Only increase learning rate from 1e-3 to 1e-2, AUC significantly better than results reported, why?

[FeiGSSS]

Hi,
Thanks for your contribution! I cloned your codes (python3 version) and trained the model using Enron_new you provided. When setting learning rate=0.001 (default value), the final AUC is around 0.86 (as you reported). However, when I increased the learning rate to 0.01 and regenerated context pairs and eval data, the final AUC reached 0.89. I have tested several times in the same environment as requirement.txt.
Did you try to tune the learning rate when you training this dataset? If not, this is a free and significant improvement, right?