超详细的PyTorch离线安装步骤

有时候为了教学的需要，机房的网络很慢或者机房的机器不允许上外网，而有需要在这些机房进行PyTorch的教学或实验，则需要以离线安装PyTorch的方式。

第一部分：首先，先找一台能够上网的机器去离线下载PyTorch所需要的安装包。

当然，首先需要在此台能够上网的机器上安装相关的python环境，本文是Python3.10.14版本。

接着，打开python.exe文件所在的位置

接着，输入在资源管理器中输入cmd命令

进入命令提示符窗口，如下图：

继续输入命令：pip download torch torchvision torchaudio -d ./pytorch_packages

将所有包下载当前python.exe所在的子目录pytorch_packages中

最后等待下载完成，如提示“Successfully downloaded.......”表示成功下载了所有的包。

下载完成后，目录pytorch_packages如下图所示：

然后，将上述整个目录复制到其它无法上网的机器上。

注意：以下操作是在无法上网的即离线的机器上操作。

首先需要在此台机器上安装Anaconda3和python3.10.14（需要的包可以在网上找到并复制到离线的机器上安装即可。）

本文以Anaconda3（Anaconda3-2021.05-Windows-x86_64.exe）为例说明。

一、安装Anaconda及python3.10虚拟环境

1、安装Anaconda3-2023.09-0-Windows-x86_64.exe，下载该文件到本地机器，然后双击打开该文件，按照提示下一步下一步即可。可以到以下国内镜像下载即可

2、安装python3.10虚拟环境

首先，打开Anaconda Prompt(Anaconda3)如下图所示：

接着，创建python=3.10虚拟环境

conda create -n pytorch python=3.10

(base) C:\Users\Administrator>conda create -n pytorch python=3.10

Collecting package metadata (current_repodata.json): done

Solving environment: done

==> WARNING: A newer version of conda exists. <==

  current version: 4.10.1

  latest version: 24.3.0

Please update conda by running

    $ conda update -n base -c defaults conda

## Package Plan ##

  environment location: d:\Anaconda3\envs\pytorch

  added / updated specs:

    - python=3.10

The following packages will be downloaded:

    package                    |            build

    ---------------------------|-----------------

    bzip2-1.0.8                |       h2bbff1b_5          78 KB

    openssl-3.0.13             |       h2bbff1b_0         7.4 MB

    pip-23.3.1                 |  py310haa95532_0         2.9 MB

    python-3.10.14             |       he1021f5_0        15.9 MB

    setuptools-68.2.2          |  py310haa95532_0         942 KB

    sqlite-3.41.2              |       h2bbff1b_0         894 KB

    tk-8.6.12                  |       h2bbff1b_0         3.1 MB

    tzdata-2024a               |       h04d1e81_0         116 KB

    wheel-0.41.2               |  py310haa95532_0         127 KB

    xz-5.4.6                   |       h8cc25b3_0         587 KB

    zlib-1.2.13                |       h8cc25b3_0         113 KB

    ------------------------------------------------------------

                                           Total:        32.1 MB

The following NEW packages will be INSTALLED:

  bzip2              pkgs/main/win-64::bzip2-1.0.8-h2bbff1b_5

  ca-certificates    pkgs/main/win-64::ca-certificates-2024.3.11-haa95532_0

  libffi             pkgs/main/win-64::libffi-3.4.4-hd77b12b_0

  openssl            pkgs/main/win-64::openssl-3.0.13-h2bbff1b_0

  pip                pkgs/main/win-64::pip-23.3.1-py310haa95532_0

  python             pkgs/main/win-64::python-3.10.14-he1021f5_0

  setuptools         pkgs/main/win-64::setuptools-68.2.2-py310haa95532_0

  sqlite             pkgs/main/win-64::sqlite-3.41.2-h2bbff1b_0

  tk                 pkgs/main/win-64::tk-8.6.12-h2bbff1b_0

  tzdata             pkgs/main/noarch::tzdata-2024a-h04d1e81_0

  vc                 pkgs/main/win-64::vc-14.2-h21ff451_1

  vs2015_runtime     pkgs/main/win-64::vs2015_runtime-14.27.29016-h5e58377_2

  wheel              pkgs/main/win-64::wheel-0.41.2-py310haa95532_0

  xz                 pkgs/main/win-64::xz-5.4.6-h8cc25b3_0

  zlib               pkgs/main/win-64::zlib-1.2.13-h8cc25b3_0

Proceed ([y]/n)? y

Downloading and Extracting Packages

pip-23.3.1           | 2.9 MB    | ############################################################################ | 100%

wheel-0.41.2         | 127 KB    | ############################################################################ | 100%

bzip2-1.0.8          | 78 KB     | ############################################################################ | 100%

openssl-3.0.13       | 7.4 MB    | ############################################################################ | 100%

xz-5.4.6             | 587 KB    | ############################################################################ | 100%

tzdata-2024a         | 116 KB    | ############################################################################ | 100%

setuptools-68.2.2    | 942 KB    | ############################################################################ | 100%

python-3.10.14       | 15.9 MB   | ############################################################################ | 100%

zlib-1.2.13          | 113 KB    | ############################################################################ | 100%

tk-8.6.12            | 3.1 MB    | ############################################################################ | 100%

sqlite-3.41.2        | 894 KB    | ############################################################################ | 100%

Preparing transaction: done

Verifying transaction: done

Executing transaction: done

#

# To activate this environment, use

#

#     $ conda activate pytorch

#

# To deactivate an active environment, use

#

#     $ conda deactivate

1. 创建环境结束后，使用以下指令查看已有的环境：conda env list

(base) C:\Users\Administrator>conda env list

# conda environments:

#

base                  *  d:\Anaconda3

pytorch                  d:\Anaconda3\envs\pytorch

(base) C:\Users\Administrator>

1. 进入已有的环境，输入以下语句切换环境：conda  activate  pytorch

(base) C:\Users\Administrator>conda  activate pytorch

(pytorch) C:\Users\Administrator>

第二部分：安装torch依赖包及torch

1、安装依赖包filelock

(pytorch) D:\pytorch_packages>

(pytorch) D:\pytorch_packages>pip install filelock-3.13.1-py3-none-any.whl

Processing d:\pytorch_packages\filelock-3.13.1-py3-none-any.whl

Installing collected packages: filelock

Successfully installed filelock-3.13.1

(pytorch) D:\pytorch_packages>

2、安装 typing_extensions

(pytorch) D:\pytorch_packages>pip install typing_extensions-4.10.0-py3-none-any.whl

Processing d:\pytorch_packages\typing_extensions-4.10.0-py3-none-any.whl

Installing collected packages: typing-extensions

Successfully installed typing-extensions-4.10.0

3、安装mpmath

(pytorch) D:\pytorch_packages>pip install mpmath-1.3.0-py3-none-any.whl

Processing d:\pytorch_packages\mpmath-1.3.0-py3-none-any.whl

Installing collected packages: mpmath

Successfully installed mpmath-1.3.0

4、 安装sympy

(pytorch) D:\pytorch_packages>pip install sympy-1.12-py3-none-any.whl

Processing d:\pytorch_packages\sympy-1.12-py3-none-any.whl

Requirement already satisfied: mpmath>=0.19 in d:\anaconda3\envs\pytorch\lib\site-packages (from sympy==1.12) (1.3.0)

Installing collected packages: sympy

Successfully installed sympy-1.12

5、安装networkx

(pytorch) D:\pytorch_packages>pip install networkx-3.2.1-py3-none-any.whl

Processing d:\pytorch_packages\networkx-3.2.1-py3-none-any.whl

Installing collected packages: networkx

Successfully installed networkx-3.2.1

6、安装MarkupSafe

(pytorch) D:\pytorch_packages>pip install MarkupSafe-2.1.5-cp310-cp310-win_amd64.whl

Processing d:\pytorch_packages\markupsafe-2.1.5-cp310-cp310-win_amd64.whl

Installing collected packages: MarkupSafe

Successfully installed MarkupSafe-2.1.5

(pytorch) D:\pytorch_packages>

7、安装Jinja

(pytorch) D:\pytorch_packages>pip install Jinja2-3.1.3-py3-none-any.whl

Processing d:\pytorch_packages\jinja2-3.1.3-py3-none-any.whl

Requirement already satisfied: MarkupSafe>=2.0 in d:\anaconda3\envs\pytorch\lib\site-packages (from Jinja2==3.1.3) (2.1.5)

Installing collected packages: Jinja2

Successfully installed Jinja2-3.1.3

8、安装fsspec

(pytorch) D:\pytorch_packages>pip install fsspec-2024.3.1-py3-none-any.whl

Processing d:\pytorch_packages\fsspec-2024.3.1-py3-none-any.whl

Installing collected packages: fsspec

Successfully installed fsspec-2024.3.1

9、安装torch

(pytorch) D:\pytorch_packages>pip install torch-2.2.1-cp310-cp310-win_amd64.whl

Processing d:\pytorch_packages\torch-2.2.1-cp310-cp310-win_amd64.whl

Requirement already satisfied: filelock in d:\anaconda3\envs\pytorch\lib\site-packages (from torch==2.2.1) (3.13.1)

Requirement already satisfied: typing-extensions>=4.8.0 in d:\anaconda3\envs\pytorch\lib\site-packages (from torch==2.2.1) (4.10.0)

Requirement already satisfied: sympy in d:\anaconda3\envs\pytorch\lib\site-packages (from torch==2.2.1) (1.12)

Requirement already satisfied: networkx in d:\anaconda3\envs\pytorch\lib\site-packages (from torch==2.2.1) (3.2.1)

Requirement already satisfied: jinja2 in d:\anaconda3\envs\pytorch\lib\site-packages (from torch==2.2.1) (3.1.3)

Requirement already satisfied: fsspec in d:\anaconda3\envs\pytorch\lib\site-packages (from torch==2.2.1) (2024.3.1)

Requirement already satisfied: MarkupSafe>=2.0 in d:\anaconda3\envs\pytorch\lib\site-packages (from jinja2->torch==2.2.1) (2.1.5)

Requirement already satisfied: mpmath>=0.19 in d:\anaconda3\envs\pytorch\lib\site-packages (from sympy->torch==2.2.1) (1.3.0)

Installing collected packages: torch

Successfully installed torch-2.2.1

(pytorch) D:\pytorch_packages>

第三部分：安装torchvision

1、安装依赖numpy-1.26.4-cp310-cp310-win_amd64.whl

 (pytorch) D:\pytorch_packages>pip install numpy-1.26.4-cp310-cp310-win_amd64.whl

Processing d:\pytorch_packages\numpy-1.26.4-cp310-cp310-win_amd64.whl

Installing collected packages: numpy

Successfully installed numpy-1.26.4

1. 安装依赖pillow-10.2.0-cp310-cp310-win_amd64.whl

(pytorch) D:\pytorch_packages>pip install pillow-10.2.0-cp310-cp310-win_amd64.whl

Processing d:\pytorch_packages\pillow-10.2.0-cp310-cp310-win_amd64.whl

Installing collected packages: pillow

Successfully installed pillow-10.2.0

1. 安装torchvision

(pytorch) D:\pytorch_packages>pip install torchvision-0.17.1-cp310-cp310-win_amd64.whl

Processing d:\pytorch_packages\torchvision-0.17.1-cp310-cp310-win_amd64.whl

Requirement already satisfied: numpy in d:\anaconda3\envs\pytorch\lib\site-packages (from torchvision==0.17.1) (1.26.4)

Requirement already satisfied: torch==2.2.1 in d:\anaconda3\envs\pytorch\lib\site-packages (from torchvision==0.17.1) (2.2.1)

Requirement already satisfied: pillow!=8.3.*,>=5.3.0 in d:\anaconda3\envs\pytorch\lib\site-packages (from torchvision==0.17.1) (10.2.0)

Requirement already satisfied: filelock in d:\anaconda3\envs\pytorch\lib\site-packages (from torch==2.2.1->torchvision==0.17.1) (3.13.1)

Requirement already satisfied: typing-extensions>=4.8.0 in d:\anaconda3\envs\pytorch\lib\site-packages (from torch==2.2.1->torchvision==0.17.1) (4.10.0)

Requirement already satisfied: sympy in d:\anaconda3\envs\pytorch\lib\site-packages (from torch==2.2.1->torchvision==0.17.1) (1.12)

Requirement already satisfied: networkx in d:\anaconda3\envs\pytorch\lib\site-packages (from torch==2.2.1->torchvision==0.17.1) (3.2.1)

Requirement already satisfied: jinja2 in d:\anaconda3\envs\pytorch\lib\site-packages (from torch==2.2.1->torchvision==0.17.1) (3.1.3)

Requirement already satisfied: fsspec in d:\anaconda3\envs\pytorch\lib\site-packages (from torch==2.2.1->torchvision==0.17.1) (2024.3.1)

Requirement already satisfied: MarkupSafe>=2.0 in d:\anaconda3\envs\pytorch\lib\site-packages (from jinja2->torch==2.2.1->torchvision==0.17.1) (2.1.5)

Requirement already satisfied: mpmath>=0.19 in d:\anaconda3\envs\pytorch\lib\site-packages (from sympy->torch==2.2.1->torchvision==0.17.1) (1.3.0)

Installing collected packages: torchvision

Successfully installed torchvision-0.17.1

第四部分：安装torchaudio-2.2.1-cp310-cp310-win_amd64.whl

1. 直接安装torchaudio

(pytorch) D:\pytorch_packages>pip install torchaudio-2.2.1-cp310-cp310-win_amd64.whl

Processing d:\pytorch_packages\torchaudio-2.2.1-cp310-cp310-win_amd64.whl

Requirement already satisfied: torch==2.2.1 in d:\anaconda3\envs\pytorch\lib\site-packages (from torchaudio==2.2.1) (2.2.1)

Requirement already satisfied: filelock in d:\anaconda3\envs\pytorch\lib\site-packages (from torch==2.2.1->torchaudio==2.2.1) (3.13.1)

Requirement already satisfied: typing-extensions>=4.8.0 in d:\anaconda3\envs\pytorch\lib\site-packages (from torch==2.2.1->torchaudio==2.2.1) (4.10.0)

Requirement already satisfied: sympy in d:\anaconda3\envs\pytorch\lib\site-packages (from torch==2.2.1->torchaudio==2.2.1) (1.12)

Requirement already satisfied: networkx in d:\anaconda3\envs\pytorch\lib\site-packages (from torch==2.2.1->torchaudio==2.2.1) (3.2.1)

Requirement already satisfied: jinja2 in d:\anaconda3\envs\pytorch\lib\site-packages (from torch==2.2.1->torchaudio==2.2.1) (3.1.3)

Requirement already satisfied: fsspec in d:\anaconda3\envs\pytorch\lib\site-packages (from torch==2.2.1->torchaudio==2.2.1) (2024.3.1)

Requirement already satisfied: MarkupSafe>=2.0 in d:\anaconda3\envs\pytorch\lib\site-packages (from jinja2->torch==2.2.1->torchaudio==2.2.1) (2.1.5)

Requirement already satisfied: mpmath>=0.19 in d:\anaconda3\envs\pytorch\lib\site-packages (from sympy->torch==2.2.1->torchaudio==2.2.1) (1.3.0)

Installing collected packages: torchaudio

Successfully installed torchaudio-2.2.1

第五部分：验证Pytorch是否安装成功

1、输入python查看版本

(pytorch) D:\pytorch_packages>python

Python 3.10.14 | packaged by Anaconda, Inc. | (main, Mar 21 2024, 16:20:14) [MSC v.1916 64 bit (AMD64)] on win32

Type "help", "copyright", "credits" or "license" for more information.

>>>

2、导入torche,输入代码定义一个张量并输出张量的维度及张量的类型，如果现实如下结果表示环境验证成功。

(pytorch) D:\pytorch_packages>python

Python 3.10.14 | packaged by Anaconda, Inc. | (main, Mar 21 2024, 16:20:14) [MSC v.1916 64 bit (AMD64)] on win32

Type "help", "copyright", "credits" or "license" for more information.

>>> import torch

>>> t=torch.tensor([[1,2,3],[4,5,6]])

>>> print(t,t.shape,t.dtype)

tensor([[1, 2, 3],

        [4, 5, 6]]) torch.Size([2, 3]) torch.int64