安装vllm的CPU版本

原创已于 2025-12-03 17:58:05 修改 · 252 阅读

0 ·

CC 4.0 BY-SA版权

文章标签：

#人工智能 #linux #conda

于 2025-12-01 11:31:03 首次发布

大模型专栏收录该内容

2 篇文章

订阅专栏

安装vllm的CPU版本

intel

硬件信息

cpu: Intel(R) Xeon(R) Gold 6348 CPU @ 2.60GHz
系统: Ubuntu 22.04.5 LTS

安装步骤

配置g++和numa的依赖

#创建一个conda环境并配置gcc/g++版本，为了防止出现问题，请使用 g++>=12.3.0
conda create -n vllm_cpu python=3.10
conda activate vllm_cpu
conda install gxx

#配置numa.h的依赖包
conda install conda-forge::libnuma
#check下libnuma是否安装好
find ~/miniforge3/envs/llm -name "numa.h" 2>/dev/null
find ~/miniforge3/envs/llm -name "*numa*" 2>/dev/null

#完了记得加入环境变量，后面编译时会用到
export C_INCLUDE_PATH="~/miniforge3/envs/llm/include:$C_INCLUDE_PATH"
export CPLUS_INCLUDE_PATH="~/miniforge3/envs/llm/include:$CPLUS_INCLUDE_PATH"

配置cmake和pip依赖，并编译

#克隆仓库并进入
git clone https://github.com/vllm-project/vllm.git vllm_source
cd vllm_source/

#安装依赖的cmake和setuptools-scm
pip install --upgrade pip
pip install "cmake>=3.26" wheel packaging ninja "setuptools-scm>=8" numpy -i https://pypi.tuna.tsinghua.edu.cn/simple/

#按照requirements/cpu.txt安装其他依赖包
pip install -v -r requirements/cpu.txt --extra-index-url https://download.pytorch.org/whl/cpu -i https://pypi.tuna.tsinghua.edu.cn/simple/

#编译，如果找不到numa.h，参考上一步的配置好numa的环境变量再编译
VLLM_TARGET_DEVICE=cpu python setup.py install

查看是否安装成功,

(llm) ziyi.he@mlc05:~/Code/vllm_source$ vllm --version
INFO 12-01 11:24:36 [importing.py:44] Triton is installed but 0 active driver(s) found (expected 1). Disabling Triton to prevent runtime errors.
INFO 12-01 11:24:36 [importing.py:68] Triton not installed or not compatible; certain GPU-related functions will not be available.
0.11.2.dev417+gf72a817bd.cpu

如果有报错，需要更新下环境变量

export LD_LIBRARY_PATH="$CONDA_PREFIX/lib:$LD_LIBRARY_PATH"