Deep learning notes that outrun the textbook release cycle

What it does This repository is a curated, continuously updated study guide covering deep learning from PyTorch fundamentals through diffusion models, CLIP, and SAM. The author maintains it in Quarto Markdown—treating lecture notes like source code—so the content stays version-controlled and publishes to a static site or exports to Jupyter Notebooks. It is explicitly positioned as a practical successor to introductory texts like Dive into Deep Learning, updated for the Transformer era.

The interesting bit Rather than collecting disconnected blog posts, the author treats the repo as a single coherent curriculum with custom utility code (dnnlpy), mathematical derivations, and engineering notes in one place. The Quarto-first workflow is a pragmatic choice: plain-text source files render to a website, but also convert cleanly to Colab-ready notebooks via a sync’d companion repository.

Key highlights

• Covers territory often missing from older textbooks: Attention, LoRA, Stable Diffusion, vLLM, and multimodal models like CLIP.
• Ships with a custom dnnlpy library containing implementations and utilities used across the lessons.
• Dual-format publishing: Quarto source for the website, with a sync’d Jupyter Notebook repo for immediate Colab execution.
• Explicitly tested against Python 3.14, PyTorch 2.12, and Transformers v5.
• Includes practical workflow notes from data processing through training, inference, and deployment.

Caveats

• The author openly warns that some explanations may be imprecise or incomplete because the project is built while learning.
• Many examples depend on the dnnlpy library; the notes will not run in a vanilla environment without it.
• Uses Transformers v5, whose API differs significantly from the v4 examples still common in most online tutorials.

Verdict Worth bookmarking if you are self-studying modern deep learning and want a structured alternative to chasing scattered tutorials. Skip it if you need a peer-reviewed reference or a drop-in library.