在Jetson Nano上运行YOLOv5时遇Torch依赖库缺失错误求助

环境与操作流程 #

• 硬件：Jetson Nano GPU
• 系统：Ubuntu 20.04
• Python版本：3.8
• 已完成的环境搭建命令：

git clone https://github.com/ultralytics/yolov5
cd yolov5
pip install -r requirements.txt

错误信息 #

运行detect.py时触发以下错误：

Traceback (most recent call last):
  File "detect.py", line 37, in <module>
    import torch
  File "/home/jetson/.local/lib/python3.8/site-packages/torch/__init__.py", line 198, in <module>
    _load_global_deps()
  File "/home/jetson/.local/lib/python3.8/site-packages/torch/__init__.py", line 151, in _load_global_deps
    ctypes.CDLL(lib_path, mode=ctypes.RTLD_GLOBAL)
  File "/usr/lib/python3.8/ctypes/__init__.py", line 373, in __init__
    self._handle = _dlopen(self._name, mode)
OSError: libcublas.so.11: cannot open shared object file: No such file or directory

解决方案 #

原因分析 #

Jetson Nano默认JetPack自带的CUDA版本为10.x（对应依赖库libcublas.so.10），但通过pip install安装的通用版torch要求CUDA 11.x（依赖libcublas.so.11），二者版本不匹配导致依赖缺失。

修复步骤 #

1. 卸载当前不兼容的torch

pip uninstall torch -y

2. 安装Jetson适配的torch版本
   针对Python3.8、Ubuntu20.04的适配版本：

wget https://nvidia.box.com/shared/static/p57jwntv436lfrd78inwl7iml6p13fzh.whl -O torch-1.9.0-cp38-cp38-linux_aarch64.whl
pip install torch-1.9.0-cp38-cp38-linux_aarch64.whl

3. 安装匹配的torchvision

sudo apt-get install libjpeg-dev zlib1g-dev -y
git clone --branch v0.10.0 https://github.com/pytorch/vision torchvision
cd torchvision
export BUILD_VERSION=0.10.0
python setup.py install --user

4. 重新安装YOLOv5依赖

cd ../yolov5
pip install -r requirements.txt --force-reinstall

5. 验证CUDA加载状态
   运行以下Python代码确认torch正常识别CUDA：

import torch
print(torch.cuda.is_available())  # 输出True即为成功

内容的提问来源于stack exchange，提问作者Sean Mabli