At present, the Paddle ROCm version is based on Haiguang CPU (X86) and DCU to support single-machine single-card/single-machine multi-card training and inference of the following models.

#Propeller #Frame #ROCm #Version #Support #Model