Encoder-Decoder Model for text generation gives repetitive and bad predictions

Question

I am trying to generate text using an encoder-decoder model, inspiring myself from this tutorial. I have problems with the loss and accuracy of my model. First, the accuracy was