Questions about reproducing the WISE model on the Temporal dataset.

[Eliaukyh]

Hello!

Q1: I noticed that the code doesn't include an OOD Generalization evaluation metric, which is preventing me from reproducing the results on the temporal task. Could you update the code?

Q2: I’ve run several tests with the WISE model on LLaMA3-8B, but the results haven’t been very good. Both the rewriter_acc and rephrase_acc are around 50. From the perspective of scaling laws, shouldn't a larger model lead to better performance? What steps should I take to improve WISE's performance on LLaMA3?

I look forward to your response, and I apologize for any inconvenience caused. Thank you for your understanding!