Transformers and Attention
Lesson notes
Overview
Attention lets every token weigh every other token when building meaning. This topic is rated advanced and is number 109 of 117 in Computer Science. Deeper models, formal results and harder reasoning.
What you are seeing
The panel above shows the steps of a process, laid out in order. Click a step to see what happens next. The animation is drawn live from the model, so everything you see responds to the sliders. Press Space to pause, use Restart to replay from the start, and change the speed to slow down quick events.
In depth
A transformer processes a sequence by letting every token attend to every other. Each token is turned into a query, key and value vector, and attention weights are computed as softmax(QKᵀ/√d), then used to take a weighted sum of the values. Stacking many layers of attention and feed-forward networks lets the model build context-dependent meanings. Unlike recurrent networks, transformers process all positions in parallel, which allows training on huge data sets, and they underlie modern language models.
Reading the relationship
The expression "softmax(QKᵀ/√d)V" is a compact summary of the idea. It uses the symbols s, o, f, t, m, a; name what each one stands for before you start. Notice that it contains a square root, so the result grows more slowly than the quantity inside it, and a ratio, so increasing the denominator makes the result smaller. Then use the sliders to see how changing one quantity changes the others.
A guided exploration
1. Run the animation once with the default settings and describe in words what happens. 2. Change one slider at a time and note which quantity responds most. 3. Push each slider to its extreme and decide whether the result still makes sense.
Try this
Try to explain why the pattern holds, not just what it is. Look for limiting cases and what happens at the edges.
Common pitfalls
Test small inputs by hand and then the largest ones. Off-by-one errors and forgetting how cost grows with input size are the most common traps.
How it connects
It builds on “Convolutional Neural Networks”, the topic before it in Computer Science. Once you are comfortable, continue with “Reinforcement Learning Loop”.
Why it matters
Computing ideas decide how fast, how safely and how well software can solve problems.
Where you might use it
- Revising “Transformers and Attention” before a class, exam or certification
- Choosing the right algorithm or data structure when writing software or apps
- Debugging programs and judging how fast or how safely code will run at scale
- Understanding how search engines, games, phones and AI systems make decisions
Frequently asked questions
What is Transformers and Attention?
Attention lets every token weigh every other token when building meaning. It is an advanced topic in Computer Science.
How do I use the Transformers and Attention simulation?
The panel at the top shows the steps of a process, laid out in order. Click a step to see what happens next. Move the sliders one at a time, then press Space to pause the animation and look closer.
What does the formula mean?
The expression softmax(QKᵀ/√d)V packs the idea into symbols. Name each symbol first, then change one quantity with the sliders and see which others respond.
Who is this topic for?
Deeper models, formal results and harder reasoning. If the quiz feels hard, reread the lesson notes and try the previous topic first.
What should I learn before this?
Start with “Convolutional Neural Networks”, the previous topic in Computer Science. Each topic in a subject builds on the ones before it.
How do I track my progress?
Press “Mark as learned” on this page. Progress is stored only in this browser and is never sent anywhere. The Browse page can show topics you have learned or still need to learn.