🧠 Deep Learning from Scratch

Minimal PyTorch implementations of core deep learning components, built from scratch to understand how they work.

📓 Notebooks

• llm.ipynb
  A minimal decoder-only Transformer for language modeling, with code for architecture, training, and generation.

• adamw.ipynb
  A from-scratch implementation of the AdamW optimizer, comparing it against L2-regularized Adam on a toy regression task.

💡 Tip: Run these notebooks directly in Google Colab — no setup required.