mirror of
https://github.com/Helsinki-NLP/OPUS-MT-train.git
synced 2024-09-17 15:17:22 +03:00
1.1 KiB
1.1 KiB
Things to do
Issues
- get rid of BPE to simplify the scripts
General settings
- better hyperparameters for low-resource setting (lower batch sizes, smaller vocabularies ...)
- better data selection (data cleaning / filtering); use opus-filter?
- better balance between general data sets and backtranslations
Backtranslation
- status: basically working, need better integration?!
- add backtranslations to training data
- can use monolingual data from tokenized wikipedia dumps: https://sites.google.com/site/rmyeid/projects/polyglot
- https://dumps.wikimedia.org/backup-index.html
- better in JSON: https://dumps.wikimedia.org/other/cirrussearch/current/
Fine-tuning and domain adaptation
- status: basically working
- do we want to publish fine-tuned data or rather the fina-tuning procedures? (using a docker container?)
Show-case some selected language pairs
- collaboration with wikimedia
- focus languages: Tagalog (tl, tgl), Central Bikol (bcl), Malayalam (ml, mal), Bengali (bn, ben), and Mongolian (mn, mon)
Other requests
- Hebrew-->English and Hebrew-->Russian (Shaul Dar)