I was wondering about the differences between using tf.saved_model.save and model.save under the hood. In my experiments, I came across seeing differences for myself when trying to save a subclassed model. It would be helpful if you could shed some light on how and why these two work differently?
PS: I had raised an issue (Differences between tf.saved_model.save and model.save · Issue #52128 · tensorflow/tensorflow · GitHub) which contains exactly what I tried but this is only about making changes to the docs and not about trying to understand why and how these are different thus this message.