OpenNMT Forum

Pretrained model for German-to-Englsih?

Hi,

We need a MT tool for German-to-English.

Do you have any pretrained model for German-to-English that is ready to be used? We plan to run it on Linux without GPU. Thank you so much in advance!

Have a look here or here.
Don’t expect very high translation quality though, these are basic setups. Also, not sure how they’d perform on CPU.

Many Thanks, François Hernandez!

However, I got lots of ??
1). Is this expected?
2). If so, should I train the model with the domain corpus for better result?
3). Which model would you suggest to train for German-to-English?

Any suggestion will be appreciated!

We posted good quality EN to DE but not DE to EN.

@chrislumd Regarding the ??s, do you apply the proper sentencepiece tokenization before inference?

I used the translate.py (from OpenNMT) with options:
–model pretrained_model.pt --src inFile --output outFile

The input file contains 60 lines. Each line is a German sentence.
I was assume that this is a good test for the start…
Did I miss something? Thank you!

You have to tokenize your sentences with the provided sentencepiece model before translating them.
You can use sentencepiece directly or OpenNMT’s tokenizer for this.

I trained a DE->EN model for my LibreOffice translator (based on OpenNMT-py)

it’s on the release page.

Any feedback is very welcome.

Best regards

Thomas