[NLPL Task Force (A)] Translation activity milestones

[Scherrer, Yves]

Hi all,

I’d like to announce the availability of the updated MT modules and datasets on the new clusters Puhti and Saga.

The following modules have been installed: nlpl-moses, nlpl-efmaral, nlpl-mttools, nlpl-opennmt-py, nlpl-marian-nmt. Further instructions are available on http://wiki.nlpl.eu/index.php/Infrastructure/software/catalogue and http://wiki.nlpl.eu/index.php/Translation/home
(@Bjørn: this completes milestones B1.5 and B3.4)

In terms of datasets and pretrained models, we now have the following offerings:
- Documented datasets from the 2017 and 2018 competitions (WMT, IWSLT) have been copied over to the new clusters and updated with the University of Helsinki WMT2019 participation (German-English + Finnish-English).
- The pretrained OpenNMT-py, Marian and Moses models have been copied over to the new clusters, but have not been tested yet. In particular, the scripts will need to be updated due to some API changes in OpenNMT-py. I will finalize this in the coming weeks.
- As a joint OPUS/MT initiative, Jörg has made available Marian models pretrained on the OPUS datasets for more than 150 language pairs. The models, including download links, are documented on https://github.com/Helsinki-NLP/Opus-MT/tree/master/train/models
(@Bjørn: this completes milestone B2.3)

As always, please don’t hesitate to contact us if anything is missing or not working.

I wish you all a nice and relaxing Christmas time,
Yves
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.nlpl.eu/archives/infrastructure/attachments/20191220/2aa3ac4e/attachment.htm>