There is another question about

bentoml.pytorch.save_model

. In the documentation and code, it gets the model as a

torch.nn.Module

instance, and save it with

torch.save

. However, the recent PyTorch does not allow to save model instance directly instead of state_dict. Therefore, it returns the following error. RuntimeError: Serialization of parametrized modules is only supported through state_dict(). Is there anyone who can fix this problem?