How is SVD from scikit-surprise handles empty values

I am studying the surprise lib for recommender system. SVD from this lib doesn’t require all value input in user-item matrix. But it is a must of the original SVD method. The official doc doesn’t explain this. Anyone knows how surprise works?