C5_W3_A1 attention model "May 33"

Later in the assignment, we are asked to load the existing weights and take the model for a spin. One of the examples is incorrectly translated by the system as “yyyy-05-33”
source: 3 May 1979
output: 1979-05-33
In the visualization it seems like that first initial digit in the input, activates both “3” values in the output and also digits “x97x” in the year have a significant contribution. Is this all normal, or did I wire my model incorrectly?

I have the same results as yours.

1 Like