What is the best attention calculation formula in NLP?

[ji_cocaer]

Has anyone done a similar study or comparison?Some related papers are weclome.