#14TageLLM
14 days, 3–4 short lessons a day.
The challenge takes you through the 48 free lessons of AI Steps: first how large language models are built, then PyTorch from tensors and autograd to a training loop. In the free learning room “A model learns”, you then train real weights yourself. Free, in your browser, without payment details.
- 14 days
- 48 free lessons
- 16 units
- 20–30 minutes a day
How it works
- Create a free account. You need no payment details.
- Each day, do the lessons from the plan below. One lesson takes 5–9 minutes, and your answers are checked right away.
- Start whenever you like: day 1 is the day you begin. Your streak shows whether you keep going.
Your plan, day by day
The days follow the order of the course. The minutes are the lesson lengths stated in the course.
Days 1–7: Language model foundations
7 units, 21 lessons
- Day 13 lessons, 22 min
What is a language model?Concept, Predict & trace, Build & verify
- Day 23 lessons, 23 min
Applications and limitationsConcept, Predict & trace, Build & verify
- Day 33 lessons, 23 min
Stages of building and using an LLMConcept, Predict & trace, Build & verify
- Day 43 lessons, 22 min
The Transformer architectureConcept, Predict & trace, Build & verify
- Day 53 lessons, 21 min
Large datasets and scalingConcept, Predict & trace, Build & verify
- Day 63 lessons, 22 min
GPT architecture at a glanceConcept, Predict & trace, Build & verify
- Day 73 lessons, 21 min
The complete build roadmapConcept, Predict & trace, Build & verify
Days 8–14: PyTorch foundations
9 units, 27 lessons
- Day 84 lessons, 24 min
Python readiness and the course environmentConcept, Predict & trace, Build & verify
Scalars, vectors, matrices and tensorsConcept
- Day 94 lessons, 24 min
Scalars, vectors, matrices and tensorsPredict & trace, Build & verify
Tensor operations and broadcastingConcept, Predict & trace
- Day 104 lessons, 29 min
Tensor operations and broadcastingBuild & verify
Computation graphs and automatic differentiationConcept, Predict & trace, Build & verify
- Day 114 lessons, 27 min
Multilayer neural networksConcept, Predict & trace, Build & verify
Datasets and DataLoaderConcept
- Day 124 lessons, 26 min
Datasets and DataLoaderPredict & trace, Build & verify
A reproducible training loopConcept, Predict & trace
- Day 134 lessons, 30 min
A reproducible training loopBuild & verify
Saving and loading modelsConcept, Predict & trace, Build & verify
- Day 143 lessons, 20 min
CPU, accelerator and multi-GPU foundationsConcept, Predict & trace, Build & verify
Free learning room: A model learns, 18 min
After 14 days
- You can explain how a language model is built: from predicting the next token and the transformer to pretraining and fine-tuning.
- You have worked through tensors, autograd, a multilayer network, datasets and a training loop in exercises in your browser.
- In the learning room “A model learns”, you have trained real weights in your browser and checked what your metric actually proves.
- You know which building blocks a model of your own needs, and in which order they are built.
What the challenge does not do
You do not build a language model of your own in these 14 days yet. That happens in Plus (not bookable yet): 8 more sections with 202 lessons, from text data and attention to pretraining and fine-tuning, plus 3 more learning rooms.
View plans and availabilityJoin in public
Post your start with #14TageLLM on X. The link opens X with a prepared text; nothing is sent to us.