Nicola Bertoldi

Results 40 comments of Nicola Bertoldi

Hi @Fitz13009 Etienne is almost right how to reset an engine. You have to delete directories ```engine/ENGINE``` and ```runtime/ENGINE```, where `ENGINE` is the name you give to your engine. By...

Hi @Fitz13009 from the training.log it is seems that training worked properly And so, the system should have stored the model in the directory ``engines/default`` Specifically, in this directory there...

@Fitz13009 Are you sure that you the the installation succeded? What the output of the command ```./mmt status```? Which is the content of ``models/decoder/model.conf`` and ``engine.xconf``? PS: In the future,...

@mzeidhassan In ModernMT v4.X, all well-formed xml tags ('', '', '') are not considered during the translation process. ModernMT removes these tags from the source text, translates the text (without...

@popstick Surely, the output is wrong. The most likely reason for the babble is the amount of training data, especially if the vocabulary is large. Another reason could be a...

@popstick The training seems end correctly. So again I reconfirm that the problem is due to the lack of data. Are you sure that the dictionary of the source and...

@popstick I would suggest to remove apostrophes internal to words, and re-train your system.

I confirm that numbers are not transformed into '0'. They are kept in their original form.

@LoekvanKooten actually we never tried testing our software on the RTX3-30 series. So we do not know which is the right configuration you require. Concerning GEForce 2080 Ti, the requirements...

@LoekvanKooten we do not have short-time plan to test ModernMT on this kind of GPUs (especially because we not such GPU available) and modified the code to be compliant with...