Hugging Face Learn

What is in the catalogue

The courses are written by Hugging Face staff and the community, and are free to read and run:

• LLM Course: the long-standing flagship (formerly the NLP course), teaching transformers, tokenizers, datasets, fine-tuning and sharing models with the Hugging Face libraries.
• Agents Course: building agents with smolagents, LangGraph and LlamaIndex, with a graded final project and free certificate.
• Context Course: context engineering for code agents.
• a smol course: a short course on post-training models.
• Diffusion Course: image generation with the Diffusers library.
• Deep RL Course: deep reinforcement learning from the basics upwards.
• Computer Vision and Audio courses: applying transformers and related models to images and sound.
• Robotics Course: building robots with the LeRobot library.
• ML for Games and ML for 3D: more specialist tracks.
• Open-Source AI Cookbook: standalone notebooks rather than a course, useful for specific recipes.

Most courses are a mix of written chapters and notebooks you can run on a free Colab GPU or locally.

Which track to start with

• You want to build with language models: the LLM Course, then the Agents Course.
• You already ship LLM features and use coding agents: the Context Course is short and immediately useful.
• You are curious about images: the Diffusion Course.
• You like maths and simulation: the Deep RL Course, which is one of the more complete free RL resources.

Before you start

You need comfortable Python. For the LLM and diffusion tracks, some familiarity with PyTorch helps but the courses explain what they use. A basic grasp of what a neural network does, for example from the 3Blue1Brown videos, makes the early chapters much smoother.

How to get value from it

The catalogue is broad enough to browse forever, which is the main way people waste it. Pick one track, finish it, and publish something small at the end: a fine-tuned model, a dataset or a Space demo on your own Hugging Face profile. A public artefact you built is worth more than several half-read courses, and it is something you can show when applying for roles.

Run every notebook cell yourself and change at least one thing per chapter, such as the dataset, the model size or a training setting, then see what breaks.

Common questions

Do the courses give certificates? Some do, such as the Agents Course, which offers free certificates for completing graded work. Check each course's introduction.

How long does a track take? It varies by course. Plan on several weeks of part-time study for a full track such as the LLM Course or Deep RL Course.

Do I need a GPU? For most chapters, the free tier of Google Colab or a Hugging Face Space is enough. Some fine-tuning chapters run slowly without a better GPU.

Which is the best first course? For most software engineers today, the LLM Course, because it underpins the agents, context and fine-tuning material.

Best for: developers moving into applied machine learning, especially anyone working with open models.