|
Mila
Deep Neural Network Library
|
Configuration for the attention output-gate component. More...
Classes | |
| class | Mila::Dnn::AttentionOutputGateConfig |
Namespaces | |
| namespace | Mila |
| Mila main API namespace. | |
Configuration for the attention output-gate component.
The gate has no structurally required parameters: its width comes from the input shape at build time and its activation is a compile-time template parameter, the same shape Swiglu carries for its gate function. The default constructor is correct.