发帖
雷达卡
PipeTransformer: Automated Elastic Pipelining for
Distributed Training of Large-scale Models
Chaoyang He 1 Shen Li 2 Mahdi Soltanolkotabi 1 Salman Avestimehr 1
Abstract Transformer (ViT) (Dosovitskiy et al., 2020) also achieved
89% top-1 accuracy in ImageNet, outperforming state-of-
The size of Transformer models is growing at an the-art convolutional networks ResNet-152 (He et al., 2016)
unpreceden ...
京公网安备 11010802022788号
