Chapter 2: Installation & Setup

“A neural net’s journey begins with a single tensor.”

---

2.1 Installing PyTorch the Right Way

✅ Step 1: Visit the Official Installer Page

Go to: https://pytorch.org/get-started/locally

You’ll see a selector for:

• PyTorch Build (Stable / Preview)
• Your OS (Linux, Mac, Windows)
• Package Manager (pip, conda)
• Language (Python, C++)
• Compute Platform (CPU, CUDA 11.8, CUDA 12.x, etc.)

Recommendation

For most ML developers (especially if you’re using a GPU):

# For pip + CUDA 11.8
pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

If you don’t have a GPU, install CPU-only:

pip3 install torch torchvision torchaudio

⚠️ Heads-up: Make sure your Python version is 3.8–3.12. PyTorch isn’t too happy with older versions.

2.2 Virtual Environment Setup (Highly Recommended)

To avoid breaking other Python projects:

python -m venv torch_env
source torch_env/bin/activate  # On Windows: torch_env\Scripts\activate

Then install PyTorch inside that environment.

2.3 Verify the Installation

Let’s test it right away:

import torch
x = torch.rand(3, 3)
print("Tensor:\n", x)
print("Is CUDA available?", torch.cuda.is_available())

You should see a 3×3 matrix of random values and a True/False flag about CUDA.

2.4 Your First Tensor Playground

Let’s walk through a few core operations to prove it works — and start building intuition.

➤ Create a Tensor

a = torch.tensor([[1, 2], [3, 4]], dtype=torch.float32)

➤ Do Some Math

b = torch.ones_like(a)
c = a + b
print("Addition:\n", c)

➤ Matrix Multiplication

d = torch.matmul(a, c.T)  # Transpose and multiply
print("Matrix product:\n", d)

➤ Move to GPU (if available)

if torch.cuda.is_available():
    a = a.to("cuda")
    print("Tensor on GPU:", a)

2.5 CPU vs CUDA: Why It Matters

Operation	CPU	CUDA (GPU)
Matrix multiplication	Slower for large matrices	Highly optimized
Memory access	Direct system memory	VRAM on GPU
Use-case	Lightweight ML / debugging	Training large models

TL;DR: Use CUDA if available. It’s fast. Like, ridiculously fast.

2.6 Troubleshooting Tips

❌ “torch not found” Make sure:

• You activated the correct environment

• You installed it in the right Python version

❌ CUDA not available - Check your driver:

nvidia-smi

Make sure your CUDA toolkit matches the version you selected during install.

2.7 Quick Recap

By now, you should:

✅ Have PyTorch installed in a clean environment

✅ Know how to write and run basic tensor operations

✅ Understand the role of CUDA and how to check if it’s working