yes, that is clearer indeed. However S4 and Mamba class models have also performed well at small scale and started lagging with larger models and larger context sizes, or at particular tasks.