PyTorch 101: Zero to Hero

A beginner-to-advanced learning path for PyTorch — built as reference material, a technical portfolio, and a foundation for future articles.

Welcome! This article is designed to take you from "What even is a tensor?" to confidently building, training, and deploying deep learning models in PyTorch. Think of it less like documentation and more like a mentor sitting next to you — we'll always explain why a concept exists before showing you how to code it.

Each module is a self-contained, article-quality deep dive. Read them in order if you're new, or jump around if you're brushing up on a specific topic.

---

🚀 Quickstart

1. Install PyTorch

The safest way to install PyTorch is to grab the exact command from the official install selector, which tailors it to your OS and hardware. For a quick CPU-only start:

# Create an isolated environment (recommended)
python -m venv .venv
source .venv/bin/activate        # Windows: .venv\Scripts\activate

# Install PyTorch + companions
pip install torch torchvision torchaudio

2. Verify your installation

import torch

print(torch.__version__)                 # e.g. 2.3.0
print("CUDA available:", torch.cuda.is_available())   # NVIDIA GPU?
print("MPS available:", torch.backends.mps.is_available())  # Apple Silicon?

If either prints True, you have hardware acceleration available. If not, no worries — every example here runs fine on a CPU.

3. Start learning

Head into docs/01-tensors/ and work your way up. 🎓

---

🗺️ The Roadmap / Syllabus

The curriculum is split into three phases, moving from raw building blocks to full deployment.

Phase 1 — Fundamentals

The bedrock. Master these and everything else clicks into place.

#	Module	What you'll learn
01	Tensors	Creation, dtypes, slicing, reshaping, matrix math, device placement (CPU/GPU/MPS/CUDA)
02	Autograd	The computational graph, requires_grad, forward/backward passes, gradients, torch.no_grad()
03	Neural Networks	Custom architectures with nn.Module, the forward() method, Linear/ReLU/Dropout

Phase 2 — Data & Training

Turning models into things that actually learn.

#	Module	What you'll learn
04	Data Handling	torch.utils.data, custom Dataset classes, DataLoader, batching, shuffling, transforms
05	The Training Loop	Loss functions, optimizers (Adam, SGD), zeroing grads, backprop, train() vs eval()
06	Saving & Loading	state_dict best practices, checkpoints, resuming training

Phase 3 — Advanced Architectures & Ecosystem

Real-world models and the tools around them.

#	Module	What you'll learn
07	Computer Vision	CNNs, Conv2d, MaxPool2d, transfer learning
08	NLP & Transformers	Text data, embeddings, LSTMs, intro to nn.Transformer
09	Ecosystem: Lightning	Refactoring boilerplate loops into PyTorch Lightning
10	Deployment & Optimization	Exporting models, PyTorch 2.x torch.compile for speed

---

📁 Article Structure

Each module is a folder containing a short README.md index plus 2–4 sequentially numbered sub-module files that each go deep on one concept.

Every sub-module file follows the same predictable structure: The 'Why' → Core Concepts → Code in Action → Common Pitfalls → Further Reading & Watch List.

---

📚 Top-Tier Resources (used throughout)

These are the resources I reference again and again. Bookmark them.

• 📖 Deep Learning with PyTorch — the official Manning book (free PDF available).
• 🌐 Learn PyTorch for Deep Learning: Zero to Mastery — Daniel Bourke's superb web book.
• 🎓 Practical Deep Learning for Coders — the fast.ai course.
• ▶️ Neural Networks: Zero to Hero — Andrej Karpathy's legendary YouTube series.
• 🛠️ Official PyTorch Tutorials — straight from the source.

---

🤝 How to Use This Article

• Total beginner? Start at Module 01 and don't skip ahead — each module assumes the previous ones.
• Need a refresher? Jump straight to the module you need; each is self-contained.
• Building a portfolio? Each module doubles as a polished write-up you can point to.

Happy learning. Let's build something. 🧠