Library analogy

Attention is often explained with three roles. A library analogy keeps them memorable:

• Query — what you are looking for (“information about Transformer attention”).
• Key — the descriptive label on each book used for matching.
• Value — the useful content inside the books you matched.

You compare your query with keys. Stronger matches get more attention. Values from those matches are combined into a new representation.

Query
  ↓
compare with Keys
  ↓
Attention Scores
  ↓
Weight Values
  ↓
Combined Representation