* add an interface to predict the latency of compressed model

* just for rerun

* delete temporary model files
Co-authored-by: Nceci3 <ceci3@users.noreply.github.com>