Better way to forward sparse matrix

@11130 you could think of contributing to this thread - Sparse tensor use cases