tiedemann

This is indeed strange. I don't really know what is happening in those cases. I need to investigate this. The only reason I can think of is that the sentencepiece...

Is that a question about finetuning in OPUS-CAT or using the procedures in OPUS-MT-train?

Could we move this issue to https://github.com/Helsinki-NLP/OPUS-CAT?

The models in OPUS-MT may not be directly nematus compatible but that is something that you could ask at the marian-nmt mailing list. I am not aware of any conversion...

This is the one I have: https://object.pouta.csc.fi/Tatoeba-MT-models/eng-kor/opusTCv20210807+bt-2021-11-10.zip You can check with the leader board what kind of models are available for a language pair you are interested in: https://opus.nlpl.eu/leaderboard/index.php?metric=bleu&src=eng&trg=kor&test=all

I have some trouble converting the model to pytorch but I can tell you that I’m also training a new model at the moment and hopefully it’ll come up here...

Yes, that looks a bit weird. The model at huggingface does not seem to handle that kind of input well. At least a newer OPUS-MT model does not do that...

I have no experience with iOS and I wonder whether the conversion to huggingface will help you. Did you see our models in their HUB: https://huggingface.co/Helsinki-NLP?

Integrating bitextor here would be a major task and I would rather like to keep the bitext harvesting procedures outside of the core translation service. It could be an interesting...

Looks like marian-server is not available / installed in your colab instance.