深度学习工程实现框架调研
郝伟 2022/08/11

深度学习主流框架：PyTorch VS TensorFlow

主要框架对比

两个最流行的框架 PyTorch 和 TensorFlow，在阅读了以下文章后：

可以得到初步的结论:

性能和功能两者相关不大
PyTorch 因为简单，逻辑清晰，易于上手等优势，在学术界的市场份额已经从2018年10%的提高到90%，
Tensorflow 复杂，功能相对强大，且易于扩展等优点所以更合适用于实际工程。

选择结论

所以作为当前深度学习从研究和人员配置情况来考虑，暂时先使用PyTorch进行相关研究认证，等对深度学习有足够了解后，根据实际项目需求再考虑迁移至 TensorFlow框架上。

安装

需要安装Anaconda https://www.anaconda.com/ （不是必需，但是比较易用）
教程 https://anaconda.cloud/getting-started
安装 CUDA https://blog.csdn.net/m0_45447650/article/details/123704930
需要安装以下两个组件，可以参考：非常详细的安装过程（真的非常详细）
安装指南 https://docs.nvidia.com/cuda/cuda-installation-guide-microsoft-windows/index.html#introduction

先安装 CUDA Toolkit: https://developer.nvidia.com/cuda-downloads?target_os=Windows&target_arch=x86_64&target_version=11&target_type=exe_local
下载文件： cuda_11.7.1_516.94_windows.exe
再安装cnDNN https://developer.nvidia.com/cudnn
下载文件： cudnn-windows-x86_64-8.5.0.96_cuda11-archive

安装 PyTorch

conda install pytorch torchvision torchaudio cudatoolkit=11.7 -c pytorch -c conda-forge

PS C:\Users\hao> conda install pytorch torchvision torchaudio cudatoolkit=11.7 -c pytorch -c conda-forge
Collecting package metadata (current_repodata.json): done
Solving environment: done


==> WARNING: A newer version of conda exists. <==
  current version: 4.12.0
  latest version: 4.13.0

Please update conda by running

    $ conda update -n base -c defaults conda


WARNING conda.gateways.disk.delete:unlink_or_rename_to_trash(143): Could not remove or rename C:\Users\hao\anaconda3\pkgs\pytorch-1.12.1-py3.9_cuda11.3_cudnn8_0.tar.bz2.  Please remove this file manually (you may need to reboot to free file handles)

## Package Plan ##

  environment location: C:\Users\hao\anaconda3

  added / updated specs:
    - cudatoolkit=11.7
    - pytorch
    - torchaudio
    - torchvision


The following packages will be downloaded:

    package                    |            build
    ---------------------------|-----------------
    conda-4.13.0               |   py39hcbf5309_1         1.0 MB  conda-forge
    cudatoolkit-11.7.0         |      ha6f8bbd_10       965.6 MB  conda-forge
    libuv-1.44.2               |       h8ffe710_0         362 KB  conda-forge
    python_abi-3.9             |           2_cp39           4 KB  conda-forge
    pytorch-1.12.1             |      py3.9_cpu_0       133.6 MB  pytorch
    pytorch-mutex-1.0          |              cpu           3 KB  pytorch
    torchaudio-0.12.1          |         py39_cpu         3.5 MB  pytorch
    torchvision-0.13.1         |         py39_cpu         6.2 MB  pytorch
    ------------------------------------------------------------
                                           Total:        1.08 GB

The following NEW packages will be INSTALLED:

  cudatoolkit        conda-forge/win-64::cudatoolkit-11.7.0-ha6f8bbd_10
  libuv              conda-forge/win-64::libuv-1.44.2-h8ffe710_0
  python_abi         conda-forge/win-64::python_abi-3.9-2_cp39
  pytorch            pytorch/win-64::pytorch-1.12.1-py3.9_cpu_0
  pytorch-mutex      pytorch/noarch::pytorch-mutex-1.0-cpu
  torchaudio         pytorch/win-64::torchaudio-0.12.1-py39_cpu
  torchvision        pytorch/win-64::torchvision-0.13.1-py39_cpu

The following packages will be UPDATED:

  conda              pkgs/main::conda-4.12.0-py39haa95532_0 --> conda-forge::conda-4.13.0-py39hcbf5309_1


Proceed ([y]/n)? y


Downloading and Extracting Packages
torchaudio-0.12.1    | 3.5 MB    | ############################################################################ | 100%
pytorch-mutex-1.0    | 3 KB      | ############################################################################ | 100%
python_abi-3.9       | 4 KB      | ############################################################################ | 100%
libuv-1.44.2         | 362 KB    | ############################################################################ | 100%
conda-4.13.0         | 1.0 MB    | ############################################################################ | 100%
cudatoolkit-11.7.0   | 965.6 MB  | ######                                                                    cudatoolkit-11.7.0   | 965.6 MB  | ######                                                            cudatoolkit-   8%
cudatoolkit-11.7.0   | 965.6 MB  | ########

2. PyTorch 学习网站

https://pytorch.org/tutorials/

Bokeh 官网 https://bokeh.org/
什么是 Bokeh https://blog.csdn.net/weixin_38037405/article/details/121498202

创建NVidia账号: hwaust@hotmail.com, me2022 (创建过程还需要填写资料和调研，好麻烦)

PS C:\Users\hao> nvidia-smi
Wed Aug 10 12:23:15 2022
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 511.81       Driver Version: 511.81       CUDA Version: 11.6     |
|-------------------------------+----------------------+----------------------+
| GPU  Name            TCC/WDDM | Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                               |                      |               MIG M. |
|===============================+======================+======================|
|   0  NVIDIA GeForce ... WDDM  | 00000000:01:00.0  On |                  N/A |
| N/A   44C    P8    14W /  N/A |   2280MiB /  6144MiB |      3%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
|        ID   ID                                                   Usage      |
|=============================================================================|
|    0   N/A  N/A      1820    C+G                                   N/A      |
|    0   N/A  N/A      1904    C+G   ...ekyb3d8bbwe\YourPhone.exe    N/A      |
|    0   N/A  N/A      3020    C+G   ...m\XMP\Program\APlayer.exe    N/A      |
|    0   N/A  N/A     13052    C+G   C:\Windows\explorer.exe         N/A      |
|    0   N/A  N/A     13460    C+G   ...icrosoft VS Code\Code.exe    N/A      |
|    0   N/A  N/A     14840    C+G   ...bbwe\Microsoft.Photos.exe    N/A      |
|    0   N/A  N/A     17136    C+G   ...8bbwe\WindowsTerminal.exe    N/A      |
|    0   N/A  N/A     19276    C+G   ...cal\Feishu\app\Feishu.exe    N/A      |
|    0   N/A  N/A     22280    C+G   ...n1h2txyewy\SearchHost.exe    N/A      |
|    0   N/A  N/A     26252    C+G   ...y\ShellExperienceHost.exe    N/A      |
|    0   N/A  N/A     27680    C+G   ...293.47\msedgewebview2.exe    N/A      |
|    0   N/A  N/A     28792    C+G   ...cw5n1h2txyewy\LockApp.exe    N/A      |
|    0   N/A  N/A     30728    C+G   ...in7x64\steamwebhelper.exe    N/A      |
|    0   N/A  N/A     35032    C+G   ...2txyewy\TextInputHost.exe    N/A      |
|    0   N/A  N/A     37956    C+G   ...tracted\WechatBrowser.exe    N/A      |
|    0   N/A  N/A     40208    C+G   ...ge\Application\msedge.exe    N/A      |
|    0   N/A  N/A     45860    C+G   ...artMenuExperienceHost.exe    N/A      |
|    0   N/A  N/A     46828    C+G   ...al\Programs\Veee\Veee.exe    N/A      |
|    0   N/A  N/A     53808    C+G   ...d\runtime\WeChatAppEx.exe    N/A      |
|    0   N/A  N/A     54168    C+G   ...n\Application\Maxthon.exe    N/A      |
|    0   N/A  N/A     58744    C+G   ...lPanel\SystemSettings.exe    N/A      |
+-----------------------------------------------------------------------------+
PS C:\Users\hao>

PS C:\Users\hao> conda install pytorch torchvision torchaudio cudatoolkit=11.7 -c pytorch -c conda-forge
Collecting package metadata (current_repodata.json): failed

CondaHTTPError: HTTP 000 CONNECTION FAILED for url <https://conda.anaconda.org/pytorch/win-64/current_repodata.json>
Elapsed: -

An HTTP error occurred when trying to retrieve this URL.
HTTP errors are often intermittent, and a simple retry will get you on your way.
'https://conda.anaconda.org/pytorch/win-64'

2022/08/12 完成成功！

PS C:\Users\hao> conda install pytorch torchvision torchaudio cudatoolkit=11.7 -c pytorch -c conda-forge
Collecting package metadata (current_repodata.json): done
Solving environment: done

## Package Plan ##

  environment location: C:\Users\hao\anaconda3

  added / updated specs:
    - cudatoolkit=11.7
    - pytorch
    - torchaudio
    - torchvision


The following packages will be downloaded:

    package                    |            build
    ---------------------------|-----------------
    pytorch-1.12.1             |      py3.9_cpu_0       133.6 MB  pytorch
    torchvision-0.13.1         |         py39_cpu         6.2 MB  pytorch
"By downloading and using the CUDA Toolkit conda packages, you accept the terms and conditions of the CUDA End User License Agreement (EULA): https://docs.nvidia.com/cuda/eula/index.html"

done
PS C:\Users\hao>

狂肝两万字带你用pytorch搞深度学习！！！