SPD: Sync-Point Drop for Efficient Tensor Parallelism of Large Language Models.

Han-Byul Kim, Duc N. M. Hoang, Arnav Kundu, Mohammad Samragh, Minsik Cho

Browse the full ICML paper archive.