// the find
mrdbourke/learn-huggingface
Repo designed to help learn the Hugging Face ecosystem (transformers, datasets, accelerate + more).
A personal, hands-on companion repo to the Hugging Face docs, walking through full end-to-end projects (text classification, object detection, LLM/VLM fine-tuning, multimodal RAG) with matching datasets, trained models, and live demos on the Hub. Built for beginners who already have a few months of Python and one ML/DL course under their belt, not for people who've already shipped fine-tuning pipelines before.
Every project is actually finished end-to-end — dataset, trained model, and a working Spaces demo are all linked and live, not just notebook code that references a model that may or may not exist. The data→model→demo structure is consistent across every example, so once you've done one you know how to navigate the rest. It covers real breadth of the current HF stack (transformers, PEFT, accelerate, timm, Spaces, Evaluate) across both text and vision/multimodal tasks, including more current material like VLM fine-tuning and multimodal RAG, not just the standard text-classification tutorial everyone writes.
It's a companion to a paid video course (ZTM) — several entries point you to a paid platform for the actual walkthrough, so the free repo alone is thinner than it looks. Content is uneven: notebooks are added over roughly a year and a half with gaps of months, so some (RAG, VLM) are newer and less battle-tested than the original text classification one. No tests, no CI beyond notebook-to-markdown sync, and the TODO list in the README is still half-unchecked, so don't expect a maintained library-grade reference, just a stack of personal tutorial notebooks. Being Jupyter-notebook-only means no reusable Python package to pip install — you're copy-pasting cells, not importing a module.