Understand AI.
Build it yourself.
Discover how language models work. Short lessons, real exercises and a clear path from your first token to your own model.
- sections
- lessons
- exercises
- learning rooms
- German & Englishcourse languages
Built for keeping at it.
A daily routine
Your daily goal: one lesson. Your streak counts the days you learn in a row, XP your progress.
Find your first stepUnderstand your next step
Try it, get feedback and understand why. Exercises check your answers. Open tasks such as explanations are self-assessed.
Try an exampleA real project for your computer
With Plus you download tiny-gpt-cpu: a Python project in 10 steps from tensors to a small GPT, checked with local CPU tests. Tested on macOS with Apple Silicon.
Explore Plus
See what happens inside.
A sentence becomes tokens. Attention connects them. Follow each step and try the interactive example right here.
From a sentence to the next token.
Sample input
KI lernt aus Daten.
One sentence. Three steps. Explore the building blocks behind it.
„KI lernt aus Daten.“
- kiID 1
- lerntID 2
- ausID 3
- datenID 4
- .ID 5
The tokenizer splits the German sample into tokens and gives each one an ID.
What you learn here
Text becomes computable.
A language model works with numbers. Token IDs connect pieces of text to the vocabulary.
Each idea builds on the last.
10 sections250 lessons
Free: sections 1–2With Plus: sections 3–10
Language model foundations
FreeCapabilities, limitations and transformer architecture
7 units, 21 lessons
PyTorch foundations
FreeTensors, automatic differentiation and training loops
9 units, 27 lessons
Learning room: A model learns Free, 18 minWorking with text data
PlusTokenization, token IDs and embeddings
8 units, 24 lessons
Learning room: Text becomes tokens Plus, 12 minImplementing attention
PlusSelf-attention, causal masks and multiple heads
11 units, 33 lessons
Learning room: Who can see whom? Plus, 15 minBuilding a GPT model
PlusTransformer blocks, normalization and text generation
7 units, 21 lessons
Pretraining on unlabeled data
PlusLoss, evaluation and the training loop
9 units, 27 lessons
Fine-tuning for classification
PlusPrepare labeled examples and adapt the model
8 units, 24 lessons
Fine-tuning for instructions
PlusInstruction datasets, response generation and evaluation
9 units, 27 lessons
Efficient training and PEFT
PlusMemory budgets, quantization and efficient adaptation
10 units, 30 lessons
Learning room: Will the model fit? Plus, 15 minCapstone: project plan and local checks
PlusPlan your project and verify it in your own environment
8 units, 16 lessons
Try each step for yourself.
Split text into tokens, build an attention mask, train a small set of weights or plan a memory budget. The 4 learning rooms turn course concepts into focused experiments.
These experiments run in your browser. For your own Python and PyTorch projects, you need a suitable local environment or notebook.
Tokenization
Text becomes tokens
Split text, trace token IDs, and repair a small vocabulary.
12 minPlus
Causal attention
Who can see whom?
Build an attention mask and stop the model from seeing the future.
15 minPlus
Model memory
Will the model fit?
Plan weights, runtime reserve, and context within a fixed memory budget.
15 minPlus
Training & evaluation
A model learns
Train real weights in your browser and learn what a metric actually proves.
18 minFree
Before you begin.
What experience do I need?
You can start without prior knowledge of LLMs. Basic Python helps with the later coding exercises. The PyTorch section introduces tensors, autograd and training loops.
Which languages can I learn in?
Choose English or German for your learning session. Keep quoted text and specified code values exactly as shown in an exercise, even when those values are in another language.
Can I run full PyTorch training here?
The course teaches Python and PyTorch through explanations and exercises. Selected JavaScript exercises and small experiments run in your browser. Full PyTorch training runs in your own environment.
What can I learn for free?
Free includes LLM and PyTorch foundations: 16 units and 48 lessons with their guides, plus the learning room “A model learns”. Plus opens the full path and all 4 rooms. View plans and availability
Start with the foundations.
The LLM and PyTorch foundations are free: 16 units and 48 lessons.