huggingface/course: The Free Hugging Face Course on Transformers and NLP

What This Repository Is and How It Relates to the Course Website

The huggingface/course repository holds the MDX content files that are used to build the Hugging Face course website at huggingface.co/course. The repository itself is not a Python package and does not run locally as an application. Its purpose is to store the text and code samples that appear on the course website.

The README describes the course as teaching developers how to apply Transformers to various tasks in natural language processing and beyond, and how to use the Hugging Face ecosystem: the Transformers library, the Datasets library, the Tokenizers library, the Accelerate library, and the Hugging Face Hub. The course is free and open-source under the Apache-2.0 license.

Developers who want to read the course should go to huggingface.co/course, not to the GitHub repository. The GitHub repository is the source of truth for translators, contributors fixing errors, and anyone who wants to run the content formatting tools locally.

Repository Structure: Chapters, Languages, and Subtitles

The top-level directory structure is: chapters/ for course content, subtitles/ for video subtitles, and utils/ for internal tooling. The chapters/ directory is subdivided by language code: chapters/en/ contains the English text, chapters/fr/ the French text, chapters/de/ the German work-in-progress, chapters/es/ the Spanish work-in-progress, and so on for roughly fifteen languages in total.

Within a language directory, each chapter is a directory of MDX files. MDX combines Markdown with JSX components, which allows the course pages to embed interactive widgets alongside prose and code examples.

The requirements.txt for development work contains three packages:

nbformat>=5.1.3
PyYAML>=5.4.1
black>=24.1.1

These are tooling dependencies for the code formatter (utils/code_formatter.py), not course dependencies. The course itself runs on the Hugging Face website infrastructure; readers do not install anything to access the content.

What the Course Covers

The README states the course covers applying Transformers to various tasks in NLP and beyond, using the Hugging Face ecosystem. Based on the repository structure and what the README describes, the core content includes working with pretrained models, fine-tuning Transformer models on specific tasks, sharing models and datasets on the Hugging Face Hub, building end-to-end pipelines, and using the tokenization and data processing APIs in the Hugging Face libraries.

The course teaches through practical examples. Code samples in the chapters are formatted and validated by the repository's code formatter script. The Makefile documents two targets:

make quality
make style

The quality target runs utils/code_formatter.py --check_only to verify that code blocks in the chapters are correctly formatted. The style target runs the formatter to apply fixes automatically. Contributors use these when submitting pull requests.

The subtitle files in subtitles/ correspond to video content that accompanies the text chapters, though the video hosting itself is external to the repository.

Language Availability and Translation Status

The README documents over fifteen language translations. Completed translations include English and French. Work-in-progress translations include German, Spanish, Persian, Bengali, Gujarati, Hebrew, Hindi, Bahasa Indonesia, Italian, Korean, Portuguese, Romanian, Russian, Turkish, Chinese (Simplified), and Japanese, among others.

Each translation is maintained in its own subdirectory within chapters/. The README lists the contributors for each language alongside the language entry. Translation completeness varies: the English and French versions are the most complete, while most other languages cover a subset of chapters.

The repository uses GitHub Actions for quality checks on contributions. Translators working on a new language or updating existing translations follow the same workflow as any other contributor: fork the repository, edit the relevant files in chapters/[language-code]/, and open a pull request.

When This Course Is Not the Right Resource

The Hugging Face course focuses on applying pretrained Transformer models and using the Hugging Face ecosystem. It does not cover: training models from scratch, deep reinforcement learning, deploying models to production servers at scale, MLOps workflows, or infrastructure for serving models in production. Teams who need guidance on those topics would need to look at Hugging Face's separate documentation, cloud provider tutorials, or specialized resources for each concern.

The course is structured for learners who are comfortable with Python but may not have prior deep learning experience. Practitioners who already understand backpropagation, attention mechanisms, and the Transformer architecture and want to focus on advanced fine-tuning techniques may find the early chapters too introductory. The course also does not address model quantization, pruning, or efficiency techniques for edge deployment.

The repository also does not include a local site-building workflow for readers. The content is MDX, which requires a JavaScript build system to render as HTML. Readers who want to read the course locally should check whether the Hugging Face website allows downloading content, or simply use the online version at huggingface.co/course.

Alternative: fast.ai Practical Deep Learning and Maintenance Status

The fast.ai Practical Deep Learning course is a comparable free resource. It differs in approach: fast.ai starts from practical results and working code, then works toward theory, and uses the FastAI library rather than the Hugging Face ecosystem. fast.ai also covers computer vision more deeply in its early chapters. For a developer specifically interested in the Hugging Face library APIs and the Hub, the Hugging Face course maps more directly to those tools. For a developer who wants to understand the theory alongside the practice and is comfortable with a different library, fast.ai is a reasonable alternative.

The last push to huggingface/course was on 2026-09-23, indicating the content is actively updated. The repository has no GitHub releases. Apache-2.0 is a permissive license that permits reuse and modification with attribution. Because the course content is Apache-2.0 licensed, educators who want to use sections of it in their own teaching materials or courses may do so with attribution, though the code samples in the chapters have their own sources and readers should verify the license of each model and dataset the chapters reference on the Hugging Face Hub.

Contributing to the repository follows a standard GitHub workflow. The Makefile provides make quality and make style as the two primary local checks a contributor runs before opening a pull request. The code formatter script at utils/code_formatter.py validates and reformats Python code blocks embedded in the MDX chapter files, so that all code examples in the course follow a consistent style. This matters for a course that readers use as reference material: inconsistent formatting in code samples would create confusion about syntax.

Editorial conclusion

The Hugging Face course is a good fit for a developer who wants a structured, free introduction to applying Transformer models and the Hugging Face ecosystem: Transformers, Datasets, Tokenizers, Accelerate, and the Hub. It is less appropriate for a developer who wants deployment infrastructure, production MLOps, or deep coverage of computer vision and audio beyond NLP. The course is read at huggingface.co/course; this repository holds the source content that builds the site. Developers who want to contribute a translation or fix an error should work from chapters/ in this repository. The last push was on 2026-09-23.

Frequently asked questions

Official sources

1. huggingface/course on GitHub
2. Issues
3. License: Apache-2.0
4. Project website
5. README