EPLoN: Exploiting Efficient Parallelism with Selective Rematerialization for Lightning Attention on Ascend NPU.
Han Bao, Zhenfeng Su, Alexander Setyaev, Stanislav Kamenev, Alexander Gneushev, Kang Zhao, Junmin Xiao, Haoran Lin, Anastasiya Bistrigova, Sergey Buzykanov, Evgeny Tetin, Guangming Tan, Boxiao Liu, Xueyi Zou, Zhenhua Dong, Constantine Korikov, Xianzhi Yu, Zhongzhe Hu
Browse the full ICS paper archive.