I am wondering if the model needs to be trained by a large dataset before using it to compare the similarities between 2 assembly functions?