You can resolve the error by explicitly reshaping Y_pred or pred.
But logically, number of columns in Y and Y_pred should be same to start with.
Can you share link to your notebook?
Not accessible. Make the link accessible to anyone.
ffsnn_multi.fit(X_train, Y_train, epochs=2000, learning_rate=0.005, display_loss=True)
Here, instead of
Y_train, you have to send one-hot-encoded vector i.e.