How to set the retrain parameters of ResNetv2-152?

[homles11]

Before retraining the model, I got almost the same accuracy as the log of 152-layers ResNet: 76.63%(Top-1 Validation Accuarcy). Then I retrained the model with lr=0.0005 and wd=0.001. But after one epoch, I found that my top-1 validation doesn't increase like the the log recorded (only increase 0.1% or doesn't increase). I wonder if this setting is wrong? And which setting of lr and wd should I use so that I can the same performance as the log recorded.