About the Baseline of Qwen

[visionxyz]

Hello, thanks for your work!

I am trying to reproduce the baseline of Qwen without training based on the framework of JudgeLM, however, I find that the Qwen model can not output two scores correctly with training (JudgeLM itself is okay).

I am wondering whether there is anything I should do about it, thank you so much if you could give me some advice!