前言

vLLM在开启多显卡并行模式下，-tp 2 或者 --tensor-parallel-size 2，运行报错提示如下：

The above exception was the direct cause of the following exception:Traceback (most recent call last): File "/usr/lib/python3.8/runpy.py", line 194, in _run_module_as_main return _run_code(code, main_globals, None, File "/usr/lib/python3.8/runpy.py", line 87, in _run_code exec(code, run_globals) File "/usr/local/lib/python3.8/dist-packages/vllm/entrypoints/openai/api_server.py", line 236, in <module> engine = AsyncLLMEngine.from_engine_args(engine_args) File "/usr/local/lib/python3.8/dist-packages/vllm/engine/async_llm_engine.py", line 628, in from_engine_args engine = cls(parallel_config.worker_use_ray, File "/usr/local/lib/python3.8/dist-packages/vllm/engine/async_llm_engine.py", line 321, in init self.engine = self._init_engine(*args, **kwargs) File "/usr/local/lib/python3.8/dist-packages/vllm/engine/async_llm_engine.py", line 369, in _init_engine return engine_class(*args, **kwargs) File "/usr/local/lib/python3.8/dist-packages/vllm/engine/llm_engine.py", line 126, in init self._init_workers_ray(placement_group) File "/usr/local/lib/python3.8/dist-packages/vllm/engine/llm_engine.py", line 304, in _init_workers_ray self._run_workers("init_model", File "/usr/local/lib/python3.8/dist-packages/vllm/engine/llm_engine.py", line 1041, in _run_workers driver_worker_output = getattr(self.driver_worker, File "/usr/local/lib/python3.8/dist-packages/vllm/worker/worker.py", line 94, in init_model init_distributed_environment(self.parallel_config, self.rank, File "/usr/local/lib/python3.8/dist-packages/vllm/worker/worker.py", line 275, in init_distributed_environment cupy_utils.init_process_group( File "/usr/local/lib/python3.8/dist-packages/vllm/model_executor/parallel_utils/cupy_utils.py", line 79, in init_process_group raise ImportError( ImportError: NCCLBackend is not available. Please install cupy. 

一般报错的原因情况有三种

• 未安装cuda-toolkit
• cuda和cupy的版本不匹配
• 为设置 cuda的环境变量 LD_LIBRARY_PATH

安装cuda-toolkit

ubuntu20.04系统，只要是桌面版，都有nouveau驱动，这是一个第三方搞的驱动，开源，很久前很火，但现在英伟达都有很完善的驱动方法，我们不用nouveau，马上卸载禁用！

不做的话，后面cuda环境有变数，概率出现不支持设备情况！

先打开terminal命令窗！输入指令，卸载已有的旧驱动！

dpkg -l | grep -i nvidia
sudo apt-get purge nvidia* libnvidia* -y
sudo apt autoremove && sudo reboot

继续！输入指令，彻底禁用nouveau驱动！
打开blacklist.conf文件

​sudo vi /etc/modprobe.d/blacklist.conf

添加blacklist nouveau，后:wq保存

blacklist nouveau

接着执行以下命令

echo options nouveau modeset=0 | sudo tee -a /etc/modprobe.d/nouveau-kms.conf

然后更新内核！+重启！

sudo update-initramfs -u && sudo reboo

重启后，看是否成功，命令窗输入下面指令，无回复内容，则成功！

lsmod | grep nouveau

接下来，安装显卡驱动！

驱动跟cuda是不同的东西！cuda是一个并行计算平台和编程模型，cuda要用显卡资源来计算，就要通过驱动来链接GPU！

每个cuda版本都有一个相匹配的显卡驱动，cuda安装程序已经把显卡驱动都打包在一起了！

建议一起安装！避免出现版本冲突问题！

到这里下载即可！不要下最新的！很多工具都没适配，会报错！

官网地址：https://developer.nvidia.com/cuda-toolkit-archive
选择你要下载的cuda toolkit版本 以12.1.0为例。vllm


复制粘贴 Base Installer中的命令

wget https://developer.download.nvidia.com/compute/cuda/12.1.0/local_installers/cuda_12.1.0_530.30.02_linux.run
sudo sh cuda_12.1.0_530.30.02_linux.run

上面的窗口输入到窗口，会自动下载！要等一会，耐心等待！

到了这个界面，输入accept，回车确认安装！

来到配置页面，按下图确认！安装！

要等2分钟！成功会有提示！

最后执行命令,配置系统变量

export LD_LIBRARY_PATH=/usr/local/cuda-12.3/lib64:$LD_LIBRARY_PATH

• /usr/local/cuda-12.3/lib64 要和你实际cuda路径相符

打印变量命令

echo $LD_LIBRARY_PATH