State_dict() got an unexpected keyword argument 'destination'

vision
1

I am training an ensemble of two deep learning models and the workflow is working fine and saving ‘.pth’ file when trained on M1 mac but on GPU I am getting this error. i tried changing version to 1.12.0 (torch), same error. I am writing full error and code snippet. Please assist. Thank you.

# save model function outside Trainer Class
def save_model(model, model_filename, path='.'):
    if not model_filename.endswith('.pth'):
        model_filename += '.pth'
    filepath = os.path.join(path, model_filename)
    io.create_path(filepath)
    torch.save(model.state_dict(), filepath)


# save_model method inside Trainer class

def save_model(self, model_filename=None, *, models_dir='models'):
        if model_filename is None and self.model_filename is None:
            raise ValueError('Param "model_filename" is None')
        elif model_filename is None:
            model_filename = self.model_filename
        path = os.path.join(self.path, models_dir)
        save_model(self.model, model_filename, path)


# last part of training method where it saves the model


 # update ReduceLROnPlateau scheduler (if available)
            if (valid_loss is not None and scheduler is not None
                and isinstance(scheduler, lr_scheduler.ReduceLROnPlateau)):
                scheduler.step(valid_loss)

            if self.primary_metric in scores:
                # save model checkpoint
                is_better = scores[self.primary_metric] > best_score
                if is_better:
                    best_score = scores[self.primary_metric]
                    best_state_dict = deepcopy(self.model.state_dict())
                    # best_state_dict = self.model.state_dict()
                    if self.model_filename is not None:
                        print(f'Epoch {epoch+1} - Save Checkpoint with Best Recall Score: {best_score:.6f}')
                        self.save_model(self.model_filename, models_dir='models')

        if best_state_dict is not None:
            self.model.load_state_dict(best_state_dict)

Traceback (most recent call last):
  File "/notebooks/ensamble_model/ensamble_models/fine_tune.py", line 129, in <module>
    trainer.train(no_epochs=config.no_epochs, lr=config.learning_rate)
  File "/notebooks/ensamble_model/ensamble_models/src/core/training.py", line 259, in train
    best_state_dict = self.model.state_dict()
  File "/notebooks/.env/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1381, in state_dict
    module.state_dict(destination=destination, prefix=prefix + name + '.', keep_vars=keep_vars)
TypeError: state_dict() got an unexpected keyword argument 'destination'
2

Could you check if you have overridden the state_dict method in your custom module class?

3

Thank you for your response @ptrblck. I found the error is about the pre-trained architecture I am using in my Model Class. I was combining two models one of which was “vit_base_patch16_384” from timm library and I found if I change to “vit_small_patch16_384” or some other like I tried “efficientnet_b0”. then my code works fine even on GPU. So, it was about architecture which I was using in my new Model Class.