PyTorch Tutorial Series

Overview

• Introduction to PyTorch
• Popular machine learning and deep learning framework
• Focus on practical applications
• Aim to cover the necessary basics

Installation of PyTorch

1. Official Website

   • Go to pytorch.org
   • Click on "Get started"
   • Select the newest PyTorch build (currently version 1.3)

2. Select Operating System

   • Choose your OS (example: Mac)
   • Recommended package manager: Anaconda
   • If Anaconda is not installed, watch the tutorial provided in the description

3. Select Python Version

   • Use the newest Python version (example: Python 3.7)
   • Note: On Mac, only CPU version is available

4. GPU Support (Linux/Windows)

   • Install the CUDA toolkit for GPU support
   • Requires an NVIDIA GPU
   • Visit NVIDIA CUDA Downloads
     • Important: Latest supported CUDA version by PyTorch is CUDA 10.1
     • For latest CUDA version, go to legacy releases and download CUDA 10.1
     • Follow instructions for installation

Creating a Virtual Environment

• Open terminal
• Create a conda environment with: conda create -n pytorch python=3.7
• Activate the environment: conda activate pytorch

Installing PyTorch

• Copy the installation command from the PyTorch site
• Paste the command in the terminal
• Wait for the installation to complete

Verification

1. Start Python in the activated environment: python
2. Import the torch module: import torch
   • If installation is correct, no error should occur
3. Create a tensor: x = torch.rand(3) print(x)
4. Check if CUDA is available: torch.cuda.is_available()
   • Returns false on non-GPU setups; should return true if CUDA toolkit and GPU support are correctly installed

Conclusion

• PyTorch installation is complete
• Ready to begin working with PyTorch in upcoming tutorials
• Encouragement to enjoy the learning process
• See you in the next tutorial!