Basically, I have a trained .h5 model (I'm not married to that format, but it's how I inherited the codebase) and want to provide a way for other people in my lab to use this file for prediction. Right now, the call to load_model() inside keras.saving.save or keras.models tries to run import tensorflow.compat.v2 as tf and throws an error if tensorflow hasn't been installed.
This is annoying, especially since tensorflow is so big and it isn't listed as a required dependency for keras. Is there a way to load this model for making predictions without the huge overhead of downloading tensorflow? Should I be saving it to a different format to avoid the call to load_model()?
