Is there a smaller data set that supports GRPO training?

[laogonggong847]

Hello @lewtun，Thank you for your excellent work！

I have a question if there is a smaller scale data set to support GRPO training, I am currently training GRPO on 8 A800s and it takes 3 days (not code data set)
In addition, I have run GRPO training twice on the same machine with the same configuration, and I find that there is a big difference in the convergence of the final loss and reward. Is this normal for reinforcement learning? How should I make sure that my reinforcement learning has an approximate effect every time (or that loss and reward converge as closely as possible)?