I really don’t see how this kind of toolkit could actually be used by a translator directly.
Most of them 1) won’t be Linux savvy enough 2) won’t have the GPU to train a model.
There are 2 directions I think:
sdl studio integration as discussed in the other thread
create kind of “script recipes” where you can easily see all the necessary steps from
getting the corpus, pre-procesing, training, evaluating, spitting the bleu scores.
These “recipes” could make it easier for “openNMT newbies” to step in the project and replicates some results.
@srush if you’re familiar with Kaldi, this is the kind of “recipes” I have in mind.