high learning rate in later stages

[coldgemini]

Hi,
why did you use much higher (4X) learning rate from 2 stages and leave only the first stage a small learning rate? If it's kind of a fine tune strategy, why did you not just freeze the first stage? Is it due to some empirical results?

Thank you a lot.

[1Rookiechang]

have you download the pretrained model？can you share it with me.
because the download site cannot open.thank you very much!
827927604@qq.com