greetings, does the benchmark include using multiple optmizers/loses(specifically custom ones) or it should only focus on the models tested