Optimizing tensorflow model using tflite but keeping .pb format

Question

I have trained a keras model and for the inference I have to use the saved_model.pb format, but before that I would like to optimize my model using tflite. Is there a way to quantize/prune my model without turning it into .tflite format? There is stuff like this that turns .tflite to .pb. But I would like to skip that part entirely. Thanks in advance!