Transformers · Level 4 of 5
Grouped-Query Attention
Attention where multiple query heads share fewer key-value heads.
This can reduce cache memory while retaining multiple query projections.
Listen to the definition and example
Audio transcript
Grouped-Query Attention. Attention where multiple query heads share fewer key-value heads. This can reduce cache memory while retaining multiple query projections. For example: Eight query heads use only two sets of key-value projections.
Explore this concept
Why it matters
This helps you trace how token representations exchange and transform information.
Start with
Related concepts
Quick recall question
Try answering before looking back at the definition.