Model returns NaN values when deployed

My image classification app works fine when deployed locally , but predicts all NaN values when deployed on streamlit cloud.I have made sure I am using all correct dependencies( made a requirements.txt using pip freeze).Also if it matters, I used mixed precision while training the model on google collab.

feel free to clone and experiment with it GitHub - viraj-19/food101-tfds: This project is based on kaggle food101 dataset.This project was a part of the tensorflow course by Daniel Bourke from Zeroi to Mastery academy