Plan to release other pre-trained models?

[sorami]

Plan to release other pre-trained models?

[captainvera]

Hi @sorami,

Let us discuss this, and I'll get back to you early next week 🙂

[captainvera]

Hello,
I know that my colleague has already answered your email, but in an effort to be as open as possible to anyone with future interest in this:

Thank you for your interest in OpenKiwi.

We don't have concrete plans for releasing more pre-trained models. On the other hand, the same parallel corpus used for pre-training the en-de Predictor model could also be used for pre-training it in the reverse direction, so it should be somewhat easy to do.

However, its usefulness really depends on the domain of the data your going to use on top of it.

We do plan on future updates to OpenKiwi later this year so stay tuned!

[captainvera]

Hello @sorami,

Yes, in the future we intend to release the models of our participation on WMT19.

That is: En-De/En-Ru/En-Fr NMT

Did you have anything specific in mind?

[sorami]

Hi @captainvera , thank you very much for your reply!

Yes, in the future we intend to release the models of our participation on WMT19.

Got it!

I was looking for De-En language pair (WMT18 De-En SMT).