arXiv - CSCL: "DeepSpeed Ulysses: System Optimizations for Enabl…" - CleverLibre Social

arXiv - CSCL @arxiv_cscl@social.cleverlibre.org

DeepSpeed Ulysses: System Optimizations for Enabling Training of Extreme Long Sequence Transformer Models. (arXiv:2309.14509v2 [cs.LG] UPDATED)

http://arxiv.org/abs/2309.14509 #arXiv #NLProc

Oct 05, 2023, 03:17 · · arxiv-cscl · · ·

Sign in to participate in the conversation