主机的网络端口映射

netsh interface portproxy add v4tov4 listenport=[宿主机监听端口] listenaddress=0.0.0.0 connectport=[子系统端口] connectaddress=[子系统IP]

8. 安装新版本的Node.js和npm

#如果之前通过 apt 安装过旧版本 Node.js，建议先卸载避免冲突
sudo apt remove -y nodejs npm
sudo apt autoremove -y

# 下载并安装 nvm（使用官方脚本，版本号可能更新，以官网为准）
curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.39.7/install.sh | bash
source ~/.bashrc

# 输出版本号即成功
nvm --version

# 安装 Node.js 20.x LTS（会自动安装对应版本的 npm）
nvm install 20

# 设置 20.x 为默认版本（避免重启终端后版本切换）
nvm alias default 20

# 应输出 v20.x.x（如 v20.17.0）
node -v

# 应输出对应的 npm 版本（如 10.8.2）
npm -v

#安装pnpm
npm install -g pnpm
pnpm --version

9. 安装uv

curl -LsSf https://astral.sh/uv/install.sh | sh
echo 'export PATH="$HOME/.local/bin:$PATH"' >> ~/.bashrc
source ~/.bashrc

安装CUDA，cuDNN，NCCL，torch

1. 安装CUDA Toolkit
查询CUDA版本

nvidia-smi

在 https://developer.nvidia.com/cuda-downloads 根据操作系统及CUDA版本，下载并执行对应的run file

#13.2
wget https://developer.download.nvidia.com/compute/cuda/13.2.0/local_installers/cuda_13.2.0_595.45.04_linux.run
sudo sh cuda_13.2.0_595.45.04_linux.run

#12.8
wget https://developer.download.nvidia.com/compute/cuda/12.8.0/local_installers/cuda_12.8.0_570.86.10_linux.run
sudo sh cuda_12.8.0_570.86.10_linux.run

#12.9
wget https://developer.download.nvidia.com/compute/cuda/12.9.0/local_installers/cuda_12.9.0_575.51.03_linux.run
sudo sh cuda_12.9.0_575.51.03_linux.run

2. 配置环境变量

#13.2
echo 'export CUDA_HOME=/usr/local/cuda-13.2' >>  ~/.bashrc
#12.8
echo 'export CUDA_HOME=/usr/local/cuda-12.8' >>  ~/.bashrc
#12.9
echo 'export CUDA_HOME=/usr/local/cuda-12.9' >>  ~/.bashrc

echo 'export PATH=$PATH:${CUDA_HOME}/bin' >>  ~/.bashrc
echo 'export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:${LD_LIBRARY_PATH}/lib64' >>  ~/.bashrc
echo 'export PATH=$PATH:/home/ubuntu/.local/bin' >>  ~/.bashrc

source ~/.bashrc

3. 检查nvcc版本

nvcc --version

4. 安装cuDNN
在 [https://developer.nvidia.com/rdp/cudnn-archive) 根据操作系统及CUDA版本下载并安装cuDNN

#解压缩
tar -xvf cudnn-linux-x86_64-8.9.7.29_cuda12-archive.tar.xz

#复制到cuda目录
#13.2
sudo cp cudnn-linux-x86_64-8.9.7.29_cuda12-archive/include/cudnn* /usr/local/cuda-13.2/include
sudo cp -P cudnn-linux-x86_64-8.9.7.29_cuda12-archive/lib/libcudnn* /usr/local/cuda-13.2/lib64
#12.8
sudo cp cudnn-linux-x86_64-8.9.7.29_cuda12-archive/include/cudnn* /usr/local/cuda-12.8/include
sudo cp -P cudnn-linux-x86_64-8.9.7.29_cuda12-archive/lib/libcudnn* /usr/local/cuda-12.8/lib64
#12.9
sudo cp cudnn-linux-x86_64-8.9.7.29_cuda12-archive/include/cudnn* /usr/local/cuda-12.9/include
sudo cp -P cudnn-linux-x86_64-8.9.7.29_cuda12-archive/lib/libcudnn* /usr/local/cuda-12.9/lib64

#修改文件权限
#13.2
sudo chmod a+r /usr/local/cuda-13.2/include/cudnn*.h /usr/local/cuda-13.2/lib64/libcudnn*
#12.8
sudo chmod a+r /usr/local/cuda-12.8/include/cudnn*.h /usr/local/cuda-12.8/lib64/libcudnn*
#12.8
sudo chmod a+r /usr/local/cuda-12.9/include/cudnn*.h /usr/local/cuda-12.9/lib64/libcudnn*

#显示版本表示安装成功
cat /usr/local/cuda/include/cudnn_version.h | grep CUDNN_MAJOR -A 2

5. 安装NCCL
在 [https://developer.nvidia.com/nccl/nccl-download/) 根据操CUDA版本下载并安装NCCL

wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/cuda-keyring_1.1-1_all.deb
sudo dpkg -i cuda-keyring_1.1-1_all.deb
sudo apt update

#13.2
sudo apt install libnccl2=2.30.3-1+cuda13.2 libnccl-dev=2.30.3-1+cuda13.2
#12.8
sudo apt install libnccl2=2.26.2-1+cuda12.8 libnccl-dev=2.26.2-1+cuda12.8
#12.9
sudo apt install libnccl2=2.30.3-1+cuda12.9 libnccl-dev=2.30.3-1+cuda12.9

6. 安装torch

#13.2
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu130
#12.8
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu128
#12.9
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu129

7. 验证脚本

import torch
import platform

def get_system_info():
    return {
        "系统": platform.system(),
        "Python 版本": platform.python_version(),
        "PyTorch 版本": torch.__version__,
        "CUDA 可用": torch.cuda.is_available(),
        "CUDA 版本": torch.version.cuda,
        "MPS 可用": hasattr(torch.backends, "mps") and torch.backends.mps.is_available(),
        "显卡信息": torch.cuda.get_device_name(0) if torch.cuda.is_available() else "无"
    }

def test_mps():
    if not torch.backends.mps.is_available():
        if not torch.backends.mps.is_built():
            print("MPS 不可用，因为当前的 PyTorch 安装未启用 MPS。")
        else:
            print("MPS 不可用，因为当前的 MacOS 版本不是 12.3+，或者此机器上没有启用 MPS 的设备。")
    else:
        mps_device = torch.device("mps")
        # 在 mps 设备上创建一个张量
        x = torch.ones(5, device=mps_device)
        # 或者
        x = torch.ones(5, device="mps")
        # 任何操作都在 GPU 上进行
        y = x * 2
        # 将模型移动到 mps 设备
        model = YourFavoriteNet()
        model.to(mps_device)
        # 现在每次调用都在 GPU 上运行
        pred = model(x)

if __name__ == "__main__":
    info = get_system_info()
    for k, v in info.items():
        print(f"{k}: {v}")
    test_mps()

本地部署Ollama

1. 安装ollama

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

2. 配置服务

sudo vi /etc/systemd/system/ollama.service

# 文件内容
[Unit]
Description=Ollama Service
After=network-online.target

[Service]
ExecStart=/usr/bin/ollama serve
User=ollama
Group=ollama
Restart=always
RestartSec=3
Environment="PATH=$PATH"
Environment="OLLAMA_HOST=0.0.0.0"
Environment="OLLAMA_ORIGINS=*"

[Install]
WantedBy=default.target

# 重载配置
sudo systemctl daemon-reload
# 启动服务
sudo systemctl start ollama.service
# 查看服务状态
sudo systemctl status ollama.service
# 设置服务开机自启动
sudo systemctl enable ollama.service

3. 下载模型

ollama pull qwen3.5:35b

4. 安装Ngnix

#安装 Nginx
sudo apt install nginx -y

#启动 Nginx 并设置开机自启
sudo systemctl start nginx
sudo systemctl enable nginx

#验证 Nginx 是否运行，如果 Nginx 正常运行，你应该看到 `active (running)` 状态。 
sudo systemctl status nginx

5. 配置 API Key 验证

sudo tee /etc/nginx/conf.d/ollama.conf <<'EOF'
server {
listen 9180;
location / {
if ($http_authorization != "[API KEY]") {
return 403;
}
proxy_pass http://localhost:11434;
}
}
EOF

6. 设置宿主机端口映射

netsh interface portproxy add v4tov4 listenport=9180 listenaddress=0.0.0.0 connectport=9180 connectaddress=172.20.149.74

本地部署Llama.cpp

1. 安装llamp.cpp

# 克隆仓库
cd /usr/local
git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp

# 使用 CMake 编译（推荐方式）
mkdir build && cd build

# 编译时启用 CUDA
cmake .. -DLLAMA_CUDA=ON
cmake --build . --config Release -j$(nproc)

export PATH=$PATH:/usr/local/llama.cpp/build/bin
source ~/.bashrc

2. 安装modelscope

sudo apt install python3-pip
pip install modelscope -i https://pypi.tuna.tsinghua.edu.cn/simple

export PATH=$PATH:/home/ubuntu/.local/bin
source ~/.bashrc

3. 从modescope下载模型

在https://www.modelscope.cn/models下载合适的模型

modelscope download --model [模型合集] [模型] --local_dir [下载路径]

modelscope download --model Qwen/Qwen3.5-27B-FP8 README.md --local_dir /usr/local/llama.cpp/build/models

4. 运行模型

# 回到 build 目录
cd /usr/local/llama.cpp/build/bin

# 基础运行方式
./llama-cli \
    -m ~/models/Llama-3.2-1B-Instruct-Q4_K_M.gguf \
    -p "你好，请介绍一下自己" \
    -n 512

# 交互式聊天模式
./llama-cli \
    -m ~/models/Llama-3.2-1B-Instruct-Q4_K_M.gguf \
    --chat-template llama3 \
    -cnv

# 启动 HTTP API 服务器
./llama-server \
    -m ~/models/Llama-3.2-1B-Instruct-Q4_K_M.gguf \
    --host 0.0.0.0 \
    --port 8080 \
    -c 4096

5. 配置服务

sudo vi /etc/systemd/system/llama-server.service

设置API端口为9191

[Unit]
Description=llama.cpp HTTP Server
After=network.target

[Service]
Type=simple
User=llama
Group=llama
WorkingDirectory=/usr/local/llama.cpp
ExecStart=/usr/local/llama.cpp/build/bin/llama-server \
  -m /usr/local/llama.cpp/build/models/modei_file_name.guff \
  --port 9191 \
  --host 0.0.0.0 \
  -c 163840 \
  -np 4 \
  --threads 12 \
  --n-gpu-layers 35 \
  --cont-batching \
  -ngl 99999 \
  -b 4096

Restart=always
RestartSec=5
Environment=LD_LIBRARY_PATH=/usr/local/cuda/lib64
LimitNOFILE=65536

[Install]
WantedBy=multi-user.target

6. 安装Ngnix

#安装 Nginx
sudo apt install nginx -y

#启动 Nginx 并设置开机自启
sudo systemctl start nginx
sudo systemctl enable nginx

#验证 Nginx 是否运行，如果 Nginx 正常运行，你应该看到 `active (running)` 状态。 
sudo systemctl status nginx

7. 配置 API Key 验证

sudo tee /etc/nginx/conf.d/llamacpp.conf <<'EOF'
server {
listen 9280;
location / {
if ($http_authorization != "[API KEY]") {
return 403;
}
proxy_pass http://localhost:9191;
}
}
EOF

8. 设置宿主机端口映射

netsh interface portproxy add v4tov4 listenport=9280 listenaddress=0.0.0.0 connectport=9280 connectaddress=172.20.149.74

本地部署OpenClaw

本地部署CoPaw

1. 安装CoPaw

curl -fsSL https://copaw.agentscope.io/install.sh | bash

2. 初始化CoPaw

/home/ubuntu/.local/bin/copaw init --defaults
/home/ubuntu/.local/bin/copaw app

3. 启动CoPaw

/home/ubuntu/.local/bin/copaw app

4. 打开控制台
http://127.0.0.1:8088

5. 创建service

sudo tee /etc/systemd/system/copaw.service <<EOF
[Unit]
Description=CoPaw Inference Service
After=network.target
[Service]
Type=simple
User=ubuntu
Group=ubuntu
Restart=always
WorkingDirectory=/home/ubuntu/.copaw
ExecStart=/home/ubuntu/.local/bin/copaw app
ExecStop=/home/ubuntu/.local/bin/copaw shutdown
Restart=always
RestartSec=5
[Install]
WantedBy=multi-user.target
EOF

sudo systemctl start copaw.service
sudo systemctl enable copaw.service

6. 安装Ngnix

#安装 Nginx
sudo apt install nginx -y

#启动 Nginx 并设置开机自启
sudo systemctl start nginx
sudo systemctl enable nginx

#验证 Nginx 是否运行，如果 Nginx 正常运行，你应该看到 `active (running)` 状态。 
sudo systemctl status nginx

7. 配置代理
CoPaw默认绑定127.0.0.1地址，修改配置文件后还是被覆盖，所以改用NG做代理

sudo tee /etc/nginx/conf.d/copaw.conf <<'EOF'
server {
listen 18088;
location / {
proxy_pass http://localhost:8088;
}
}
EOF

8. 设置宿主机端口映射

netsh interface portproxy add v4tov4 listenport=18088 listenaddress=0.0.0.0 connectport=18088 connectaddress=172.20.149.74