Transformers · Level 3 of 5
Feedforward Block
A positionwise feature transformation within a transformer layer.
It typically expands and contracts hidden features with a nonlinearity.
Example
Each token independently passes through two learned projections.
Listen to the definition and example
Audio transcript
Feedforward Block. A positionwise feature transformation within a transformer layer. It typically expands and contracts hidden features with a nonlinearity. For example: Each token independently passes through two learned projections.
Explore this concept
Why it matters
This helps you trace how token representations exchange and transform information.
Start with
Quick recall question
Try answering before looking back at the definition.