fantasyzhai / lsh_attention Goto Github PK
View Code? Open in Web Editor NEWThis project forked from somisawa/lsh_attention
Calculate Softmax layer of Attention in O(LlogL)(L=sequence length) instead of O(L^2) using crpss-polytope Locality-Sensitive Hashing(https://arxiv.org/abs/1802.05751 ).
License: MIT License