如何解决TensorRT安装过程中的依赖冲突问题

问题背景 #

安装TensorRT时遭遇系统包依赖冲突，具体错误如下：

安装tensorrt时的错误 #

The following packages have unmet dependencies:
 tensorrt : Depends: libnvinfer8 (= 8.6.1.6-1+cuda11.8) but it is not going to be installed
            Depends: libnvinfer-plugin8 (= 8.6.1.6-1+cuda11.8) but it is not going to be installed
            Depends: libnvparsers8 (= 8.6.1.6-1+cuda11.8) but it is not going to be installed
            Depends: libnvonnxparsers8 (= 8.6.1.6-1+cuda11.8) but it is not going to be installed
            Depends: libnvinfer-bin (= 8.6.1.6-1+cuda11.8) but it is not going to be installed
            Depends: libnvinfer-dev (= 8.6.1.6-1+cuda11.8) but it is not going to be installed
            Depends: libnvinfer-plugin-dev (= 8.6.1.6-1+cuda11.8) but it is not going to be installed
            Depends: libnvparsers-dev (= 8.6.1.6-1+cuda11.8) but it is not going to be installed
            Depends: libnvonnxparsers-dev (= 8.6.1.6-1+cuda11.8) but it is not going to be installed
            Depends: libnvinfer-samples (= 8.6.1.6-1+cuda11.8) but it is not going to be installed
E: Unable to correct problems, you have held broken packages

安装libnvinfer8时的关键错误 #

libnvinfer8 : Depends: libcublas.so.11 but it is not installable or
                        libcublas-11-1 but it is not installable or
                        libcublas-11-0 but it is not installable
E: Unable to correct problems, you have held broken packages

本地环境说明：

• 已安装CUDA11.2，libcublas.so.11文件存在于/usr/local/cuda/lib64目录，且已配置LD_LIBRARY_PATH
• nvcc --version输出为CUDA11.2版本

核心原因 #

你尝试安装的TensorRT 8.6.1是针对CUDA11.8编译的版本，与本地CUDA11.2不兼容。apt的依赖检查仅识别系统包管理中的库文件，不认可手动放置或CUDA本地安装的库，因此触发依赖错误。

解决步骤 #

1. 安装与CUDA11.2匹配的TensorRT版本 #

TensorRT与CUDA版本存在严格对应关系，CUDA11.2兼容的TensorRT版本为8.2.x、8.4.x（例如8.2.3.0、8.4.3.1）。

• 若使用apt安装：
  确保已添加NVIDIA官方源后，指定版本安装：

  sudo apt-get install tensorrt=8.2.3.0-1+cuda11.2
  sudo apt-get install python3-libnvinfer=8.2.3.0-1+cuda11.2
  

• 若找不到对应apt包，可下载对应版本的.deb包，用以下命令安装：

  sudo dpkg -i tensorrt_8.2.3.0-1+cuda11.2_amd64.deb
  sudo apt-get -f install
  

2. 用pip安装TensorRT（Python环境推荐） #

直接通过pip安装TensorRT Python包，避开apt的系统级依赖检查：

pip install tensorrt==8.6.1 --extra-index-url https://pypi.nvidia.com

安装完成后，确保LD_LIBRARY_PATH包含/usr/local/cuda/lib64即可正常使用。

3. 手动解压TensorRT tar包配置 #

若上述方法不适用，可下载对应CUDA11.2的TensorRT tar包：

1. 解压到指定目录，例如/opt/tensorrt：

   tar -xzf TensorRT-8.2.3.0.Linux.x86_64-gnu.cuda-11.2.cudnn8.2.tar.gz -C /opt/
   

2. 添加环境变量：

   export PATH=/opt/tensorrt/bin:$PATH
   export LD_LIBRARY_PATH=/opt/tensorrt/lib:$LD_LIBRARY_PATH
   

3. 安装Python包：

   pip install /opt/tensorrt/python/tensorrt-8.2.3.0-cp37-none-linux_x86_64.whl
   

4. 修复系统损坏的包依赖 #

先清理并修复系统包依赖，再重新安装：

sudo apt-get clean
sudo apt-get autoclean
sudo apt-get update
sudo apt-get -f install
sudo dpkg --configure -a

内容的提问来源于stack exchange，提问作者Rajesh Gupta