Datasets used to train NMT supervised task ?
#25
by
OrianeN
- opened
If I understood the paper correctly, the T5-small was trained on supervised tasks including NMT in 3 directions: EN>FR, EN>RO and EN>DE.
Yet I can't find the datasets used to train on these supervised tasks, could you please add them to the model card ?
If I'm not mistaken, the paper mentions the following training datasets for the NMT tasks:
- EN>DE: News Commentary v13, Common Crawl, Europarl v7 (+ newstest2013 for validation)
- EN>FR: WMT 2015 (+ newstest2014 for validation)
- EN>RO: WMT 2016