Back to projects

PyTorch Deep Learning Lab

40+ experiments that build intuition for autograd, optimization, and model behavior.

PythonPyTorchNumPyMatplotlib
PyTorch Deep Learning Lab
A lab-style repo designed for learning by experiment: small, controlled setups, consistent diagnostics, and clear plots.

Architecture overview

Notebook-driven lab modules: experiment → instrumentation (hooks/plots) → interpretation. Focused on gradient flow, autograd mechanics, and minimal “magic.”

Key technical decisions

  • PyTorch for transparency and easy debugging.
  • Matplotlib for direct control over diagnostics/plots.
  • Small, reproducible notebooks to keep concepts isolated.

Results / metrics

40+ experiments paired with 32+ long-form articles in “PyTorch Zero to One,” so readers can reproduce and extend each idea.

What I learned

How to design experiments that teach intuition (not just API usage), and how to build lightweight tooling that makes training dynamics visible.

GitHubBlog series
KRT