Open stsouko opened 2 months ago
In graphormer bias to attention scores are learnable.
https://github.com/microsoft/Graphormer
Yup, we want to support trainable biases. There is some subtlety with how to implement this -> Broadcasts in the forward become reductions in the backwards. We are investigating though the best way to enable this
In graphormer bias to attention scores are learnable.
https://github.com/microsoft/Graphormer