Transformer Feed-Forward Layers Are Key-Value Memories
31658Knowledge Triples
| Source Entity | Relationship | Target Entity |
|---|---|---|
|
Feed Forward Layers
Method |
captures |
Shallow Patterns
OtherEntity |
|
Feed Forward Layers
Method |
learns |
Semantic Patterns
OtherEntity |
|
Key Value Memories
Method |
correlatesWith |
Training Examples
Material |