GuKerui

Hi, I have a follow-up to better understand this issue. 1) I can run inference. 2) When I set distributed=False, I can run MMDP on a single GPU.