Hi authors, thanks for MathVerse!
Following up on #10 — would it be possible to release the remaining portion of the dataset beyond testmini? Based on the paper (~15K samples total vs ~3.9K released), there's a significant portion that would be very useful for studying problem distribution and for training-eval separation.
Totally understand if there are leaderboard integrity concerns. Even a smaller held-out train split (clearly labeled) would be a huge help to the community.
Thanks!
Hi authors, thanks for MathVerse!
Following up on #10 — would it be possible to release the remaining portion of the dataset beyond testmini? Based on the paper (~15K samples total vs ~3.9K released), there's a significant portion that would be very useful for studying problem distribution and for training-eval separation.
Totally understand if there are leaderboard integrity concerns. Even a smaller held-out train split (clearly labeled) would be a huge help to the community.
Thanks!