LLM courses: learn large language models from the basics to RAG and agents
Academl's LLM academy is a path through 15 online courses: how a large language model works, prompting, Hugging Face transformers, open-source and local models with Ollama, embeddings and RAG, tool-using agents, fine-tuning and production. 111 lessons in total, about 52 hours. The first two lessons of every course are free.
The LLM learning path: 7 stages
Stages go from simple to advanced. The first two are for beginners and need no coding. From stage 3 the lessons include Python code.
Stage 1. How an LLM works5 h
Tokens and context, why models hallucinate and how to choose a model for the task.
Stage 2. Prompting4 h
Prompt anatomy, zero-shot and few-shot, chain-of-thought, ReAct, structured output and defending against injection.
Stage 3. Transformers in practice6 h
The Hugging Face Pipeline API, BPE, WordPiece and SentencePiece tokenizers, fine-tuning with Trainer and LoRA, demos on Spaces.
Stage 4. Open-source and local models2 h
A map of open-weight models (DeepSeek, Qwen, Mistral), running them with Ollama and LM Studio, hardware choices, a local RAG stack.
Where to start for your task
Take only the stages you need: find your task and start at the stage it points to.
How the courses work
Every lesson has the same shape: short theory, a ready prompt or code, a hands-on task and a quiz. AI checks the task and suggests what to improve. You can ask the AI tutor about the lesson right on the page. Progress saves on its own, and every course ends with a certificate you can verify by code.
How long it takes to learn LLMs
The full path is about 52 hours by lesson estimates. At 30 minutes a day, five days a week, that is usually about 21 weeks, and you set the pace: stages can be taken selectively. The basics (stages 1–2) take about 9 hours.
| Stage | Courses | Time |
|---|---|---|
| 01 · How an LLM works | 2 | 5 h |
| 02 · Prompting | 1 | 4 h |
| 03 · Transformers in practice | 1 | 6 h |
| 04 · Open-source and local models | 2 | 2 h |
| 05 · Embeddings and RAG | 4 | 13 h |
| 06 · Agents and tools | 2 | 10 h |
| 07 · Fine-tuning and production | 3 | 12 h |
| Total | 15 | ~52 h |
Pricing
You can start for free: the first two lessons of every course are always open. Sign-up gives 7 days of full access, no card needed. Pro opens every lesson and certificate for $9 a month or $79 a year.
LLM courses: FAQ
What is an LLM and what do LLM courses teach?
An LLM (large language model) is the kind of model behind ChatGPT, Claude, Gemini, Llama and Qwen. Academl's LLM courses teach you to use these models and build on them: write prompts, run open models, connect your documents through RAG, give the model tools, fine-tune it and evaluate quality.
Where should a beginner start?
At stage 1: two courses on how language models work and on generative AI basics explain tokens, context, hallucinations and how to choose a model, with no math. Then stage 2, the prompt engineering guide. The courses of both stages are marked Beginner in the catalog.
Do I need to know how to code?
Not for stages 1–2. From stage 3 on (Hugging Face, embeddings, RAG, agents, fine-tuning) the lessons include Python code, and basic Python makes the practice much easier.
Can I learn LLMs for free?
Partly. The first two lessons of every course are always open, and sign-up gives 7 days of access to every lesson with no card. After that, Pro opens all lessons and certificates for $9 a month or $79 a year.
Is there a course on local models and Ollama?
Yes, that is stage 4. The local LLM course covers running a model on your own computer with Ollama and LM Studio, choosing hardware and a local RAG stack. The DeepSeek, Qwen and Mistral course helps you pick an open model, and 'LLM Engineer' at stage 7 goes from Ollama to production with vLLM.
What is the difference between RAG and fine-tuning?
RAG adds retrieved passages from your documents to the prompt so the model answers from them. Fine-tuning changes the model's weights for a format or style. For answers from documents you usually start with RAG (stage 5) and move to fine-tuning when you need a stable format (stage 7).
Related
Start with stage one today
Sign up with email or Google and get 7 days of full access to every course on the path, no card needed.
Sign up for free →