DenseNet - Prediction on new image failing

My project is about image classification.
I’ve trained my model on kaggle and used learn.export() to save it as export.pkl.

Now, I’ve tried to load it and use it to predict on a single image, in a new Kaggle notebook. But I get this error:

AttributeError: '_DenseLayer' object has no attribute 'memory_efficient'

This is my code:

!pip install fastai==1.0.40
from fastai.basic_train import *
from pathlib import Path
path = Path()
model_export = load_learner(path/'../input/trainingmodel')
from import * # open_image
cat, tensor, probs = model_export.predict(open_image(f"../input/test/imagepng"))

I used 1.0.40 to train my model and for the testing part as well.
I tried going by this solution from a post, but I still have the same issue:

!pip install --upgrade torchvision