Comment by Xunjin

Do you have other ideas of "combinations" for this kind of benchmark?

I'm wondering if this is being trained on by the models today.