Jerin Philip

Perhaps the paper linked below will answer the corpus used. * [Revisiting Low Resource Status of Indian Languages in Machine Translation](https://arxiv.org/abs/2008.04860). Regarding the data/training: * [fairseq/data/cvit/corpora.py](https://github.com/jerinphilip/fairseq-ilmt/blob/master/fairseq/data/cvit/corpora.py) has dataset with a...

[This](https://kheafield.com/papers/edinburgh/transfer.pdf) paper might have some useful information, I think. I'd just retrain with new vocabulary, turnaround is approximately 1 day or something on 4 GPUs to start getting reasonable numbers....

This is weird, it should work. Maybe some fairseq-padding option. Try to use the more recent ones. ``` "mm-all-iter0" "mm-all-iter1" "mm-to-en-iter1" ```

Check https://github.com/jerinphilip/ilmulti/issues/1. Why are you using colab? Switch the install to torch=1.0.0 at the start of the notebook, that should work. Before installing ilmulti

Okay, do a bunch of package changes (fairseq==0.8.0 to a fork). I have managed a final self-contained colab-notebook implementation. https://colab.research.google.com/drive/1KOvjawhzPXOQ6RLlFBFeInkuuR0QAWTK?usp=sharing

`ILMULTI_CORPUS_ROOT` is only used for training. I have set this to a hard-failure to ensure scripts fail early than at a later stage. For inference purposes, this can be removed...

> figuring out the mess that's the Linux Desktop Ecosystem, sounds indeed a bit farfetched I have plans of some daemon/service/bus and hook ibus or something for outbound translation. I...

> My only pain is that the MacOS version will have more features than the linux version xD How would we do this? dbus is a good idea?

Yes. I am proposing something like librepo. Qt is not just ok, but based on cross-platform troubles I understand from exchanges appears to be the most viable option. What I'm...

> Even when deleting all other models so that the English-Finnish is loaded first (and therefore there's no swapping issues), I still can't get it to translate anything right, even...