Sagemaker deployment

[Olamyy]

I removed the ONMT folder and added it to the requirements.txt

[ruohoruotsi]

Since there are no automated tests or CI/CD, did you verify if the training scripts still work correctly? It would be essential if they do, as this is the primary usage of the repo (i.e. model building) thus far.

[Olamyy]

Yeah. I haven't tested that. Will get to it.

[ruohoruotsi]

okay, you'll be entering researchy code, but you should be able to pull all the data & at least see if you enter the first epoch. Ensure you have the repos checked out correctly as in https://github.com/Niger-Volta-LTI/yoruba-adr/blob/master/src/aggregate_corpora_make_parallel_text.py#L22

If it doesn't work, it may be best to separate concerns (1) ensure training isn't broken (2) deploy

[Olamyy]

How about the easier alternative of keeping two branches.
I can confirm this (my) branch works for sagemaker deployment.
So, you could create a new branch and I'll merge this to it.
Then, we keep master as is until you can clear the opennmt folder yourself. I feel that'll ensure I don't break anything.

[ruohoruotsi]

Sounds good, I'll make a new branch as you suggest

[ruohoruotsi]

new branch here: https://github.com/Niger-Volta-LTI/yoruba-adr/tree/sagemaker-deployment

mirroring yours @ Olamyy