torch.load(model_dict.pt) Error It doesn't load

code:

model.load_state_dict(torch.load(path+'model_dict.pt', map_location=device))

error:
    raise TypeError("Expected state_dict to be dict-like, got {}.".format(type(state_dict)))
TypeError: Expected state_dict to be dict-like, got <class 'transformers.models.bert.modeling_bert.BertForTokenClassification'>.

An error occurred while loading a pt file that was trained with PyTorch.

Please have a look at this https://pytorch.org/tutorials/recipes/recipes/saving_and_loading_a_general_checkpoint.html

It seems like you did not save the checkpoint correctly. From what you get as an error I suppose you did

torch.save(model, some_path)

If you try to load this via model.load_state_dict() you will get this kind of error.

How you save a checkpoint is by doing

torch.save(model.state_dict(), some_path)

This can then be loaded and uses

model_state_dict = torch.load(some_path)
model.load_state_dict(model_state_dict)

In general it often makes sense to not only save the model state dict, but more information like number of epochs so far, the optimizer state dict if it has one, e.g. if you are using ADAM. Generally everything you need to use the model and maybe also resume a training. This is what the official documentation also does. This is something along the line of

torch.save({'epoch': epoch,
            'model_state_dict': model.state_dict(),
            'optimizer_state_dict': optimizer.state_dict()
            }, some_path)

# and then you can load it via
checkpoint = torch.load(some_path)
model.load_state_dict(checkpoint['model_state_dict'])
optimizer.load_state_dict(checkpoint['optimizer_state_dict'])
epoch = checkpoint['epoch']