Your work is awesome. I have noticed the model parameters you share in this repo, the re-implement results is the same as which of your paper. However, when I try to train your model by using the hparam.json in PFPL, I can't get such good results. So, would you please share some tricks which you use in your training procedure?
Your work is awesome. I have noticed the model parameters you share in this repo, the re-implement results is the same as which of your paper. However, when I try to train your model by using the
hparam.jsonin PFPL, I can't get such good results. So, would you please share some tricks which you use in your training procedure?