The code (command) to reproduce the official benchmark

[MorishT]

Hi,

Thanks for the nice framework.

What is the exact command to evaluate a new model to compare it fairly with the models shown on the official leaderboard?

[MorishT]

From the paper, I guess we have to search for the best top_p and temperature values for each model. Do we have other things to do?