AI - Win10 機(jī)器學(xué)習(xí)環(huán)境 (TensorFlow GPU, JupyterLab, VSCode)

原文：https://makeoptim.com/deep-learning/tensorflow-gpu-on-win10

介紹
所需軟件
Nvidia GPU driver
CUDA ToolKit
cuDNN
環(huán)境變量
Miniconda
虛擬環(huán)境
安裝 TensorFlow
安裝 JupyterLab 和 matplotlib
安裝 pywin32
在 JupyterLab 中運(yùn)行 TensorFlow
安裝 VSCode
VSCode 運(yùn)行 TensorFlow
小結(jié)
延伸閱讀
參考鏈接

介紹

Win10
GTX 1050ti
tensorflow 2.4.1

所需軟件

NVIDIA? GPU drivers：CUDA? 11.0 需要 450.x 或更高版本辅鲸。
CUDA? Toolkit：TensorFlow 支持 CUDA? 11（TensorFlow 2.4.0 及更高版本）
CUDA? Toolkit附帶的CUPTI端壳。
cuDNN SDK 8.0.4 cuDNN 版本).
（可選）TensorRT 6.0割岛，可縮短用某些模型進(jìn)行推斷的延遲時(shí)間并提高吞吐量。

Nvidia GPU driver

前往 https://www.nvidia.com/download/index.aspx?lang=en-us 下載并安裝對應(yīng)的顯卡驅(qū)動(dòng)。

image

重啟并使用以下命令檢查 GPU 是否可見。

PS C:\Users\Administrator> nvidia-smi.exe
Wed Apr 28 22:28:23 2021
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 466.11       Driver Version: 466.11       CUDA Version: 11.3     |
|-------------------------------+----------------------+----------------------+
| 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 |
| 30%   34C    P0    N/A /  75W |    455MiB /  4096MiB |      1%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
|        ID   ID                                                   Usage      |
|=============================================================================|
|    0   N/A  N/A      1532    C+G   C:\Windows\System32\dwm.exe     N/A      |
|    0   N/A  N/A      2800    C+G   ...Central\Razer Central.exe    N/A      |
|    0   N/A  N/A      4576    C+G   C:\Windows\explorer.exe         N/A      |
|    0   N/A  N/A      5104    C+G   ...5n1h2txyewy\SearchApp.exe    N/A      |
|    0   N/A  N/A      6112    C+G   ...lPanel\SystemSettings.exe    N/A      |
|    0   N/A  N/A      7588    C+G   ...me\Application\chrome.exe    N/A      |
|    0   N/A  N/A      8568    C+G   ...arp.BrowserSubprocess.exe    N/A      |
|    0   N/A  N/A      9076    C+G   ...nputApp\TextInputHost.exe    N/A      |
|    0   N/A  N/A      9096    C+G   ... Host\Razer Synapse 3.exe    N/A      |
|    0   N/A  N/A     10152    C+G   ...y\ShellExperienceHost.exe    N/A      |
+-----------------------------------------------------------------------------+

CUDA ToolKit

前往 http://developer.nvidia.com/cuda-downloads 下載 CUDA Toolkit 11.3 并安裝。

image

cuDNN

前往 https://developer.nvidia.com/rdp/cudnn-archive 下載 cuDNN 并解壓柬帕。

image

將解壓后的 lib、bin、include陷寝，拷貝到 C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.1 對應(yīng)的目錄下锅很。

image

復(fù)制 C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.1\bin 下的 cusolver64_11.dll，重命名為 cusolver64_10.dll 盼铁。

image

環(huán)境變量

SET PATH=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.1\bin;%PATH%
SET PATH=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.1\extras\CUPTI\lib64;%PATH%
SET PATH=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.1\include;%PATH%
SET PATH=C:\tools\cuda\bin;%PATH%

Miniconda

前往 https://docs.conda.io/en/latest/miniconda.html 下載并安裝 Miniconda粗蔚。

image

虛擬環(huán)境

打開 Anaconda Powershell Prompt。

image

創(chuàng)建一個(gè)名稱為 tensorflow 的虛擬環(huán)境饶火。

$ conda create -n tensorflow python=3.8.5
$ conda activate tensorflow

注：如果使用 Power Sehll鹏控，需要執(zhí)行 conda init powershell 和 set-executionpolicy remotesigned, 否則會(huì)出現(xiàn)無法 activate 成功地問題，詳見 https://stackoverflow.com/questions/62082873/conda-not-activate-in-power-shell

PS C:\Users\Administrator> conda init powershell

PS C:\Users\Administrator> set-executionpolicy remotesigned

安裝 TensorFlow

(tensorflow) PS C:\Users\Administrator> pip install tensorflow==2.4.1

驗(yàn)證安裝

(tensorflow) PS C:\Users\Administrator> python -c "import tensorflow as tf;print('Num GPUs Available: ', len(tf.config.list_physical_devices('GPU')))"
2021-05-05 21:38:12.335369: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library cudart64_110.dll
2021-05-05 21:38:14.569229: I tensorflow/compiler/jit/xla_cpu_device.cc:41] Not creating XLA devices, tf_xla_enable_xla_devices not set
2021-05-05 21:38:14.571864: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library nvcuda.dll
2021-05-05 21:38:14.605855: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1720] Found device 0 with properties:
pciBusID: 0000:01:00.0 name: NVIDIA GeForce GTX 1050 Ti computeCapability: 6.1
coreClock: 1.392GHz coreCount: 6 deviceMemorySize: 4.00GiB deviceMemoryBandwidth: 104.43GiB/s
2021-05-05 21:38:14.606025: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library cudart64_110.dll
2021-05-05 21:38:14.682846: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library cublas64_11.dll
2021-05-05 21:38:14.682987: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library cublasLt64_11.dll
2021-05-05 21:38:14.721429: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library cufft64_10.dll
2021-05-05 21:38:14.732614: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library curand64_10.dll
2021-05-05 21:38:14.818913: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library cusolver64_10.dll
2021-05-05 21:38:14.842604: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library cusparse64_11.dll
2021-05-05 21:38:14.848834: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library cudnn64_8.dll
2021-05-05 21:38:14.849000: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1862] Adding visible gpu devices: 0
Num GPUs Available:  1

安裝 JupyterLab 和 matplotlib

(tensorflow) PS C:\Users\Administrator> pip install jupyterlab matplotlib

安裝 pywin32

(tensorflow) PS C:\Users\Administrator> cd 'C:\ProgramData\Miniconda3\Scripts\'
(tensorflow) PS C:\ProgramData\Miniconda3\Scripts> python pywin32_postinstall.py -install
Parsed arguments are: Namespace(destination='C:\\ProgramData\\Miniconda3\\envs\\tensorflow\\Lib\\site-packages', install=True, quiet=False, remove=False, silent=False, wait=None)
pywin32_postinstall.py:164: DeprecationWarning: the imp module is deprecated in favour of importlib; see the module's documentation for alternative uses
  import imp
Copied pythoncom38.dll to C:\WINDOWS\system32\pythoncom38.dll
Copied pywintypes38.dll to C:\WINDOWS\system32\pywintypes38.dll
Registered: Python.Interpreter
Registered: Python.Dictionary
Registered: Python
-> Software\Python\PythonCore\3.8\Help[None]=None
-> Software\Python\PythonCore\3.8\Help\Pythonwin Reference[None]='C:\\ProgramData\\Miniconda3\\envs\\tensorflow\\Lib\\site-packages\\PyWin32.chm'
Pythonwin has been registered in context menu
Creating directory C:\ProgramData\Miniconda3\envs\tensorflow\Lib\site-packages\win32com\gen_py
Can't install shortcuts - 'C:\\ProgramData\\Microsoft\\Windows\\Start Menu\\Programs\\Python 3.8' is not a folder
The pywin32 extensions were successfully installed.

在 JupyterLab 中運(yùn)行 TensorFlow

jupyter lab

JupyterLab 將自動(dòng)在瀏覽器打開肤寝。

從 https://www.tensorflow.org/tutorials/images/cnn 下載并導(dǎo)入 CNN notebook当辐。

image

執(zhí)行 Restart Kernel and Run All Cells。

image

當(dāng)訓(xùn)練開始, 檢查 GPU 進(jìn)程鲤看，可以看到 ...nvs\tensorflow\python.exe 表示正在使用 GPU 訓(xùn)練模型缘揪。

image

PS C:\Users\Administrator> nvidia-smi.exe
Wed May  5 21:47:50 2021
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 466.11       Driver Version: 466.11       CUDA Version: 11.3     |
|-------------------------------+----------------------+----------------------+
| 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 |
| 30%   39C    P0    N/A /  75W |   3695MiB /  4096MiB |     50%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
|        ID   ID                                                   Usage      |
|=============================================================================|
|    0   N/A  N/A      1008    C+G   C:\Windows\explorer.exe         N/A      |
|    0   N/A  N/A      3620    C+G   ...8bbwe\WindowsTerminal.exe    N/A      |
|    0   N/A  N/A      6936    C+G   ...nputApp\TextInputHost.exe    N/A      |
|    0   N/A  N/A      7024    C+G   ...5n1h2txyewy\SearchApp.exe    N/A      |
|    0   N/A  N/A      9648    C+G   ...me\Application\chrome.exe    N/A      |
|    0   N/A  N/A     12128    C+G   C:\Windows\System32\dwm.exe     N/A      |
|    0   N/A  N/A     12248    C+G   ...y\ShellExperienceHost.exe    N/A      |
|    0   N/A  N/A     14384      C   ...nvs\tensorflow\python.exe    N/A      |
+-----------------------------------------------------------------------------+

安裝 VSCode

前往官網(wǎng)下載并安裝 VSCode。

打開 VSCode 并安裝 Python 支持义桂。

image

選擇某個(gè)文件夾（這里以 ~/tensorflow-notebook/01-hello 為例）找筝，新建文件 hello.ipynb。

import tensorflow as tf
hello = tf.constant('Hello, TensorFlow!')
hello.numpy()

使用 VSCode 打開剛才創(chuàng)建的 ~/tensorflow-notebook/01-hello/hello.ipynb慷吊，并選擇 Python 為創(chuàng)建的虛擬環(huán)境袖裕。

image

VSCode 運(yùn)行 TensorFlow

image

小結(jié)

至此急鳄，開發(fā)環(huán)境已經(jīng)搭建完畢堰酿。大家可以根據(jù)自己的習(xí)慣疾宏，選擇使用命令行触创、JupyterLab 或者 VSCode 進(jìn)行開發(fā)。

參考鏈接

https://www.tensorflow.org/install
https://www.tensorflow.org/guide/gpu
https://docs.nvidia.com/cuda/cuda-installation-guide-microsoft-windows/
http://developer.nvidia.com/cuda-downloads

AI - Win10 機(jī)器學(xué)習(xí)環(huán)境 (TensorFlow GPU, JupyterLab, VSCode)

介紹

所需軟件

Nvidia GPU driver

CUDA ToolKit

cuDNN

環(huán)境變量

Miniconda

虛擬環(huán)境

安裝 TensorFlow

安裝 JupyterLab 和 matplotlib

安裝 pywin32

在 JupyterLab 中運(yùn)行 TensorFlow

安裝 VSCode

VSCode 運(yùn)行 TensorFlow

小結(jié)

延伸閱讀

參考鏈接