PyTorch installation and environmentconfiguration

PyTorch installation and environmentconfiguration

1. installation before 准备

in installationPyTorch之 before , 我们需要确保system已经installation了Pythonenvironment. PyTorchsupportPython 3.6及以 on version, 推荐usingPython 3.8 or 更 high version.

1.1 checkPythonversion

首先, 我们需要checksystemin is 否已installationPython以及Python version:

python --version
#  or 者
python3 --version

1.2 installationpip

pip is Python packagemanagementtool, 我们需要using它来installationPyTorch. such as果systemin还没 has installationpip, 可以按照以 under 步骤installation:

# Windowssystem
python -m ensurepip --upgrade

# macOS and Linuxsystem
sudo apt install python3-pip  # Ubuntu/Debian
brew install python3-pip      # macOS (usingHomebrew)

2. installationPyTorch

PyTorchproviding了 many 种installation方式, includingusingpip, condaetc.packagemanagementtool. 我们可以根据自己 operationsystem and 硬件environment选择合适 installation方式.

2.1 usingpipinstallation

usingpipinstallation is 最common 方式, 我们可以根据 is 否需要CUDAsupport选择不同 installationcommands:

2.1.1 installationCPUversion (不supportGPU加速)

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu

2.1.2 installationCUDAversion (supportGPU加速)

such as果你 system has NVIDIA GPU, 并且installation了CUDA, 可以选择installationCUDAversion以获得GPU加速:

# CUDA 11.8
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

# CUDA 12.1
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121

2.2 usingcondainstallation

such as果你usingAnaconda or Miniconda, 可以usingcondacommandsinstallationPyTorch:

2.2.1 installationCPUversion

conda install pytorch torchvision torchaudio cpuonly -c pytorch

2.2.2 installationCUDAversion

# CUDA 11.8
conda install pytorch torchvision torchaudio cudatoolkit=11.8 -c pytorch -c nvidia

# CUDA 12.1
conda install pytorch torchvision torchaudio cudatoolkit=12.1 -c pytorch -c nvidia

3. verificationinstallation

installationcompletion after , 我们可以throughrun一个 simple Python脚本来verificationPyTorch is 否installation成功:

import torch
import torchvision
import torchaudio

print("PyTorch version:", torch.__version__)
print("TorchVision version:", torchvision.__version__)
print("TorchAudio version:", torchaudio.__version__)
print("CUDA available:", torch.cuda.is_available())

# such as果CUDA可用, 打印CUDAversion
if torch.cuda.is_available():
    print("CUDA version:", torch.version.cuda)
    print("Number of GPUs:", torch.cuda.device_count())
    print("Current GPU:", torch.cuda.current_device())
    print("GPU name:", torch.cuda.get_device_name(0))

4. creation虚拟environment

for 了避免依赖conflict, 推荐 in 虚拟environmentininstallationPyTorch. 我们可以usingvenv or condacreation虚拟environment.

4.1 usingvenvcreation虚拟environment

# creation虚拟environment
python -m venv pytorch-env

# 激活虚拟environment
# Windows
pytorch-env\Scripts\activate

# macOS and Linux
source pytorch-env/bin/activate

#  in 虚拟environmentininstallationPyTorch
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu

4.2 usingcondacreation虚拟environment

# creation虚拟environment
conda create -n pytorch-env python=3.8

# 激活虚拟environment
conda activate pytorch-env

#  in 虚拟environmentininstallationPyTorch
conda install pytorch torchvision torchaudio cpuonly -c pytorch

5. commoninstallationissues及solution

5.1 CUDAversion不匹配

5.2 memory不足

pip install --no-cache-dir torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu

5.3 networkissues

# using国 in 镜像sources
pip install torch torchvision torchaudio -i https://pypi.tuna.tsinghua.edu.cn/simple

6. 推荐 Developmentenvironment

除了installationPyTorch本身, 我们还推荐installation以 under tool来提升Developmentefficiency:

6.1 Jupyter Notebook/Lab

Jupyter is a 交互式计算environment, 非常适合for深度Learning实验:

pip install jupyter jupyterlab

6.2 常用 Pythonlibrary

in 深度Learningprojectin, 我们经常会用 to 以 under library:

pip install numpy pandas matplotlib scikit-learn tqdm

# 生成requirements.txt
pip freeze > requirements.txt

#  from requirements.txtinstallation依赖
pip install -r requirements.txt

7. summarized

本tutorial介绍了PyTorch installationmethod, including:

• usingpip and condainstallationPyTorch
• installationCPUversion and CUDAversion PyTorch
• creation and management虚拟environment
• 解决common installationissues
• 推荐 Developmentenvironment and tool

installationcompletion after , 你就可以开始usingPyTorchfor深度Learningproject Development了.