pytorch-encoding安装踩坑记录

最新推荐文章于 2024-11-21 10:58:01 发布

胖虎的编程日记

最新推荐文章于 2024-11-21 10:58:01 发布

阅读量3.2k

点赞数 2

分类专栏： Linux

本文链接：https://blog.csdn.net/Markfieraloha/article/details/108369816

版权

这篇博客记录了在非root环境下安装CUDA 10.1和9.2的过程，以及在安装PyTorch 1.4.0和PyTorch-Encoding时遇到的CUDA_HOME设置错误。通过修改.bashrc文件解决了编译问题。还探讨了在安装过程中因环境中的包名与文件夹名冲突导致的import错误，并提出了解决方案。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

安装非root的cuda10.1：(失败)

sh cuda_10.1.105_418.39_linux.run

提示如下错误：

Log file not open.
Segmentation fault (core dumped)

可能是数组越界，按照网上的方法进行尝试解决：

#设置core大小为无限      ulimit -c unlimited

#设置文件大小为无限       ulimit unlimited

仍然提示一样的错误！

安装非root的cuda9.2（成功）
建立conda环境：pytorch-encoding_python3.6
- 安装pytorch 1.4.0 （成功）
- 安装pytorch-encoding（失败成功）
  - 失败原因：CUDA_HOME 设置错误，重新编写.bashrc 之后成功编译！

-- Building version 1.2.2b20200901
c++:  ['/home/yuanquan/PyTorch-Encoding/encoding/lib/cpu/syncbn_cpu.cpp', 
'/home/yuanquan/PyTorch Encoding/encoding/lib/cpu/encoding_cpu.cpp', 
'/home/yuanquan/PyTorch-Encoding/encoding/lib/cpu/rectify_cpu.cpp', 
'/home/yuanquan/PyTorch-Encoding/encoding/lib/cpu/roi_align_cpu.cpp', 
'/home/yuanquan/PyTorch-Encoding/encoding/lib/cpu/nms_cpu.cpp', 
'/home/yuanquan/PyTorch-Encoding/encoding/lib/cpu/operator.cpp']
cuda:  ['/home/yuanquan/PyTorch-Encoding/encoding/lib/gpu/operator.cpp', 
'/home/yuanquan/PyTorch-Encoding/encoding/lib/gpu/nms_kernel.cu', 
'/home/yuanquan/PyTorch-Encoding/encoding/lib/gpu/roi_align_kernel.cu', 
'/home/yuanquan/PyTorch-Encoding/encoding/lib/gpu/lib_ssd.cu', 
'/home/yuanquan/PyTorch-Encoding/encoding/lib/gpu/rectify_cuda.cu', 
'/home/yuanquan/PyTorch-Encoding/encoding/lib/gpu/encoding_kernel.cu',
'/home/yuanquan/PyTorch-Encoding/encoding/lib/gpu/activation_kernel.cu', 
'/home/yuanquan/PyTorch-Encoding/encoding/lib/gpu/syncbn_kernel.cu']
running install
running bdist_egg
running egg_info
writing torch_encoding.egg-info/PKG-INFO
writing dependency_links to torch_encoding.egg-info/dependency_links.txt
writing requirements to torch_encoding.egg-info/requires.txt
writing top-level names to torch_encoding.egg-info/top_level.txt
reading manifest file 'torch_encoding.egg-info/SOURCES.txt'
writing manifest file 'torch_encoding.egg-info/SOURCES.txt'
installing library code to build/bdist.linux-x86_64/egg
running install_lib
running build_py
copying encoding/version.py -> build/lib.linux-x86_64-3.6/encoding
running build_ext
building 'encoding.gpu' extension
gcc -pthread -B /home/yuanquan/anaconda3/envs/pytorch-encoding/compiler_compat 
-Wl,--sysroot=/ -Wsign-compare -DNDEBUG -g -fwrapv -O3 -Wall -Wstrict-prototypes -fPIC -DWITH_CUDA -I/home/yuanquan/PyTorch-Encoding/encoding/lib/cpu 
-I/home/yuanquan/anaconda3/envs/pytorch-encoding/lib/python3.6/site-packages/torch/include 
-I/home/yuanquan/anaconda3/envs/pytorch-encoding/lib/python3.6/site-packages/torch/include/torch/csrc/api/include 
-I/home/yuanquan/anaconda3/envs/pytorch-encoding/lib/python3.6/site-packages/torch/include/TH 
-I/home/yuanquan/anaconda3/envs/pytorch-encoding/lib/python3.6/site-packages/torch/include/THC 
-I/home/yuanquan/PyTorch-Encoding/encoding/lib/gpu 
-I/home/yuanquan/anaconda3/envs/pytorch-encoding/lib/python3.6/site-packages/torch/include 
-I/home/yuanquan/anaconda3/envs/pytorch-encoding/lib/python3.6/site-packages/torch/include/torch/csrc/api/include 
-I/home/yuanquan/anaconda3

最低0.47元/天解锁文章