LSTM/GRU/RNN (dropout) Pytorch implementation following TensorFlow one

Do you have some code of your model to show? There are typically 2 issues that often crop up yielding subpar results: