What it does This repo bundles lecture notes (PDFs) and Jupyter notebooks into a structured course on deep learning. Topics run from classic MLPs, CNNs, and RNNs up to modern additions like Mamba, SAM2, flow matching, diffusion models, and a from-scratch GPT-2 trained on TinyStories. A “Practice” section adds tooling coverage: PyTorch Lightning, ONNX export, TensorRT, Triton serving, and even a Docker guide.

The interesting bit Most tutorials pick a lane—math or code. This one pairs a PDF explainer with a runnable notebook for nearly every topic, so you can read the theory, then immediately break the implementation. The 2025 revision also keeps the content current; Mamba and flow matching sit alongside transformers and GANs rather than being scattered across blog posts.

Key highlights

• Theory + code pairs for 15+ topics, from autoencoders to LLM agents
• Includes newer architectures: Mamba, SAM2 segmentation, normalizing flows, diffusion/flow matching
• LLM section covers training GPT-2 from scratch, fine-tuning, and validation on TinyStories
• Practice modules on deployment: ONNX, TensorRT, PyTriton, Docker
• Citable academic-style notes with a BibTeX entry provided

Caveats

• Video links are listed but uniformly blank ("-") in the current README
• Some topics (Agents, Detection, Optimization, Regularization) have notes but no code yet
• “HuggingFcae” typo suggests light proofreading; external PDFs on Google Drive could drift or break

Verdict Ideal for self-taught developers or students who want a coherent, free curriculum with working PyTorch examples. Skip it if you need production-hardened libraries or video-first learning; the value here is the structured breadth, not depth on any single frontier model.