Hi,
As far as my understanding, the state dict should contain all the information about the model. I am sure that all the parameters will be contained.
But does the state dict contain other settings, like model.eval()?
The state_dict
contains all parameters and buffers registered to your module (not the architecture or some other code).
The training
attribute is not part of the state_dict
and will be set to True
in newly initialized models.