AttributeError: 'float' object has no attribute 'copy'

Hi everyone, I copied the weights of my trained model using “model.state_dict()” and then I want to update these weights using model.load_state_dict(weights) but I’m getting an error "AttributeError: ‘float’ object has no attribute ‘copy’ " as shown in the images below. Any ideas as to how can I can fix this?

Could you post the model definition of x_global_model, as it seems the state_dict() operation returns a float, while an OrderedDict would be expected?

It seems you are manually appending some “weights” to g_global_weights and try to treat it as a state_dict, which is causing the issue.
You would have to use the workflow posted previously, i.e. create the state_dict via:

sd = model.state_dict()

and load it via:


Currently you are replacing the initial state_dict with an empty list and append the (unknown) values.

PS: you can post code snippets by wrapping them into three backticks ```, which would make debugging easier.