move tensor if necessary (like device_map: auto with >1 gpu)

#2
No description provided.
RangiLyu changed pull request status to merged

Sign up or log in to comment