Pytorch安装教程

一、参考资料

超简单的pytorch(GPU版)安装教程(亲测有效)
pytorch官网
pytorch GetStarted
pytorch镜像地址
PyTorch 中文手册(pytorch handbook)
pytorch笔记
Pytorch入门教程
pytorch GetStarted

二、准备工作

pytorch版本,CUDA版本,python版本要匹配

# 博主的环境
windows10
64位系统
cuda11.1
python3.8

# 下载对应的 pytorch、torchvision 版本
torch-1.8.1+cu111-cp38-cp38-win_amd64.whl
torchvision-0.9.1+cu111-cp38-cp38-win_amd64.whl

# 清华源镜像
pytorch-1.7.1-py3.7_cuda11.0.221_cudnn8.0.5_0.tar.bz2
torchvision-0.8.0-py37_cu110.tar.bz2

1. 查看pytorch 版本

PyTorch Version

2. 查看 torchvision 版本

Torchvision Version

3. 下载安装包

清华源镜像下载安装包: 清华源镜像
或者下载: 其他镜像

三、关键步骤

  1. 安装CPU版本的pytorch

    # 安装cpu版本的pytorch
    pip install torch
    pip install torchvision
    
    # 卸载
    pip uninstall torch
    pip uninstall torchvision
    
  2. 本地安装GPU版本的pytorch

    # 本地安装GPU版本的pytorch
    pip install torch-1.8.1+cu111-cp38-cp38-win_amd64.whl
    pip install torchvision-0.9.1+cu111-cp38-cp38-win_amd64.whl
    
    # 或者
    conda install --offline pytorch-1.7.1-py3.7_cuda11.0.221_cudnn8.0.5_0.tar.bz2
    conda install --offline torchvision-0.8.0-py37_cu110.tar.bz2
    
  3. 测试是否安装成功

    # 测试是否安装成功
    import torch
    print(torch.cuda.is_available())
    device = torch.device("cuda:0" if (torch.cuda.is_available() and True > 0) else "cpu")
    print(torch.cuda.get_device_name(0))
    

四、pytorch支持的gpu算力

import torch
torch.cuda.get_arch_list()

# 输出
['sm_37', 'sm_50', 'sm_60', 'sm_61', 'sm_70', 'sm_75', 'compute_37']

五、FAQ

Q:“libmkl_intel_lp64.so 不能被加载”的错误

Traceback (most recent call last):
  File "", line 1, in 
  File "/home/yichao/miniconda3/envs/tensorRT-pytorch/lib/python3.7/site-packages/torch/__init__.py", line 189, in 
    _load_global_deps()
  File "/home/yichao/miniconda3/envs/tensorRT-pytorch/lib/python3.7/site-packages/torch/__init__.py", line 142, in _load_global_deps
    ctypes.CDLL(lib_path, mode=ctypes.RTLD_GLOBAL)
  File "/home/yichao/miniconda3/envs/tensorRT-pytorch/lib/python3.7/ctypes/__init__.py", line 356, in __init__
    self._handle = _dlopen(self._name, mode)
OSError: libmkl_intel_lp64.so: cannot open shared object file: No such file or directory
解决办法:
方法一:
[libmkl_intel_lp64.so](https://www.cnblogs.com/denny402/p/10848506.html)

# 1. 搜索路径
sudo find /home -name libmkl_intel_lp64.so

显示路径为:
/home/yichao/anaconda3/lib/libmkl_intel_lp64.so

# 2. 添加环境变量
sudo gedit ~/.bashrc

# 2.1 添加一行
export LD_LIBRARY_PATH=/home/yichao/anaconda3/lib:$LD_LIBRARY_PATH

# 3. 更新配置,显示配置
source ~/.bashrc
echo $LD_LIBRARY_PATH

注意:如果是其他的动态链接库文件也找不到,解决方法是一样的

方法二:
如果方法一没有找到 libmkl_intel_lp64.so 文件,尝试用方法二
[Python在Ubuntu下老是报libmkl_intel_lp64.so 不能被加载的解决办法](https://blog.csdn.net/qq_22704577/article/details/53928059)

# 1. debug一下mkl
conda install --debug mkl

# 2. 重新安装mkl包
conda install mkl
  • torchvision 版本错误
RuntimeError: Couldn't load custom C++ ops. This can happen if your PyTorch and torchvision versions are incompatible, or if you had errors while compiling torchvision from source. For further information on the compatible versions, check https://github.com/pytorch/vision#installation for the compatibility matrix. Please check your PyTorch version with torch.__version__ and your torchvision version with torchvision.__version__ and verify if they are compatible, and if not please reinstall torchvision so that it matches your PyTorch install.
错误原因:
pytorch版本与torchvision版本不匹配

解决办法:
torchvision版本对齐
[torchvision](https://github.com/pytorch/vision#installation)

你可能感兴趣的:(编程工具,深度学习)