Can not reproduce results

[BwLiu01]

Thanks for your fantastic work! When I tried to reproduce the training result following the quick-start, I found the reward did not increase when using 7B model. The 14B model looked good, but reward collapse still occurred.

[BobbyZhouZijian]

Hi. Thank you for your question! We are working to reproduce the issues. It might be something with the batch size. We will address this as soon as possible.