Hao Wang (Dylan)
Hao Wang (Dylan)
Since the model was trained using DDP, the model is wrapped with `module`. We need to remove this prefix before loading the pre-trained model weights. The difference between train and...
It is not redundant, but useful to have a sanity check. You can google by yourself about PyTorch DDP.
> Hi @Dylan-H-Wang, I think it can be expanded correctly? Maybe some network issues? >  > Thanks! That is strange, it does work...