Large Language Model Course

How the course is structured

The repository is a structured reading and practice roadmap rather than a video course:

1. LLM Fundamentals (optional): the mathematics, Python and neural network basics you need before the rest.
2. The LLM Scientist: how the best models are built: architecture, pre-training, post-training datasets, supervised fine-tuning, preference alignment, evaluation, quantisation and newer trends.
3. The LLM Engineer: how to build products on top of models: running models through APIs and locally, vector storage, retrieval-augmented generation, advanced RAG, agents, inference optimisation, deployment and security.

Each topic has a short explanation and a list of the best articles, papers and tutorials to read on it.

The notebooks

A big part of the repository's value is its set of Colab notebooks, including:

• Fine-tuning: fine-tuning open models with Unsloth, preference tuning with DPO and QLoRA, and training with Axolotl.
• Quantisation: an introduction to 8-bit quantisation, GPTQ, GGUF with llama.cpp, and ExLlamaV2.
• Other tools: merging models with MergeKit, creating mixture-of-experts models and knowledge-graph experiments.

They run on free Colab GPUs for smaller models. Some notebooks reference specific model versions that were current when written, so expect to swap in newer models.

Popularity and upkeep

It is one of the most starred AI learning repositories on GitHub, with about 83,000 stars and 9,700 forks as of September 2026. The last commit at that time was in February 2026. The roadmap's structure ages well, but check dates on individual links and notebooks, since the field moves quickly. The author also co-wrote a paid book, the LLM Engineer's Handbook, and states the course itself will always stay free.

Which track should you follow?

• You build applications with LLM APIs: go straight to the LLM Engineer track. Read the RAG, agents and deployment sections, and skim the scientist track for context.
• You want to fine-tune or research models: follow the LLM Scientist track and work through the fine-tuning and quantisation notebooks.
• You are new to machine learning: do the fundamentals first, alongside an introductory course. The rest will not make sense without them.

How long it takes

There is no fixed duration. Each track links to many hours of material. A realistic plan is one section a week, reading the top two or three linked resources and running any related notebook, which takes a couple of months per track. Trying to read every link is the main way people stall.

Common questions

Is it a course with lessons and quizzes? No. It is a curated roadmap with notebooks. You need to be self-directed.

Do I need a GPU? Not for most notebooks, which are built for free Colab. Larger models need paid GPUs.

How does it compare with Hugging Face's LLM Course? Hugging Face's course teaches through its libraries step by step; this roadmap is broader and more research-oriented. They complement each other.

Best for: developers who want a map of the whole LLM field, from how models are trained to how applications are deployed.