在windows中搭建dive into deep learning中的深度学习框架

概述

因为virtualbox中无法使用cuda，见use-host-cuda-from-virtualbox。所以我想到了windows下的wsl，尝试后发现可行，记录一下过程。
后面得知，wsl2对gpu的支持是20年引入的：
https://blogs.windows.com/windowsdeveloper/2020/06/17/gpu-acc...

步骤

建议按推荐步骤安装。我首次安装时，先安装了深度学习框架，后安装gpu驱动，会有segfragment fault。重新安装深度学习框架后问题解决。

1. 安装 invida 驱动，cuda

driver:
https://www.nvidia.com/download/index.aspx
cuda toolkit:
https://developer.nvidia.com/cuda-downloads
下载点击安装。

2. 安装 wsl

wsl --install
参考：
https://learn.microsoft.com/en-us/windows/wsl/install
可以安装一下windows terminal改善体验。
https://apps.microsoft.com/detail/9N0DX20HK701
默认是ubuntu22 jammy 修改成国内源：
https://www.linuxmi.com/ubuntu-22-04-apt-sources-list.html
注意不要在wsl的linux系统里安装步骤1的内容，gpu driver以及cuda。也不要安装后面的visual code。

3. 安装深度学习框架

按dive in to deep learning 中的步骤执行如下：

sh Miniconda3-py39_4.12.0-Linux-x86_64.sh -b
~/miniconda3/bin/conda init
conda create --name d2l python=3.9 -y
conda activate d2l

https://zh.d2l.ai/chapter_installation/index.html
至此环境已经搭建完毕：

(d2l) dingxinglong@DESKTOP-PUDTV3D:~/git$ nvidia-smi
Sun Nov  5 22:41:44 2023
+---------------------------------------------------------------------------------------+
| NVIDIA-SMI 545.29.01              Driver Version: 546.01       CUDA Version: 12.3     |
|-----------------------------------------+----------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |         Memory-Usage | GPU-Util  Compute M. |
|                                         |                      |               MIG M. |
|=========================================+======================+======================|
|   0  NVIDIA GeForce GTX 1650        On  | 00000000:01:00.0 Off |                  N/A |
| 24%   34C    P8              N/A /  75W |     86MiB /  4096MiB |      0%      Default |
|                                         |                      |                  N/A |
+-----------------------------------------+----------------------+----------------------+

+---------------------------------------------------------------------------------------+
| Processes:                                                                            |
|  GPU   GI   CI        PID   Type   Process name                            GPU Memory |
|        ID   ID                                                             Usage      |
|=======================================================================================|
|  No running processes found                                                           |
+---------------------------------------------------------------------------------------+

4. 安装 visual code (可选)

注意是在windows中安装，如果在wsl的linux系统中安装过，先用下面命令删除

sudo apt purge code
sudo apt autoremove

1. 在windows中安装visual code。
2. 执行下面命令行

sudo apt-get update
sudo apt-get install wget ca-certificates

3. 在wsl terminal 中启动

code .

https://learn.microsoft.com/en-us/windows/wsl/tutorials/wsl-v...

Known Limitations

见：https://docs.nvidia.com/cuda/wsl-user-guide/index.html#gettin...

• Maxwell GPU is not supported.
• Unified Memory - Full Managed Memory Support is not available on Windows native and therefore WSL 2 will not support it for the foreseeable future.
• Pinned system memory (example: System memory that an application makes resident for GPU accesses) availability for applications is limited.
• Root user on bare metal (not containers) will not find nvidia-smi at the expected location.

• With the NVIDIA Container Toolkit for Docker 19.03, only --gpus all is supported.
  可以看到，对gpu型号的支持不完整，功能不完整，内存也不完整。对于学习没太大问题。wsl的定位就是本地开发。

  参考

  https://zh.d2l.ai/chapter_installation/index.html
  https://superuser.com/questions/1388815/use-host-cuda-from-vi...
  https://learn.microsoft.com/en-us/windows/wsl/install
  https://learn.microsoft.com/en-us/windows/ai/directml/gpu-ten...
  https://docs.nvidia.com/cuda/wsl-user-guide/index.html#gettin...
  https://learn.microsoft.com/en-us/windows/wsl/tutorials/wsl-v...
  https://blogs.windows.com/windowsdeveloper/2020/06/17/gpu-acc...