Question about data

[mzeidhassan]

Question about data

[captainvera]

Hi @mzeidhassan,

First of all, I'm sorry for my delayed response. What you are saying is absolutely correct. Be mindful however, that the predictor model is not an actual QE model (it cannot predict TER or word tags). It is merely a pretraining step or in other words, the first part of a two-part model the Predictor-Estimator.
The estimator is plugged on top of the predictor and uses the embeddings learned by the predictor to predict the actual QE tags. That said, the changes you propose to the config are absolutely correct.

Let me know if you have any other questions.

[captainvera]

hey @mzeidhassan can I close this issue?

[mzeidhassan]

Thank you so much for your support @captainvera
I appreciate it.

[mzeidhassan]

Thank you @captainvera for your kind follow-up. I am closing it now. Thanks again!