What dataset is D_G for policy SFT in 0-th iteration and further MCTS* ?

[RewindL]

Thanks for your remarkable work. I see the Policy data collected from 1st/2nd iteration MCTS*. I have a question that what is the initial question set $D_G$ you use to SFT initial policy model $\pi_{S_0}$ in Algorithm.1 and run MCTS* in iterations?

I guess user-costumed given $D_G$ works too, but I wish dataset $D_G$ can be released just like $D_{V_0}$. Do I miss something ? $D_G$ is also construct from SciInstruct, or use same train set corresponding to test set (e.g. use math-train as $D_G$ when evaluating on math-test) ?